Documentation

Knowledge base

Everything you need to know about GEO Scout

01 · 8 sections

Why your brand needs AI visibility

Where AI answers can cost a business customers, what to monitor and how GEO Scout helps identify problems and check changes.

1.1

Why now: AI already helps buyers choose products

RBC and Yandex research published in March 2026. These figures describe AI usage, not shares of orders or revenue.

35%
Online buyers use AI to choose and buy products
86%
Use AI services overall, according to the study
45%
Moscow buyers use AI for online shopping
74%
Projected share of AI-assisted shoppers in the medium term
  • The survey covered about 1,850 respondents aged 18–54 who had shopped online in the preceding three months. Consider this sample when interpreting the figures.
  • 74% is the authors’ forecast, not an observed result. Buyers already turn to AI for help choosing products.
1.2

AI answers are becoming a new traffic channel

Buyers ask AI what to choose, compare options and check terms. A brand can gain an interested visitor—or fall outside the shortlist before a website visit.

  • The channel starts with a recommendation. Check whether your brand appears in answers to buying questions.
  • Some decisions happen inside the conversation. Website referrals alone cannot show where competitors win the recommendation.
  • GEO Scout collects answers to your audience’s questions across 12 AI providers, showing which brands buyers see and why they are recommended.
1.3

A missing brand leaves the buyer with other options

Below are real AI answers to commercial questions. AI builds a shortlist and explains the options; buyers may never discover a brand that is absent.

Screenshots of real AI responses recommending GEO Scout

  • Track reach: the share of answers mentioning your brand, and the questions and providers where it is missing.
  • Compare competitors: who appears, in which use cases and with which arguments.
  • GEO Scout identifies presence gaps and saves full answers so the team can prioritize commercially important questions and investigate.
1.4

A mention can still steer the buyer toward a competitor

A name in a list is not always a recommendation. AI can praise a product while treating it as a fallback or an option for a narrow use case.

  • Track the brand’s role: primary choice, recommendation with arguments, secondary alternative or simple mention.
  • Review product associations, recommended audiences and reasons for preferring a competitor.
  • GEO Scout breaks down presence levels, arguments and caveats, helping identify benefits that need stronger evidence. See brand perception.
1.5

Negative answers create objections before purchase

Complaints about quality, support or terms can stop a buyer even when the brand is well known. Mention growth alone will not reveal this.

  • Track recurring complaints, negative answer share and the questions that trigger them.
  • Check whether negativity reflects a real product problem, outdated information or a subjective judgment. Each calls for a different response.
  • GEO Scout highlights negative answers, explains assessments and shows sources so teams can verify claims and route confirmed issues. See brand reputation.
1.6

Incorrect claims can rule out a suitable product

AI may invent a fee, miss a feature or repeat an outdated limitation, giving a buyer a reason to reject a suitable offer.

  • Track the accuracy of terms, capabilities and specifications that matter to buying decisions.
  • Check claims against current team facts and review disputed cases. Opinions are not necessarily factual errors.
  • GEO Scout compares claims with supplied facts and shows contradictions alongside original answers. Checks cover information that could be matched to those facts. See answer accuracy.
1.7

A finding should become a concrete task

Once you know where the brand is absent or misrepresented, the team needs pages and materials to work on.

  • Review cited sources: clarify your website, add evidence or propose updates to external reviews. A citation does not prove the page caused an error.
  • GEO Scout shows cited URLs and publications featuring competitors. The command center, content plans and AI agent help investigate and prioritize work.
  • Support the positioning you want with facts, tests, case studies and clear terms. See sources and the AI agent.
1.8

Check whether answers changed after your work

Publishing a material starts the check. Look for changes in shortlisting, recommendation strength and product accuracy.

  • Compare reach, share of voice, recommendation levels, negativity and contradictions. For accuracy, also consider the number of checked answers.
  • GEO Scout shows trends and citations of campaign materials. Keep questions, providers, geography and periods comparable.
  • Compare changes with referrals and leads in your analytics: visibility growth alone does not prove sales growth. Start with brand setup.
02 · 7 sections

Brand and monitoring setup

Launch GEO Scout in a few minutes: onboarding, prompts, competitors, and your first monitoring run

2.1

Run onboarding and pick prompts for your first run

Onboarding sets up GEO Scout in three steps: enter your site, review the brand data and refine the business context, then pick up to five prompts for a trial run. The whole flow takes a few minutes — then the system checks your visibility in ChatGPT, Gemini, DeepSeek, Perplexity, GigaChat, and Alice AI on its own.

1/4
  • Step 1 — enter your site URL and hit "Check". GEO Scout parses the homepage and shows a "Website verified" card with the favicon, title, and description, so you can confirm it's the right domain before the analysis starts
  • Step 2 — review the brand name. This is the exact name the system looks for in AI answers. Click the pencil if the parser picked up the wrong thing: a chunk of the page title or a tagline instead of the name ("Kwork — freelance services marketplace" instead of "Kwork"), or a spelling your audience never actually uses
  • The "Other brand names" block — add the spellings people use for you in answers: Latin and Cyrillic ("Kwork" and "Кворк"), abbreviations, a pre-rebrand name, the legal entity. We look for the brand across all of these at once, so a mention isn't lost when an AI names you differently than your site does
  • The brand description (up to 500 characters) is drafted by AI from your site. It feeds more accurate recommendations, report generation, and competitor detection — rewrite it if the system misread what you do
  • Business context is optional, but worth filling in. The business model (B2B, B2C, D2C, or Marketplace) tells the AI how to read your market when generating queries and recognising competitors in answers
  • Niche sets your industry (e.g. "B2B services" or "Finance & fintech"). It's used by the recommendation engine so advice and content ideas stay relevant to your market, and by the technical site check: each niche has its own Schema markup — medical gets MedicalOrganization, an online store gets OnlineStore, instead of a generic Organization
  • Target geography is the most important field: this is the location we set in the browser before sending the query, and it decides which locale the answer comes from. For a local city business pick "City" and name the exact city — otherwise AI answers from the wrong locale and the competitors won't be yours. For a nationwide brand, "Country" is enough
  • Step 3 — pick up to five prompts for the trial run. Generated prompts are grouped by topic, and you can add your own wording via the "Enter your prompt" field. The "Selected 5/5" counter shows what's left, and the cross removes a prompt you don't need. Narrow, niche prompts surface more accurate competitors than broad queries. The trial run is free — 5 prompts every 7 days, enough to see your first visibility patterns before upgrading to a paid plan
Tip

Start with prompts your customers would actually use to find you. A broad query like "freelance marketplace" creates noise in the competitor list; a narrow one ("where to find a freelancer for a WordPress online store") surfaces direct competitors and clean signals you can act on.

2.2

First data and your first report

As soon as the trial run finishes, the dashboard fills with the first data on your AI visibility and GEO Scout automatically assembles a starter GEO audit. A link to it sits in a row above the KPI cards, and the document itself opens as a PDF — handy to show your team or a client.

PDF guide
Sample starter GEO audit
A real GEO Scout audit from a first run: coverage per engine, Share of Voice and the competitive field, a source map and concrete steps to grow visibility.
4 pages · 0.9 MB
  • KPI cards fill up from real AI answers: answer reach, Share of Voice, domain citation rate and Citation Share — demo values are replaced by your numbers
  • «Share of Voice dynamics» shows how brand shares are distributed over time, while the «Competitor ranking» next to it lines everyone up by reach, SoV, average position and sentiment — with a «Your brand» badge in the list
  • The «Initial audit» row above the cards is your first finished document: a one-sentence takeaway, and a click opens the full PDF with a per-engine breakdown, strengths and weaknesses, and a source map
  • The audit shows not only where you appear but also where you don't: for every prompt you see which engines mentioned the brand and at which position, which competitors they named and which domains they cited
  • The «Next free run in …» counter shows when you can repeat the run on the free plan — data refreshes once a week, so draw your first conclusions from this snapshot
Tip

The first audit is built from a single run — it is a starting point, not a verdict. Look first at the competitor list and the source map: they show which platforms you need to appear on for the models to start naming you more often.

2.3

Set up prompts, clusters, and monitoring engines

In «Monitoring → Queries» you manage everything that gets measured daily: the prompt list (one by one or in bulk), clusters and their geography. The set of AI engines comes from your plan — every engine in it is queried daily for every active prompt.

1/3
  • The «Monitoring → Queries» section is split into two columns: clusters with their summary on the left («1/10 active · reach 83% · SoV 33%»), the selected cluster's prompts on the right. Above the list sit that cluster's own metrics: reach, Share of Voice, average position and domain citation rate
  • Your plan sets the limit of simultaneously active prompts. The toggle in each row turns a prompt on or off: anything above the limit is stored as inactive and never queried, so you can keep a large list and monitor selectively
  • The «Bulk» button opens batch import: one prompt per line, with duplicates skipped automatically. You pick the group right there — an existing one or a new one via «+ New group…» — so the import lands in the right cluster instead of «No group»
  • Before adding, the panel shows how many prompts will become active and how many will be created inactive because of the plan limit — you see the outcome before pressing «Add», not after
  • Split prompts into clusters by meaning — for example, «Online payments for e-commerce» and «Acquiring for freelancers». Metrics — visibility, Share of Voice, citation rate — are calculated separately for each cluster
  • Every cluster has its own geography (gear icon → «Edit group» → «Geography» block): inherit from the brand, a city, or a country. City-level geography is critical for local scenarios — for example, «choosing a store in Moscow»: GEO Scout sends queries from that city's coordinates, so answers surface local competitors instead of nationwide ones
  • Cluster geography overrides brand settings for every prompt in the group. That lets one brand run several geo directions at once — a Moscow cluster and a St. Petersburg cluster, say — without creating a separate brand per city
  • The set of AI engines is defined by your plan and is not selected inside «Queries» — you choose the mix when you buy or change a plan on the pricing page. Every engine in the plan is queried daily for every active prompt
  • On the free plan the engine set is fixed — ChatGPT, Gemini, DeepSeek, Perplexity, GigaChat and Alice AI
Tip

Import prompts in bulk one topic at a time and assign the group right away: metrics are calculated per cluster, and they only mean something when the queries inside are homogeneous. Keep the list larger than your limit — inactive prompts cost nothing — and activate the directions you are actually fighting for visibility in right now.

2.4

Grow your query set with «Prompt discovery»

The «Prompt discovery» tab inside «Monitoring → Queries» suggests new queries based on your website, clusters, geography and brand business context. It is the way to extend coverage without inventing wordings by hand.

  • Recommendations are grouped by your current clusters: the left column shows how many prompts a group already has and how many are suggested for it («10 prompts · 18 recommendations»)
  • Each suggested query carries a «Why add it» line — what exactly this prompt will reveal in AI answers. Read it before adding: that way you pick queries by meaning, not by how they sound
  • Every recommendation can be added or dismissed one by one, checked in batches, or taken all at once with «Add all (18)». Dismissed ones are never suggested again
  • Queries are generated from an analysis of your website — products, categories, use cases and detected competitors — with the brand's geography taken into account, so the wordings come out local and conversational, the way people actually type them in chats
  • Prompts you already track are filtered out of the recommendations — no duplicates will appear in the list
  • At the bottom of the cluster column sits the «Suggested new clusters» block: the system spots demand segments your groups do not cover and offers to create them together with their prompts in one click
Tip

Start with the clusters that have the fewest prompts — that is where extending coverage moves visibility the most. And dismiss what does not fit right away: dismissed queries never resurface, so the recommendation list gets sharper over time.

2.5

Telegram visibility reports

The «Settings → Notifications» tab is where you connect the Telegram bot: the weekly brand visibility report arrives straight in the messenger, with no need to open the dashboard.

  • Connecting takes a minute: press «Open Telegram bot», send /start to the bot and enter the linking code from this screen. While linking is in progress the page shows «Waiting for connection…» and picks up a successful connect on its own
  • The code is single-use and short-lived — a «Expires in …» timer runs underneath it. If you miss the window, just generate a new one
  • Telegram is connected per brand: reports cover exactly the brand you are currently in. If you run several brands, repeat the linking in each of them
  • The weekly report holds the key changes in a compact format: Share of Voice and citation shifts, prompts that gained or dropped, competitors gaining ground
Tip

Connect the bot right after your first run: the report arrives on its own, so you spot a visibility drop during the week instead of the next time you happen to open the dashboard.

2.6

Invite teammates to work on the brand together

In Settings → Team the brand owner can invite teammates by email — each gets their own access to the dashboard, monitoring, and settings, while billing and brand removal stay with the owner. Available on every paid plan.

  • Invite anyone by email. They receive a link; if they don't have a GEO Scout account yet, one is created automatically — no need to share credentials in advance
  • Members see the same as the owner: dashboard, reports, monitorings, competitors, sources, AI traffic. They can edit prompts, competitors, and brand settings — full-fledged work with the data
  • Billing and brand removal stay owner-only: the team can't accidentally change the plan or delete the brand. Managing the team — invitations and access revocation — is also reserved for the owner
  • Revoke access at any time from the same Team tab: one click and the member loses access. A pending invitation link can be cancelled the same way
Tip

If your team works across several brands in different workspaces, invite people surgically — only into the brands where they actually contribute. Credits and the subscription are scoped to the brand, not the user, so adding a teammate doesn't increase your bill.

2.7

Launch monitoring

Launching monitoring becomes available once you pay for a plan: a «Ready to launch» banner appears on the «Monitoring → Queries» page. Pick the prompts you need, press «Launch monitoring» — a first run goes through, and after that monitoring runs daily.

  • The «Ready to launch» banner appears automatically once the subscription is active. «Go to prompts» takes you to the list settings, «Launch monitoring» starts the run right away
  • Check which prompts are active: the cluster panel shows «3/10 active prompts» and «8 of 100 active prompts on your plan». Only active prompts go into the run — toggle on what you want to track, within your plan limit
  • After you press «Launch monitoring», a first run goes through the selected prompts, and then monitoring continues daily. The card metrics (reach, SoV, average position, domain citation rate) fill in by the next morning
  • Statistics accumulate from the very first run: metric dynamics by day, week and month, period comparison, and the effect of your on-site changes and external publications
Tip

Don't rush to optimize right after launch — give the system at least 3–5 days to gather data. The more observations, the more reliably real patterns and weak spots surface instead of random noise in one or two AI answers.

03 · 5 sections

Visibility analytics and competitive intelligence

How to read the dashboard, mention trends, and competitor comparisons — from the big picture to precise growth zones

3.1

Home dashboard: all your visibility in one view

The GEO Scout home page brings together the full picture of your AI visibility: from the four headline metrics and day-by-day dynamics down to the exact domains and pages the AI engines cite in their answers. The top bar holds the period and "Filters" by prompt group and provider — everything below recomputes for the selected slice, and the button on the right exports the data. Below it sit the active prompt counter against your plan limit, the "Explain visibility" button (the AI agent compares the current period with the previous one and explains what moved coverage and share of voice), and a one-line summary of the latest weekly GEO report.

1/3
  • Four KPI cards: Answer coverage — in what share of AI answers your brand is mentioned at all; Share of Voice — your share of mentions among competitors; Domain citations — the percentage of answers linking to your site; Citation Share — your domain's share of all cited domains. Each value shows the change vs the previous period
  • "Reach trend" — a daily chart with a metric switcher: coverage, Share of Voice, average position, sentiment, and Citation Share. It shows not the period average but the exact day the turn happened
  • "Competitor ranking" — a compact table next to the chart: rank, rank change (Δ), coverage, SoV, average position, and a sentiment distribution bar. Your brand carries a "Your brand" badge and is always shown alongside the top competitors, so your real place in the niche is obvious
  • "URL types" — which kinds of pages the AI engines link to: listicles and rankings, articles, homepages, product pages, how-to guides, reviews, comparisons, news. It tells you which content format AI is willing to cite in your niche at all
  • "Top domains" and "Top URLs" — the specific sites and pages the answers link to, with citation counts, share, and icons of the providers they surfaced on. A ready-made list of places your brand needs to be present on
  • Every page carries a check of whether the brand is on the page itself: "Mentioned" — the name was found in the page text (the tooltip shows which name variant and in which snippet), "Not mentioned" — the page was read in full and the brand isn't there, "Unreadable" — the page couldn't be read. "Check again" reruns the check. A page AI cites that doesn't mention your brand is the first candidate to work on
  • "Winners & Losers" — pages with the biggest citation gains and drops versus the previous period, with the same brand-on-page check. The "new" badge marks URLs that weren't in the answers before, "dropped" marks the ones that disappeared
  • "Published content performance" — how your placed materials (campaign URLs) actually make it into AI answers: citation count, share of all citations, and how many prompts they surface on. A direct link between external publications and AI visibility
  • "Brand presence quality" — the distribution of answers mentioning the brand across five levels: mention, secondary option, with arguments, primary pick, and benchmark. Each level shows its share, answer count, and change in p.p. vs the previous period. "Details" leads to Brand perception
  • "Recent responses" — the latest AI answers to your prompts: which brands were mentioned, which sources were cited, and when the answer came in. Clicking a row opens the full answer text with mentions highlighted
Tip

Start your analysis with dynamics, not absolute values. A 16% SoV is excellent if it was 5% a month ago and alarming if it was 25%. Always look at the change vs the previous period and at the "Winners & Losers" block: it shows which sources started shaping the answers and which stopped. Your page under "Dropped" often means not a drop in quality but that a competitor's fresh material displaced it — and then the work is with that material, not with your page.

3.2

Competitive intelligence: the market overview

The Competitive intelligence page opens on the Overview tab — the full picture of your market inside AI answers: who overtakes you and when, how share of voice is split between brands, and which queries drop you out of the answer entirely. The switch in the top right leads to the second tab, Top competitors, with the full ranking, and the "Find gaps" button hands the job to the AI agent: it collects the prompts where competitors appear and your brand doesn't, and shows which sources AI relies on in those answers.

1/2
  • "Competitor mention trends" — a line per brand with a "Show top 5 / 10 / 15 / 20" switch. The tooltip on any date gives the exact mention count for every brand that day, so spikes and dips are easy to line up with releases, publications, and model updates
  • The legend below the chart is interactive: hovering highlights a brand's line, clicking opens that competitor's card. Handy for isolating one player when the chart is crowded
  • "Share of Voice over time" — a stacked chart of how every brand's share is split and shifts over time, including a combined "Others" group. You see who is eating into your slice and whose share is shrinking in your favour
  • "AI visibility benchmark" — a radar pitting your brand against the top competitors across four metrics at once: mentions, position, citations, and sentiment. The shape reveals the imbalance — plenty of mentions but a domain that's barely cited, for example
  • "Competitive gaps" — prompts where AI mentions competitors but not your brand; the header shows how many gaps were found. Each row gives the gap size in percent, "Absent in 7 of 7" answers, the number of answers and competitor mentions, the providers, and the top competitors for that specific query
  • Gaps are sorted by size: the higher the percentage, the more often competitors land in the answer instead of you. Clicking a row opens the whole prompt — with the AI answers and the sources they were built from
Tip

Don't try to close every gap at once. Sort them by size and by query type — a commercial "where to buy / order" beats an informational one — and start with the top 3. Every closed gap adds to your Share of Voice instead of a vague "we're losing".

3.3

Top competitors: the full ranking of your niche

The second tab of Competitive intelligence is the full list of every player the AI engines mention on your prompts — in an active niche that easily runs past a hundred. The ranking shows who is most visible, who is growing, and where your brand stands among them.

  • The header shows how many competitors were found in total. Columns: rank, trend ("New" / moved up / moved down versus the previous period), name with domain and a link to the site, mentions, share of voice (bar and %), and average position in the answer. Your brand carries a "Your brand" badge
  • The table answers different questions: mentions — who is most visible; trend — who is new or growing fast; average position — who consistently lands in the opening lines; share of voice — each player's weight in the market
  • Search by name and domain plus pagination at the bottom: a fast way to find a specific player — a newcomer from a report or a brand from a case study
  • The toggle in each row turns tracking of a competitor on and off: a disabled competitor takes no part in comparisons. Tick several rows with the checkboxes to enable or disable them in one go
  • Clicking a name opens the competitor card with a head-to-head comparison. The "⋮" menu holds "View details", "Merge duplicates" (fold variants of one brand into a single competitor), and "This is my brand" (if the engines used that name for your own brand, it moves into your brand's name variants together with its mentions). Duplicates are also merged automatically in the background; "Merge history" at the top shows what was merged and when, and lets you undo a merge
Tip

Review the whole ranking every few weeks — the AI engines keep finding new players on your queries. A new name near the top usually means a competitor just entered your niche or started publishing heavily: an early signal worth checking in their card.

3.4

Competitor card: the head-to-head view

Clicking a competitor — in the ranking, in a chart legend, or in a gap — opens their card. It's one rival taken apart: a direct comparison with your brand, shared trends, the gaps specific to them, the pages they're cited from, and the ads they run on your prompts. The "Review this competitor" button hands the same job to the AI agent: it compares the competitor with your brand and shows where and why they win.

1/3
  • The header carries the "Active" toggle (whether the competitor takes part in comparisons), an origin label ("Automatic" or "Manual"), and the "Also known as" and "Known domains" fields. Add alternative names and domains there so the competitor's mentions and citations are counted as accurately as your own brand's; remove a wrong name variant with the cross
  • "Brand comparison" — two columns, you versus the competitor: Share of Voice, mention count and average position, "Recommended as best", and "Domain cited" (how often AI links to their site)
  • "Presence quality" — how AI presents each of the two brands: a bar split by level (benchmark, primary pick, with arguments, secondary option, mention) and the share of strong levels. Brand and competitor are scored by the same rules; the "Brand perception" link leads to the full breakdown
  • Mention sentiment — positive, neutral, and negative percentages for both brands. It matters most when Share of Voice is close: AI mentions one positively and the other neutrally or critically, and the customer picks the first
  • At the bottom of the block sits "Gap": a summary reading "Your brand is ahead by X%" or "Competitor is ahead by X%". On the right — mention trends for the two brands only, yours versus the selected one, with the points where things sped up or slowed down
  • "Competitive gaps", now narrowed to this competitor: prompts where they land in the answer and you don't, with the gap size and providers. The exact places worth winning visibility back from them
  • "Sources citing this competitor" — the pages AI pulls facts about them from: page title and address, providers, the brands mentioned (the mark on your logo shows whether your brand is on the page itself), and the citation count. The "On the page" column shows whether the competitor is actually named there, with the first snippet found; hovering reveals every snippet with highlights. Pages where the competitor is present and you aren't are a ready-made list of places to work on
  • "Competitor ads" — ad creatives caught on your monitored prompts: the ad copy, the prompt, and the impression count. Clicking one opens the full AI answer — with its sources, the "cited" / "brand" / "competitor" labels, and the ad block in context
Tip

Sentiment is routinely underrated. If your Share of Voice is 20% but a sizeable slice of the mentions is negative, the AI engines are talking about you in a bad light, and the customer will pick the competitor with the smaller share but the better image. Handling negativity is as much a part of GEO as growing mention counts.

3.5

Filters: prompt groups × providers

The analytics top bar has two filters: the period and "Filters" — the "Coverage by group and provider" matrix that replaced the separate provider and prompt dropdowns. A selection applies to the whole dashboard and every analytics section at once, letting you look at a specific slice of the data rather than just the aggregate.

  • Rows are prompt groups (the number on the right is how many prompts the group holds), columns are providers, and each cell is your brand's coverage in that group on that provider. The "All" column and row give the totals, the header shows the average coverage. The deeper the cell color, the higher the coverage: weak spots show up before you filter anything
  • One click narrows the slice: a group name — only its prompts, a provider icon — only that provider, a cell — that group on that provider. Clicking again clears the choice, and "Reset" brings back every group and provider
  • "Select multiple" switches to checkbox mode: build any combination of groups and providers and press "Save". When there are many groups, a group search appears; providers not included in your plan are marked with a lock
  • Filters apply globally: once you choose, every KPI, chart, competitor ranking, gap, and source shows data for that slice only, and the "Filters" button shows what's selected. Plus the period filter ("Last 30 days" by default)
Tip

Use the matrix as a diagnostic tool before you even filter. Average coverage may look acceptable, while the rows show at a glance that one group sags on every provider and another is strong everywhere. Click a pale cell and the whole dashboard shows who takes that spot instead of you, and on the strength of which sources.

04 · 4 sections

Brand perception: the role AI assigns to your brand

Being mentioned is not enough. This section shows whether AI recommends the brand or names it as a fallback, who gets the recommendation, which qualities describe you and which pages that comes from

4.1

Presence quality: five levels

AI may call your brand the benchmark or a fallback option with a caveat. The Presence quality block puts every answer mentioning the brand on one of five levels and shows the split for the period: the bar at the top, and below it the same ladder per provider and versus competitors. The section uses the same period, prompt-group and provider filters as the home page.

  • Benchmark — other named brands are measured against the brand (“X falls short of the brand”), or AI directly recommends it and explicitly rates it above a named rival (“better than X”). List order does not count. Primary pick — AI directly recommends the brand as its choice (“my favourite”, “I recommend”) without the signals required for Benchmark
  • With arguments — recommended under a condition ("if you need a quiet PC, take X") or with concrete reasons to choose it. This is the level that content grows most easily
  • Secondary option — the brand is called weaker than a rival or with negative sentiment, or the direct recommendation in that answer went to someone else. Mention — merely named, without arguments or comparison
  • Hover the bar or a provider's mini bar: the tooltip shows the percentage and count per level. The Strong column = benchmark + primary + with arguments, Secondary = secondary option
  • The Versus competitors tab shows the same levels for the top competitors by mentions. The brand and competitors are scored by the same rules, so the shares are directly comparable: whoever has 30% strong on the same mentions is winning the recommendation
Tip

Don't chase the mention rate: 1,000 mentions with 70% "merely named" are weaker than 300 mentions with 40% strong. The strong share is your perception metric.

4.2

Presence quality over time

The answer to the main GEO question — did it get better after the changes. The chart below the ladder shows how the levels move over time: by day for short ranges, by week and month for long ones.

  • Brand levels over time — a stacked share of the five levels among the brand's mentions in each period. The wider the green, teal and blue bands at the bottom, the more often AI recommends the brand rather than just naming it
  • Strong share: brand vs competitors — lines of the strong-level share for the brand and the top 5 competitors by mentions. You see who is growing at your expense; clicking a competitor in the legend opens its card
  • Each point's tooltip shows the mentions and answers: periods with a handful of mentions swing sharply, so use weekly and monthly buckets for conclusions
Tip

Note the date a page was published or fixed and compare the shares after two or three runs: With arguments and Primary pick should grow while Secondary option falls. If only Mention grows, the content was noticed but AI found no arguments in it.

4.3

Who gets the recommendation and what AI says against you

Two blocks under the ladder explain why the brand does not get the recommendation: who exactly takes it and in what words AI places the brand lower. Clicking a competitor keeps only the reservations from answers where the recommendation went to it. The same blocks are on the dashboard in the Reservations tab, and for a single prompt in the Perception tab on the prompt page.

  • Who gets the recommendation — competitors AI recommends outright or names stronger than your brand in answers where your brand is mentioned but not directly recommended: the number of such answers (the tooltip splits out direct recommendations), their share of the brand mentions and the qualities they are praised for
  • The “brand is called weaker” note — in how many of those answers AI also explained why your brand is weaker. These are the most valuable answers to review
  • AI's reservations about the brand — verbatim wordings: "lags behind X in cooling", "comparable to Y, but…", reasons for a negative. A weaker / on par / caveat / negative badge, provider, date and prompt; click opens the full answer
  • The Sources of negative answers button shows the pages AI cited in negative answers, with a verdict on whether the brand is mentioned on them
Tip

Two or three recurring reservations about the same thing ("lags behind in cooling") point to a topic to close with content or fix on the cited pages more precisely than any metric.

4.4

Attributes, gaps and the sources of wording

The lower part of the page is about the words AI uses to describe the brand and competitors, and where those words come from. Qualities are extracted from the answers themselves and grouped into attributes; objections ("expensive", "noisy") are listed separately.

  • How AI describes the brand — the brand's top attributes with the number of answers describing it that way; Brand shape — a per-attribute radar with relative axes: the leader on an attribute among the selected brands is 100%, the tooltip shows the share and the raw score; objections are excluded
  • What explains competitors' choice — attributes where a competitor leads the market and your brand is lower or not described that way at all. A ready-made topic list for the content plan
  • “Brand comparison by attribute” (the heatmap) — the brand and the mention leaders against every attribute: where you beat the market, where you lag, which objections are attributed to you
  • Clicking any attribute opens the "Where AI gets…" panel: the domains and pages AI cited in answers with that wording, with their share of answers and the page check result. The "brand on page · N snippets" badge expands the snippets themselves, with the brand highlighted and a note on where on the page each was found. In the panel for your own brand's attribute, pages verified not to mention the brand are hidden, and pages that do mention it come first
  • The same data is available to the in-dashboard AI agent and via MCP: ask "compare presence quality with ASUS over the month" or "where does AI get 'noisy' about us"
Tip

Start with gaps where a competitor leads and you have no attribute at all, and — in competitors' attribute panels — with pages marked "brand not on the page": these are the cheapest fixes — AI already reads those pages, it simply has nothing about you to take from them.

05 · 7 sections

Accuracy of AI answers about your brand

Compare AI answers with company facts: add reference information, select up to 20 prompts, start checking and investigate contradictions.

5.1

Why check facts in AI answers

AI can recommend your brand while getting its warranty, service area or product specifications wrong. Accuracy compares statements in collected answers with the facts you provide, helping you identify a specific mismatch and decide which public information needs clarification. The screenshots below show the live interface with real answers about Yamaha Motor collected on November 2, 2026. This is a GEO Scout demonstration, not a Yamaha customer case study. The screenshots use the Russian-language interface.

  • Open Dashboard → Accuracy, or the Accuracy tab in Brand perception. The feature currently has a Beta badge.
  • For example, your fact says the warranty is 24 months while AI says 12 months. Open the contradiction to see the statement, reference fact and original answer.
  • Your saved facts are the reference for comparison. Check that they are accurate and current before using the results.
Tip

An omitted fact is not automatically an error. The checker compares statements AI actually makes; it does not require every answer to repeat all your brand facts.

5.2

Add company facts

Click Brand facts → Add facts. Start with information customers rely on: what you offer, where you operate, and which terms or limitations apply. The screenshot shows five Yamaha Motor facts with official sources, from business areas to a specific maintenance requirement for its water purification system.

  • Manual entry: put one fact on each line. Keep delivery coverage and warranty duration as separate statements, then click Save facts.
  • CSV import: put facts in the first column and use the first row as a header. Review and edit the imported text before saving.
  • From knowledge base: select documents or paste source text, then click Prepare facts. Review the proposed statements before saving them.
  • You can enable up to 50 facts, each up to 1,000 characters. Disabling a fact frees a slot for another one. Use the checking switch to temporarily exclude a fact and the pencil to edit it.
Tip

Use specific, verifiable statements: “Model X has a 24-month warranty” is more useful than “We have the best warranty.” Saved changes apply to subsequent checks.

5.3

Select existing monitoring prompts

Click Choose prompts, check the relevant rows and save your selection. These are existing monitoring prompts: there is no need to create a separate cluster or special accuracy questions. This demonstration selects three questions about Yamaha Motor’s business, water purification, and its differences from Yamaha Corporation.

  • Select queries whose answers describe your brand, products or terms. Use search, the topic column and prompt status to find them.
  • Every subscription tier includes 20 checked prompts. This is the number of selected prompts, not the number of facts, errors or individual answers.
  • Need more than 20 prompts? Contact the support chat or support@geoscout.pro. Partner accounts use per-answer credits rather than the subscription prompt limit.
  • Monitoring keeps collecting answers on its own schedule. Selecting prompts alone does not generate new answers or activate fact checking.
Tip

If the list is empty, add prompts in Monitoring first. If no answers have been collected yet, wait for monitoring to finish: there is nothing to compare until then.

5.4

Start checking: subscriptions and partners

After adding facts and choosing prompts, the launch process depends on your account. Accuracy checks saved answers; Monitoring remains responsible for collecting new ones.

  • With a subscription, click Start checking in Accuracy. Buying a subscription or starting monitoring does not activate fact checking on its own.
  • After the first launch, checks run automatically once a week without spending credits. The section shows the next check date in your time zone. An active subscription, enabled facts and selected active prompts are required.
  • For partners, click Check answers. The dialog shows eligible saved answers for the selected period and filters, the cost and available balance. Review the amount and confirm the launch.
  • Partner pricing is 1 credit per successfully checked answer, regardless of the number of facts or errors. Answers already checked against the current facts are excluded. A batch can contain up to 500 answers; run another batch for the remainder.
  • Partner checks do not follow a weekly schedule. Credits are reserved at launch and refunded if a check fails or a waiting check is cancelled. Cancelling the queue does not interrupt answers already being processed.
Tip

If the first-launch button is unavailable, check the subscription, enabled facts and selection of active prompts. Checks run in the background, so starting a run does not make results appear immediately.

5.5

Investigate a contradiction

After processing, counters show checked answers and statements. Contradictions lists statements that conflict with your facts. Start with recurring errors that could affect a customer’s decision.

  • Open a statement or its answer count. The statement panel shows the wording, the reference fact and answers containing the claim.
  • Read the quote in context: provider, date, prompt and full answer. Check results also appear in the answer card and full-answer panel.
  • Supported means agreement with a saved fact; Contradicted means a mismatch. Not related to the fact marks a statement that should not be compared with that fact; it does not confirm that the statement is true.
  • All checked statements can be grouped by fact or topic and filtered by assessment. Use page filters to compare periods, prompts and AI providers.
  • If an assessment is wrong, use Change assessment. Manual assessments are labelled separately, and you can restore the system assessment. If the underlying fact changed, update it in Brand facts.
Tip

“No contradictions found” covers only statements that could be compared with your facts. It does not guarantee that the entire answer is accurate.

5.6

Real example: Yamaha Motor and consumables

On November 2, 2026, we asked three questions in Russian using DeepSeek, Alice AI and Search with Alice. After refining five reference facts, we checked nine saved answers. The system matched 14 claims to facts and flagged two discrepancies concerning consumables. The first screenshot shows the DeepSeek finding; the second shows the same answer in context.

1/2
  • The question was “What water purification solutions does Yamaha Motor offer?” All three engines correctly named Yamaha Clean Water Supply System. To check more specific claims, we added a maintenance fact instead of relying only on a general business description.
  • Reference fact: operating YCW after installation requires periodic purchases of chlorine powder for disinfection. The official Yamaha specifications list this among the items the customer must arrange locally.
  • The DeepSeek answer included a phrase meaning “thanks to the absence of a need for chemical reagents”. The check flagged its conflict with the chlorine powder requirement and retained the original Russian quotation and answer.
  • Open the full answer: the same model mentions chlorination earlier. This is an inconsistency within the answer, not a wholly incorrect description of the system. Read the context before changing a verdict. A second answer was separately flagged for a broad claim about not needing consumables.
  • The practical value is finding wording that could give a buyer the wrong expectation. In your own product description, distinguish between not needing replacement membranes or coagulants and needing a disinfectant. Here, the official site already makes that distinction; a flagged answer does not mean a correct source page needs changing.
Tip

This is a small demonstration run on a partner account, not a measurement of overall model accuracy. The answers were not edited and the verdicts remain automatic. Results depend on the question, date and reference facts; another run may produce different wording.

5.7

Inspect sources and record corrections

Open Sources in answers in the statement panel. These links are associated with AI answers and can help you locate outdated or ambiguous public information.

  • Labels distinguish links next to the claim, links elsewhere in the answer and retrieved but uncited pages. Start with sources next to the relevant wording.
  • Open the page and read its content. A link in an answer does not prove that the incorrect statement came from that page. Some answers will have no available source links.
  • Update your own website or contact the external page owner. Add URLs of published or updated materials under Your corrections. This records your work; it does not automatically change an AI answer.
  • Review accuracy and Review statement open the agent with the relevant context. Ask it to compare answers with facts, investigate sources and suggest corrections. Facts, prompt selection and check results are also available through MCP.
Tip

After updating content, wait for new monitoring answers and the next check. Website edits do not rewrite saved answers or guarantee that AI will change its wording immediately.

06 · 5 sections

Working with sources

The map of domains and pages AI leans on in answers to your queries: who gets cited, what's rising and falling, and which brand pages AI sees but never cites

6.1

The source map: where the engines get their facts

The "Sources" section is the full map of domains AI links to in answers to your prompts. It answers the question "where do the engines take their facts from, and where does the brand need to be": until you're inside those sources, landing in an answer is a matter of luck. The "Where to publish" button in the header hands the choice to the AI agent: it picks the outlets AI already cites in your niche where your brand isn't present yet.

  • "Top cited domains" opens with a breakdown by category: Media, Directories, Competitors, Communities, Official, Other — each with its citation count and share. That's the structure of your niche on one screen: which type of venue the engines are willing to cite at all
  • The priority for work is Communities (forums, Q&A, blogs, social, video) and Media (news, business and tech press). You can't influence competitor sites, but these you can enter with a publication, a guest piece, or a mention. "Official" are your own domains — their share shows how much AI relies on you directly
  • The "Page type" filter shows which formats make it into answers: listicles and rankings, product pages, homepages, articles, how-tos and guides, reviews, directories, news, comparisons, discussions, documentation, FAQs, research, case studies — each with its citation count and share. It tells you which format to write in to get cited
  • The domain table: domain with its category tag and number of unique URLs, citations, trend versus the previous period, citation share, the providers linking to it, a "Brand" column, and "Priority"
  • "Brand" isn't a guess from the answer text but a check of the page itself: the system opens it and looks for the brand name. "Mentioned" — found (the tooltip shows the name variant and the snippet), "Not mentioned" — the page was read and the brand isn't on it, "Not checked" — the page hasn't been checked yet or didn't open. "Check again" reruns the check
  • "Priority" (0–100) — how urgently you need to get into the source or publish an alternative piece. It's computed from verified pages and grows when the source is cited on many of your prompts and competitors are on it while you aren't. Pages that could not be read ("Unreadable") are estimated from AI answers — which companies they tie to the page — and marked "estimate". Under a domain's score you see how many pages lack you; a page gets a "verified gap" label; the tooltip breaks the score down: how many prompts cite the source, how many competitors are on the page, citations, and why the number came out as it did. Priority isn't computed for your own sites or competitor sites
  • The switches above the table: "Cited only / All sources" — only domains that actually landed in an answer, or the whole pool the engines looked through; "Gap analysis" — only verified pages where your brand is absent while competitors AI named are present, sorted by priority; "Brand mentioned" — only sources from answers where AI named the brand, i.e. what the engines rely on when they talk about you. The period and "Filters" in the top bar narrow the slice
  • The arrow to the left of a domain expands its individual URLs — with page type, brand check, and priority; the arrow on a page shows "Prompts this page feeds" with the prompt group and citation count. The ✦ icon on a page row is "Generate ideas": the system breaks down the article and collects ideas for pieces on the same topic in the Content Studio
Tip

Don't try to be everywhere. Turn on "Gap analysis": the top rows become verified pages AI cites on many of your prompts where competitors already are and you aren't — that's your work queue. Competitor domains are useful differently: their pages show the format of material the engines are willing to cite in your niche.

6.2

URL dynamics: what's rising and what's dropping out of answers

Domains answer the "where"; individual pages answer the "what exactly" the engines cite. The two blocks below show the life of specific URLs over time: which pages hold their citations, which just appeared, and which fell out of the answers.

  • "Individual URL citations" is a daily chart for the top pages from AI answers. The "Show top" switch (5 / 10 / 15 / 20) controls how many lines are drawn, and the "Citation share / URL" toggle flips between domain shares and absolute page citations
  • The point tooltip lists that day's URLs ranked, with citation counts and the outlet's icon. A crown marks the leaders: the pages the engines cite most often
  • "Winners & Losers" splits pages into "Winners" and "Losers" by the strength of their citation change versus the previous period. Every row carries the domain, page type, the brand-on-page check ("Mentioned", "Not mentioned", "Unreadable" or "Checking…", plus "Check again"), providers, and a delta like "+3 (1 → 4)"
  • A "new" badge marks URLs that weren't in the answers before, "dropped" marks the ones that disappeared. That's an early signal: a competitor's fresh listicle the engines just picked up, or your own material that stopped landing in answers
  • The comparison period is spelled out in the block header, so it's unambiguous which two intervals the gains and losses are measured between
Tip

Check "Winners & Losers" weekly — it's the fastest way to spot a shift before it shows up in the headline metrics. Someone else's fresh listicle appearing under "Gained" usually means a new ranking shipped in your niche and it's worth getting into; your own URL under "Dropped" is a reason to look at what changed on the page or at competitors.

6.3

The platform card: is it worth going there

Clicking a domain opens the platform card — one source broken down in full. It settles a practical question: will a placement there give visibility to your brand specifically, or is the platform cited on your queries while only competitors are on its pages.

1/3
  • The header shows the domain with a "Platform" badge, an "Open site" link, and four KPIs: citation share, number of prompts, the providers citing it, and "Priority" — the tooltip explains the number, e.g. "23 verified pages without you but with competitors; they feed 14 of 65 prompts". The "Review this source" button hands the breakdown to the AI agent: what AI takes from the domain, which competitors are there, and how to get in
  • "Who is on this site's pages" — only brands found in the page text itself: your brand with a "You" badge and the share of pages it's on ("on 13 of 49 pages"), and competitors — with mention counts, pages, and the fragments that name them; "N more fragments" expands the rest. If you're absent from the checked pages while competitors are present, the block says so plainly
  • "Citation dynamics" is a daily chart by provider. You can see which engines rely on the platform constantly and which picked it up recently; the tooltip breaks the day down per provider
  • "Domain pages" lists every cited page of the platform with a type filter: each shows its providers, the brand check ("Mentioned" with the matched snippet in the tooltip, "Not mentioned", "Unreadable"), priority, and citation count. Sorting by priority brings up the pages where competitors are and you aren't
  • "Winners & Losers" repeats at the domain level: which pages of the platform gained or lost citations over the period, with the same brand check. A fresh listicle starting from zero is usually exactly the one worth getting into
  • "Prompts citing this source" shows the specific queries the platform surfaces on, with providers and citation counts. Clicking a prompt opens its card
  • Below sits "Responses citing this source": the AI answer texts themselves. A quick way to check in what context the platform lands in an answer and what exactly gets pulled from it
Tip

Read the priority together with "Who is on this site's pages". A high priority while your brand is on almost none of the pages and competitors are is the strongest argument to go there with a publication: the engines already trust the source, you're simply not in it. If the brand is already on most pages, the platform has given what it can — look for the next one.

6.4

The URL card: one page broken down

From the domain table or the platform card you can drill into a specific page. This is the narrowest level: one article or listicle and everything it delivers — whether your brand and competitors are on it, on which queries, and which engines cite it.

1/2
  • The header shows the page title, a link to it, the brand check result ("Mentioned", "Not mentioned" or "Unreadable", plus "Check again"), and the same KPIs as the platform but for a single URL: citation share, prompts, providers, and priority
  • "Brand mention on the page" — the snippets where your brand was found, highlighted and labelled with the part of the page it sits in: text, title, description, markup, link, or navigation. A mention in navigation alone is nothing like a paragraph about the brand in the body of the article
  • "Who is on this page" — brands found in the page text, with fragments: on a ranking listicle that's effectively the balance of power inside the material. If you're absent while competitors are present, the block says it plainly: AI reads this page — worth getting onto it
  • "Citation dynamics" shows the page's life over time by provider: one engine may keep it in answers for weeks while another reaches for it only occasionally. "Prompts citing this source" shows which queries the page works for; if there's just one, a placement there closes a narrow slice
  • The "Generate ideas" button breaks down the article and collects ideas for pieces on the same topic in the Content Studio: for media — in the format the outlet actually publishes, for a competitor's site — for your own site, fuller and more precise
Tip

Before negotiating a placement, open the target page at this level. If your brand is already on it but AI still prefers competitors, the problem isn't the outlet — it's how you're presented inside the material: look at the snippets in "Brand mention on the page" — sometimes updating the mention beats shipping a new publication.

6.5

Pages for optimization: found, but not cited

The section's second tab is "Pages for optimization". It collects pages of your own site that the AI models found but didn't cite in their answers: the interest is there, the link isn't. It's the cheapest growth point — no new venue to win over, just work on what already enters the engines' field of view.

  • The header states how many appearances happened across how many pages, with two counters next to it: how many pages went uncited and how many are already cited. The second is your base, the first is the work queue
  • The table shows the page with its title, a "Found" column (how many times the engines saw it without linking), and the providers that found it. Sorting by "Found" puts the biggest missed potential on top
  • Under each row sits "N of M closed" — progress across that page's prompts: how many queries optimization has already addressed and how many are still open
  • Expanding a row reveals "Prompts where the page wasn't cited" — the specific queries the page surfaced on without making the answer. Closed ones are struck through, so what's already been done is visible
  • The "Optimize" button opens the content studio with that page and immediately starts generating edits, an FAQ, and schema. If the page has been optimized before, the button becomes "Open" — you land in the existing result instead of paying for a rerun
  • The optimization process itself — the edits, priorities, and moving blocks onto the page — is covered in the "Own domain visibility" guide
Tip

Start from the top rows by "Found": a page the engines saw ten times and never cited pays off more than a perfect new publication. It already has the appearances — all it lacks is wording the model can quote.

07 · 2 sections

Improving visibility on your own domain

How the GEO Scout command center turns monitoring signals into concrete actions on your site: schema markup, content, and pillar pages with ready-to-use artifacts

7.1

Page optimization: turning appearances into citations

When a page already shows up in AI answers but isn't cited, the command center raises a "Improve this page" action. AI knows about the page but doesn't consider it relevant enough to link to. Fixing exactly that page is the fastest way to lift brand visibility and direct traffic: you convert an appearance into a citation where the engines are already interested.

1/3
  • The command center is split across three tabs — "Site changes", "External publications", and "Publication plan" — with a type filter and a "Pending / Completed" switch. Actions are grouped ("Content", for example), and each card shows the providers where the page surfaced and an effort estimate
  • The "Improve this page" action appears when a page has already landed in AI answers without being cited: the card states how many times that happened and which engines saw it
  • Inside the card sits the "Prompts where the page wasn't cited" list with a closed-count. Those are exactly the queries the page needs to be reworked for — not an abstract "improve SEO"
  • The "Optimize" button opens the content studio with the source page already filled in — from the brand's sitemap or entered manually by URL (the domain must match the brand's site) — plus the "Target queries" block. "Generation history" next to it keeps every previous run
  • The agent works in real time and shows its steps: load brand → scrape page → AI rewrite (the LLM rewrites the page and generates an FAQ and a schema bundle) → save to history
  • The result opens with the "Optimization strategy": a short summary of what changes and why, plus the number of edits by priority — Critical, High, Medium
  • Every edit is its own card with a type (citability, FAQ, and so on) and a priority: WHERE to insert it (section and snippet), what's there NOW, what it SHOULD BECOME or what to INSERT, and a note on which query it works for. The "Copy" button hands over the ready block — paste it onto the page and mark the action as done
Tip

Don't rewrite the page "in general". The edits target specific uncited queries — paste in the ready paragraph that answers the prompt's intent directly, and the page starts answering the exact query AI was skipping it on. Closing query after query is how domain citability grows systematically.

7.2

Technical groundwork: Schema, llms.txt, and AI-bot access

For the AI engines to confidently read and cite your site, GEO Scout watches the technical groundwork of AI visibility: structured JSON-LD markup, an llms.txt file, and AI-bot access in robots.txt. These actions are gathered in the command center under the "Schema markup" and "Technical" groups.

1/2
  • Schema markup: GEO Scout finds pages missing the right JSON-LD types (e.g. TechArticle on /docs/payment-solution or ContactPage on /contacts) and explains why it matters — JSON-LD raises the odds of landing in AI Overviews, rich results, and AI-answer citations
  • The "Generate Schema" button (1 credit) creates the missing markup for a specific page. The result is a ready JSON-LD block with real brand data (name, description, contacts, opening hours); just paste it onto the page and mark it "Done"
  • llms.txt: if the site root has no /llms.txt, GEO Scout offers to generate it. It's an emerging standard for AI assistants — a short Markdown map of your site's important pages that helps ChatGPT, Claude, and Perplexity understand your content
  • The ready llms.txt can be copied or downloaded and uploaded to the site root as /llms.txt. Inside is a list of key pages with descriptions (payment solutions, API, guides, etc.), built from your site
  • robots.txt: GEO Scout checks whether AI bots are blocked (GPTBot, ClaudeBot, PerplexityBot, Google-Extended, YandexBot, and others). If a bot is blocked, you get a notification: the engine simply can't read your site and cite you, no matter how well the content is written
  • Technical actions are tagged "Quick fix" / "Fast": low effort, but they remove the barriers without which the rest of the visibility work has no effect
Tip

Clear the technical barriers first, then work on content. If an AI bot is blocked in robots.txt, the engine simply can't read the page — and even perfect copy won't reach the answers. Bot access and llms.txt are the foundation that page improvements and publications rest on.

08 · 5 sections

Improving visibility via external publications and mentions

How the command center turns the source-domain map into a plan for external placements: the main recommendation is the "outlet × topic" pair, and the content plan serves as a brief for the editorial team

8.1

External publications: where the brand needs to appear beyond its own site

The "External publications" tab in the command center collects recommendations on where the brand needs to appear outside its own website. For most commercial queries, AI engines cite external sources — media, directories, communities, expert domains — far more often than brand sites themselves, so presence there directly decides whether the brand makes it into the answer.

1/2
  • The recommendations grow straight out of monitoring. The system takes the prompts and providers where brand visibility sags — a low "Response coverage" — and looks at which sources the engines lean on in the answers to those exact queries. All of it is visible in the prompt card: the metrics on top and the "Sources" tab with the domains that made it into the answer. Those domains are what becomes the outlets in the recommendations
  • The tab sits second in the command center — next to "Site changes" and "Publication plan" — with a recommendation counter, an outlet-type filter, and a "Pending / Completed" switch
  • Recommendations are grouped by outlet type — "Publications", "Directories", "Reviews", "Authority". The group header shows how many sources it holds, how many queries they surface on, how many citations they produced, and on which engines
  • One outlet, one card. The header carries the linked domain, a type badge ("Community", "Editorial placement", "Directory", "Reviews", "Authority link"), the number of queries, citations, and articles, plus provider icons. Below it a line explains why this outlet type works — detailed community answers, for instance, get cited by LLMs as "lived experience"
  • Inside the card sit the outlet's specific materials, each expandable on its own: "What to do" (for example: "Perplexity and Alice AI already rely on dtf.ru in answers to your brand's queries. The brand is barely present there"), "Why it matters", "Affected queries" with coverage percentages, and "Citation examples" — the real URLs the engines already cite
  • The recommendations aren't abstract: GEO Scout takes the domains AI already cites for your queries (the same ones in the "Sources" section) and suggests strengthening presence where the model already looks for confirmation — not wherever "it would be nice to be published"
  • The "Generate ideas" button launches an AI agent: it studies the outlet's format, pulls brand facts from the knowledge base, and proposes several publication topics with an article outline. Next to it: "Done", snooze, hide the card, and flag the recommendation
  • Each type is its own workstream: editorial placements go to PR, directories to marketing, communities to the community manager, reviews to support. For directories and review sites the job is simpler — check that the brand's listing exists and its data is current
Tip

Don't treat external publications as an add-on to on-site work. For many niches this channel lifts citations faster than anything you change on your own site: AI already trusts these sources, you just need to be inside the right material. Editorial placements are the strongest — one article on an outlet the model already cites quickly raises the brand's share of mentions.

8.2

Publication plan: the order in which outlets deliver the most visibility

"Publication plan" is the most important tab in the command center. It takes the same external-outlet recommendations and lines them up by impact on visibility: step 1 is the publication that closes the most queries where the brand is missing; step 2 adds the most on top of the first, and so on. Instead of a list of "places it'd be nice to be published", you get an order of work in which every next step yields less than the previous one.

  • The plan's unit of measure is the "query × AI provider" pair. The same topic can be answered differently in ChatGPT and in Alice, so coverage is counted per pair rather than per query — that's how the real visibility gap shows up instead of an average
  • The header answers the main question in one line: how many publications close how many unique queries (out of all monitored ones) across how many AI providers. Next to it, a coverage bar by plan step shows each step's contribution as its own segment
  • The order is built on marginal coverage: each next publication closes the maximum of still-uncovered pairs. Steps are numbered, outlets influencing different providers don't crowd each other out, and an "N shared queries with step" badge honestly flags overlap with an earlier step. Every step states how many new queries it adds and how many pairs are closed cumulatively — "after this step, X of Y are closed"
  • A step reads as a task: "You need a mention or your own content similar to …" — followed by the specific outlet material the engines already cite — with an effort estimate, per-query coverage percentages, and citation examples the engines already use. A step can have alternative outlets, paged through as "Alternative 2 of 6 — if the primary outlet isn't available": if one editorial team says no, you take a replacement without losing the step's coverage
  • Only the top steps are shown — the ones carrying most of the coverage; beyond them each publication's return drops noticeably. Nothing is deleted: "Show all" expands the full list and "Collapse to top steps" brings the short version back
  • Below the plan sits the "No query data" block: recommendations with no link to prompts, which therefore can't be ranked by visibility gain. They aren't discarded — they simply stay aside so the queue isn't diluted
  • The work closes the loop right there: "Generate ideas" on a step produces topics and outlines for future materials, marking it "Done" moves the step to the "Completed" tab, and the publication's effect later shows up in marketing campaigns
Tip

Work strictly top to bottom and don't try to close the whole list. The first few steps deliver most of the gain — spending effort on the tail before the top is done isn't worth it. And don't rank steps by how prestigious an outlet feels: the plan is already sorted by how much new visibility each publication brings.

8.3

Publication ideas and article generation

From an external-publication card (or a plan step) GEO Scout assembles publication ideas with one button, and turns each idea into a finished article. The logic isn't "write some text" but to lean on what the engines already cite: the agent studies the format that works on that outlet and builds briefs for it.

1/3
  • "Generate ideas" launches the agent right inside the card, and its steps are visible live: analyze the recommendation → visit the outlet and study its format → gather brand facts → prepare topics and article outlines → save. Usually 1–2 minutes; the credit cost is shown before the run
  • The result is a set of publication ideas in the content studio. Each idea has a title, a priority (High / Medium / Low), and a format: "Listicle", "Comparison", "Guide", "How-to", "FAQ", "News". One topic can be covered by several ideas from different angles — that's how you catch more of the phrasings the engines search on
  • "Target prompts" — the queries the content must answer. These are the same monitoring prompts the outlet already gets cited on, so the article is written against a specific visibility gap rather than "the topic in general"
  • "Article outline" — the skeleton of the future piece: sections with their key points, and citable claims already distributed across sections. A brief like this can go to a copywriter or the outlet's editors as is
  • "Citable claims" — what AI providers can quote: concrete numbers and wording about the brand pulled from the knowledge base rather than the model's general knowledge. Next to it, "FAQ questions" — the block engines are especially eager to break down into answers
  • "Recommended placements" — where to take the piece. An idea grown from an external-publication card carries a single placement: the very outlet the recommendation came from. Next to it sit its type (editorial placement, community, directory) and a note on what the suggestion rests on — "Confirmed in AI responses" or "Signal-based suggestion"
  • The article-generation button delivers the finished piece in 30–60 seconds: it opens in a panel with "Article" and "Schemas" tabs, the generation timestamp, and exports — "Copy article", "Download", "Copy schemas" (the JSON-LD markup for the piece)
Tip

An article from an idea is a starting point, not a publish-ready piece. Fact-check the claims and figures and adapt the text to the outlet's requirements before sending it to editors. And the fuller the brand knowledge base, the less the model has to improvise — fill it in before the first generation.

8.4

Knowledge base: verified facts for generation

The knowledge base lives in the content studio — the "Knowledge base" tab next to "Drafts", "Published", and "Campaigns". It holds the verified sources that briefs and articles rely on: site pages, links, and uploaded files. Facts for the content come from here rather than being invented: you index your brand materials once, and every generation then draws on them.

1/3
  • The summary at the top shows the state of the base: how many sources are indexed, how many chunks there are in total, and when the base was last updated. It also restates the cost — indexing and re-indexing take 1 credit per source
  • Three source types: "Add link" (any page by URL), "Add site pages" (pick from the brand's sitemap), and "Upload files" (a deck, a guide, a price list). Everything important about the brand lives in one place
  • Each source is split into chunks the model later uses to find exact facts. The table shows the source title and address, the status, the chunk count, and the update date; columns are sortable, and the source search helps once the list grows
  • Statuses run "Not indexed" → "Queued" → "Indexing" → "Indexed". "Failed" means the page couldn't be read (blocked, redirected, empty), "Stale" means the source changed and is worth re-indexing. Tick rows with the checkboxes to index or re-index them in bulk
  • Indexed sources are blended into brief and article generation as verified brand facts. These are the "Citable facts" from the content plan — numbers and wording come from your materials, not from the model's general knowledge
  • Example: the brand site states "over 322,000 stores and services accept payments with YooKassa". Once indexed, that fact lands in the generated article verbatim and correctly — no risk of the model inventing a different number
  • The fuller the knowledge base, the more accurate and convincing the content: the model doesn't hallucinate but assembles the material from confirmed facts about your brand — which directly raises the chance of being cited and the trust in the publication
Tip

Load the key facts up front — numbers, pricing, advantages, differences from competitors — and re-index a source whenever the page's data changes. Then every generated article speaks about the brand with the same accuracy, and you won't have to fix facts by hand after each generation.

8.5

Marketing campaigns: tracking the effect of publications

Once a piece is live — on an external outlet or on your own site — add its URL to "Marketing campaigns" (a content studio tab) and track how the engines react: how often it lands in answers, which prompts it works on, and whether the brand is actually named where the piece is cited. This is where the loop closes: recommendation → publication → measurable effect.

  • The "Add URL" button takes both external placements (an article on an outlet, a directory listing, a community thread) and pages of your own site. A campaign groups placements within one release, PR wave, or quarterly plan
  • Four campaign KPI cards: "Citations" — how many times the URLs landed in AI answers; "Citation rate" — what share of answers on the monitored prompts contains them; "Prompts" — how many distinct queries AI used the material on; "Brand mentions in citing responses" — in what share of citing answers the brand is actually named
  • That last metric is the most underrated: an engine can link to your article without naming the brand. 100% means every citation works on recognition; a low share signals the brand is mentioned too much in passing inside the material
  • "Citation dynamics" — a daily chart broken down by provider: you can see when ChatGPT picked the piece up, when Alice AI did, and where a provider ignores it. The point tooltip breaks the day down per provider
  • The "URL sources" tab is a table of tracked links grouped by domain: citations, share of answers, and prompt count. A domain row expands to the individual URLs — showing which material actually earns citations and which sits idle
  • The "Prompts" and "Responses" tabs show which queries cite the placement and what the answers themselves look like. If the piece closes priority prompts, the loop is complete; if it lands on irrelevant ones, that's a signal to rework the content or the topic
Tip

This is the metric that makes GEO measurable: recommendation → publication → URL in a campaign → citations in the next monitoring cycle. Mind the lag: it takes 2–3 weeks on average (depending on the outlet) before a piece starts being cited — the engines need time to index it and pull it into their sources. If a URL still gains no citations after that, rework the material or try a different outlet.

09 · 1 section

Brand reputation monitoring in AI

How the "Reputation" tab on the home page shows the tone in which the AI engines talk about your brand, and exactly where negativity surfaces

9.1

The "Reputation" tab: where and why AI talks about the brand negatively

A mention isn't the same as a positive mention. The "Reputation" tab on the home page (next to "Overview" and "Reports") shows not just that the brand is mentioned, but in what tone — and gathers every negative AI answer into one feed with an explanation of what exactly is negative in each.

  • At the top — a reputation-risk summary: "Negatives" (count of negative answers), "Share of negatives among mentions" (%), and "Worst provider" — the engine that most often speaks about the brand negatively
  • «Negative answers» is a stream of every response with negative sentiment. Each shows the prompt («best platforms to order design in Russia»), the provider (DeepSeek, Claude…), when it happened, and «Negative» and «Brand mentioned» badges
  • «Why negative» — the system explains what exactly works against the brand in the answer: for example, «the author flags the platform's high competition and its 20 percent commission as significant drawbacks». No need to reread the whole response by hand
  • You see the full answer text with the brand highlighted and a "Sources" block — the pages the model leaned on to form the negativity. That's the lead: negativity usually traces back to a specific outdated review or a competitor comparison
  • Why it matters: an engine can consistently frame the brand negatively on specific queries and providers, which coverage-level dashboards don't reveal. Here you see exactly where, why, and from which sources — so you can address it with content and publications
Tip

Start working on negativity from the "Sources" block, not the answer itself. Open what the model referenced: it's often an old review or a competitor's roundup. Closing the negativity means giving the engine a fresh, more authoritative piece it will cite instead of the outdated one.

10 · 3 sections

AI advertising

How the "AI advertising" tab surfaces the ad blocks AI engines place next to the answers on your prompts — and who is buying them

10.1

Why watch the ads next to AI answers

AI engines increasingly show paid ads right next to the organic answer — and those ads compete for the same attention as your brand mention. The "AI advertising" tab collects every ad block seen on your prompts: you can see who buys promotion around your topics and how often their ads surface next to the answers.

  • Ads in AI answers are a separate layer from organic: even when your brand is mentioned in the answer itself, a competitor's paid block can sit above or below it on the same query
  • GEO Scout captures these blocks automatically during daily monitoring runs — there's nothing extra to collect, ads land in the section alongside the regular answers
  • For now, Alice AI is tracked: Yandex Direct text ads labeled "Promo" and sponsored product carousels
  • The section shows the picture for the period selected top-right ("Last 30 days"), so you can watch how ad pressure on your prompts changes over time
  • An empty section is fine — it just means no ads have appeared on your prompts yet. Data accumulates over time and can't be backfilled
Tip

A competitor advertising on your branded or comparison prompt is a signal the topic is already being monetized. That's a cue to strengthen your organic presence there: the more confidently the AI answers with your brand, the less weight the paid block beside it carries.

10.2

Metrics and advertisers: how to read the section

Up top are four cards summarizing ad pressure for the period; below is the "Advertisers" table showing exactly who buys impressions on your prompts.

  • "Answers with ads" — how many answers in the sample carried at least one ad block (e.g. 19 of 50). The higher the share, the more densely your prompts are plastered with paid ads
  • "Ad blocks" — the total number of blocks shown next to answers for the period: a single answer can carry several ads at once
  • "Advertisers" — how many unique domains advertised on your prompts. This is your list of who's competing for attention in your niche
  • "Product cards" — how many products landed in Alice's sponsored carousels (relevant for e-commerce; usually 0 for services)
  • The "Advertisers" table shows, per domain: number of impressions, how many prompts it appeared on, formats (e.g. "Direct"), and the date of the last impression — sortable by clicking a header
Tip

Focus not on one-off impressions but on advertisers with a high prompt count: if a single domain buys ads across many of your topics, that's your systemic competitor for paid traffic from AI engines.

10.3

Ad formats and the answer card

Open any answer from the list and you see exactly how the ad block looked: its format, label, and placement relative to the organic answer.

  • "Direct" — a Yandex Direct text ad labeled "Promo": headline, advertiser domain, description, and sitelinks (quick links to site sections)
  • Sponsored product carousels — a strip of products with prices inside Alice's answer; feeds the "Product cards" metric
  • Each block shows its "Placement" — above or below the answer: this affects how much the ad intercepts attention ahead of the organic result
  • Ad blocks appear right in the answer card next to the organic sources — so you can compare who is cited for free and who paid to show up on the same prompt
Tip

Use the section as reconnaissance on the paid channel: domains that consistently buy "Promo" on your queries reveal which topics AI engines already treat as commercially valuable — and where growing your organic visibility pays off most.

11 · 6 sections

E-com: product carousels in AI

How the E-com section shows the product carousels Alice AI and ChatGPT assemble on your prompts: which products and stores make it in, where your products rank, and how to link your own catalog

11.1

Why track product carousels

On shopping queries (“which monitor for work”, “best vacuums under $300”) AI engines increasingly add a product carousel to the text answer: photo, price, store, a buy button. It is a separate visibility layer: your brand may be mentioned in the text, yet the carousel next to it shows competitors’ products, and that is where the user clicks through to a store. The E-com section collects every such carousel from your prompts and shows who is in it and where.

  • Carousels are captured automatically during daily monitoring runs — nothing to configure, data accumulates from the moment the brand is connected
  • Two providers currently show product carousels: Alice AI and ChatGPT. Google AI Overview and Yandex search return no product cards on Russian queries; Yandex carousels are ads only and live in the “AI advertising” section
  • Every card is recorded with its position, price, store and, when present, rating and badge — all section metrics are computed from these
  • Products are grouped by card title: the same model across stores and providers counts as one product, different configurations count as different ones
  • Carousels cannot be backfilled: an empty section means none appeared on your prompts yet — add shopping queries to monitoring and let runs accumulate
Tip

Once a carousel appears on a prompt, the engine already treats the query as purchase intent. Position decides the rest: users rarely scroll past the third card, so the goal is not just to get in but to hold one of the first three slots.

11.2

Metrics and chart: reading the overview

At the top of the overview are four cards for the selected period; below them, the share of answers with a carousel per provider and the Top 7 products AI shows most often.

  • “Answers with carousel” — the share of non-empty Alice AI and ChatGPT answers that contained a product carousel (e.g. 40% — 896 of 2,259). This is the size of the e-com layer on your prompts
  • “Carousels” and “Unique products” — how many carousels and distinct products appeared; “Stores” — how many unique stores AI offers as the place to buy
  • The chart shows the share of answers with a carousel per day (per week on long ranges) for each provider; dips are days when the provider answered in text, a gap in the line is a day without runs
  • Each Top 7 product has three numbers: visibility (share of carousels it appears in), average position, and the share of carousels where it was first; the “Mine” badge marks products from your catalog
  • The provider switch at the top right narrows the whole page to Alice AI or ChatGPT — their assortments and stores differ noticeably
Tip

Compare providers separately: ChatGPT tends to build carousels from large marketplaces, Alice AI from Russian retailers and niche shops. If your product shows up in only one of them, the other is your growth point.

11.3

Showcase, products and stores

Below the chart is the Showcase with product cards as the shopper sees them, lists of stores and prompts with carousels, and full tables beneath. Every product and store has its own page.

1/4
  • The Showcase shows the best (or worst) products of the period with photo, price, visibility and average position — a quick view of who dominates the shelf
  • “Stores” — where AI sends people to buy: number of cards and the store’s share; clicking opens the store page with its products and prompts
  • “Prompts with carousels” — which queries AI answers with a carousel; clicking opens the prompt page, which has a dedicated “Products” tab
  • The “Products” table under “Details” lists every product of the period with appearances, visibility, first-place share, average position, price range, stores and providers; next to it are the “Stores”, “Prompts” and “Carousels” tabs with full answers
  • A product page shows visibility over time, position distribution, “carousel neighbours” — products shown next to it, stores, and every answer it appeared in
Tip

Look at the carousel neighbours on your own product page: they are the direct competitors at the moment of choice. If they consistently rank higher, compare descriptions and prices on the store pages AI cites — that is where the engine takes the cards from.

11.4

“My products”: catalog and card linking

To get metrics for your own assortment, upload your catalog on the “My products” tab. The matcher finds your products among carousel cards on its own and asks you to confirm the ambiguous ones.

1/3
  • Three ways to mark a product as yours: upload a feed (“Upload feed”), add a product manually (“Add product”), or click the “My product” star on any card in the “Products” table
  • A feed is accepted by public URL or as a file: Yandex Market YML, Google Merchant (RSS/Atom) and CSV/TSV with a title column; the format is detected automatically, up to 25 MB and 20,000 products
  • The matcher links cards automatically when the SKU or model code and the brand match (confidence ≥ 90%); title-only matches go to “Matcher suggestions” with “It’s mine” / “No” buttons — rejected ones are never suggested again
  • The “Catalog” table shows each product’s linked cards, appearances, visibility, position and a comparison of the feed price with the price in AI
  • A URL feed can be re-synced with the refresh button: new products are added, products missing from the feed are removed along with their links
  • After every new monitoring run click “Refresh links” — the matcher goes through fresh cards and picks up new appearances of your products
Tip

Include the SKU or model code (vendorCode, mpn, sku) and the brand in your feed — linking is most accurate by those. Stores spell carousel titles differently, but the model code is the same everywhere.

11.5

“My products” mode: what it shows

The “All products / My products” switch in the section header changes the metric set: instead of the overall picture, only cards linked to your catalog.

  • “Share of my cards” — what part of all carousel cards in the period belong to your products; “Visibility of my products” — the share of carousels with at least one of your products
  • “Average position” and “First place” — where your products usually sit and how often they open the carousel; the chart, showcase, lists and tables also show only your assortment in this mode
  • If the catalog is uploaded but nothing is linked yet, the section shows a hint with an “Open catalog” button — metrics appear after the first link
  • The number next to “My products” is the catalog size; products that never appeared in carousels are marked “Not seen” in the Catalog table — candidates to work on
Tip

Keep both modes in view: “All products” shows who occupies the shelf in AI engines, “Mine” shows how much of it is yours. The gap between them is your growth potential in the e-com layer.

11.6

Filters, the prompt tab and the assistant

The section follows the global header filters — the period and the “prompt groups × providers” matrix — and carousel data is also available on the prompt page and in the built-in assistant.

1/3
  • The period is set in the header (“Last 30 days” by default); for ranges over two months the chart switches to weekly points
  • The “Filters” control in the header narrows the section to selected prompt groups and providers: handy for looking at carousels in a single product category
  • If the filter leaves only providers without product cards (e.g. Gemini or DeepSeek), the section shows a hint — add Alice AI or ChatGPT
  • Every prompt page has a “Products” tab: carousels for the last 180 days on that exact query, with positions and stores
  • The built-in assistant (“Agent” in the sidebar) and MCP tools answer questions about carousels: “which of my products get into carousels most often”, “which stores does AI suggest buying from”, “where do we lose on position”
Tip

Ask the assistant: “Compare where my products rank in Alice AI and ChatGPT carousels and suggest which prompts to start with.” It pulls the catalog, carousels and prompts itself and returns a prioritised list.

12 · 3 sections

Query fan-out

How to see what the AI actually searched for on your prompts — and how firmly your brand sits inside its own queries

12.1

What fan-out is and why it matters

An AI engine almost never searches your question verbatim. Before answering, it breaks the question into a series of its own search queries — sub-queries — and gathers sources from those. The "Query Fan-Out" section shows those sub-queries: the actual wordings the AI took to search while answering your prompts.

  • A single prompt expands into several sub-queries: the model translates the question, adds a year, a city, a category — and often brand names you never mentioned
  • This is the closest thing to classic SEO search queries, except the wording comes from the model, not the user — and it decides whose pages make it into the answer
  • Sub-queries are captured automatically on every monitoring run for the providers that expose their search: Yandex (Search with Alice), Alice AI, ChatGPT, Perplexity, Gemini, Claude, DeepSeek
  • The section obeys the global filters up top — period, providers, prompts: look at fan-out across the whole brand or narrow it to a single topic
  • A specific prompt's sub-queries also live in its own card, on a separate fan-out tab — handy when you're digging into why you weren't shown on that topic
Tip

When a sub-query repeats your prompt word for word, the model simply searched it as-is. The valuable rows are the ones where it added something of its own: a year, a region, "reviews", "comparison", or a competitor's name. Those are the queries that actually decide the answer.

12.2

Brands in sub-queries: how well the model knows you and your competitors

The "Brands in sub-queries" card counts how often the AI, unprompted, puts a brand name into its own search query. This isn't about being mentioned in the answer or cited as a source — it's about what the model pulls from its own memory when deciding who is even worth checking on your topic. In effect, a measurable LLMO signal: how firmly the brand is embedded in the model's knowledge.

  • On the left, "Frequent phrases": recurring word combinations from sub-queries, with the number showing how many of the prompts run in this period contain the phrase. It's the topic's vocabulary as the model sees it
  • On the right, you against your competitors. In the example the brand name landed in 36 of 3,190 sub-queries (1%), competitor names in 620 (19%). A 19x gap means the model treats other names as the anchors of this niche
  • "Top competitors in sub-queries" ranks who the AI substitutes most often (Alfa-Capital — 161, VTB — 93, VIM Investments and UK Pervaya — 54 each). These are the brands the model starts its search from on your topic
  • Watch your share relative to competitors and how it moves, not the absolute number. Zero sub-queries with your name while competitors are active is a direct signal that the model doesn't know you in this category
  • "Brands in sub-queries over time" shows the same data by day in two modes — "Share, %" and "Occurrences" — for you and the top 5 competitors on one chart: you can see whether you're gaining weight in the model's memory or losing it
  • There's a warning icon next to the cards for good reason: branded prompts inflate the metric (you named the brand yourself, so the model echoed it into search), and names that double as common words sometimes match ordinary text. The cleanest read comes from non-branded prompts
Tip

The goal isn't to appear in a sub-query once — it's to close the gap with the top competitors. The same thing works here as with human brand awareness: reviews, comparisons and roundups on third-party sites where your name sits next to the category leaders — that's exactly the text the model learns the niche from.

12.3

The sub-query table: what to do with it

Below the charts is a table of every sub-query for the period: the wording itself, the provider that generated it, and how many times it came up. It's a ready-made content backlog written by the AI engines rather than by you or a keyword tool.

  • "Runs" is how many times the wording appeared in the period. Sorting by that column lifts the durable topics to the top and pushes one-offs down
  • The provider icon shows whose model phrases it that way: the same topic sounds different across providers — one drills in with follow-up queries, another searches almost verbatim from your prompt
  • The table search is the fastest way to pull every wording containing "reviews", "comparison", "price", or a specific competitor's name
  • You can lift the wordings as they are into headings, H2s and FAQ blocks: these are literally the strings the model uses to look for sources
  • The filters up top apply here too: pick a single prompt to see only its fan-out; the same table is available inside the prompt's own card
Tip

A simple routine: sort by "Runs", walk the top 20–30 wordings, and honestly ask whether you have a page that answers exactly that. Every frequent wording you don't cover is a topic the model is already asking about — and finding the answer somewhere else.

13 · 3 sections

Weekly reports

Baseline and visibility dynamics reports: metrics across all answers, period comparisons, and brand and competitor profiles.

13.1

Branded clusters: separate report metrics

If questions directly concern your brand, mark their cluster as branded. Reports will show branded and other prompt results separately.

  • Enable Branded cluster in cluster settings. The label applies to all prompts in that cluster.
  • Reports calculate visibility, competitor comparisons and summaries for branded prompts separately from other prompts.
  • The same prompt wording can exist in different clusters, for example for different regions. Duplicate wording within one cluster remains restricted.
Tip

This label does not start fact checks or change the monitoring schedule. To check accuracy, select existing prompts in Accuracy.

13.2

Baseline report: your first visibility measurement

The baseline report shows how a brand appears in AI answers during the selected period. It brings together metrics, provider and competitor comparisons, presence quality, and qualitative profiles. Find it in the Reports tab or download the PDF.

PDF guide
MSI — visibility baseline report
A real GEO Scout report for MSI covering share of mentions, answer coverage, providers, topic clusters, and brand perception. This sample PDF is in Russian.
16 pages · 0.6 MB
  • The cover shows share of mentions, answer coverage, rank among participants, and the share of answers linking to the brand website.
  • Provider breakdowns show where the brand appears and how many substantive answers mention it. Missing data is labelled separately.
  • The topic table includes every cluster in the selected set, the brand’s metrics, and leaders by mention frequency.
  • Competitor comparisons include qualitative profiles describing how AI presents the brand and alternatives, including advantages and reservations.
  • Presence quality is separate from mention frequency: a mention, an alternative, an argued or primary recommendation, or benchmark status.
  • Numeric metrics cover the complete dataset. Qualitative profiles use diverse answer examples; the sample size is stated in the report.
Tip

Start with the baseline report to establish the initial position. Use a dynamics report with comparable periods to assess change.

13.3

Visibility dynamics: comparing two periods

The dynamics report compares an earlier period with the current one. It shows changes in share and coverage across providers, topics, and competitors, alongside a detailed view of the brand’s current position. The weekly report helps teams and clients discuss these changes.

PDF guide
MSI — AI visibility dynamics
A real comparison of two MSI monitoring periods. Both date ranges are stated in the PDF. This sample is in Russian.
17 pages · 0.7 MB
  • Before-and-after figures and percentage-point changes distinguish share growth from changes in the total number of answers.
  • When provider or prompt coverage differs, changes use common prompt–provider pairs. Full-current-period metrics are labelled separately.
  • Tables show changes across providers and topics, competitor positions, and brand presence quality.
  • Concise findings explain the main differences between periods. Website citations and brand mentions are not presented as visits or sales.
Tip

Compare similar periods and check coverage. This example demonstrates the report format; its periods are stated on the cover.

14 · 5 sections

Brand names, variations, and support

Brands go through rebrands, abbreviations, and alternative spellings — AI mentions them in any form. In GEO Scout you can mark such variations as your brand or merge them into a single competitor. Plus how to report errors in response processing and reach support.

14.1

A variation of your brand landed in competitors — mark it as yours

AI often uses alternative spellings of a brand: the old name after a rebrand, an abbreviation, a translation into another language. Some cases are trickier: the service "Яндекс.Касса" was renamed "ЮKassa" — same brand, and both forms should be merged. But the similarly-named "ЮMoney" (formerly "Яндекс.Деньги") is a separate, independent brand that spun off in 2020 — it shouldn't be merged in. Until the system knows "ЮKassa" and "Яндекс.Касса" are the same brand, the old name shows up as a new competitor. One click marks it as yours — the full mention history transfers to your main brand.

  • Go to Monitoring → Competitors and find the entry that's actually your brand
  • Open the competitor menu and pick "Mark as our brand"
  • The competitor is removed from the list, and its name is added as a variation of your brand
  • All accumulated mentions are automatically transferred to your brand — no data is lost
Tip

To prevent this from happening again, add all possible brand spellings in advance at Settings → Brand → Name variations.

14.2

Rebrands and competitor variations — merge them into one

Competitors run into the same issue, but real cases go beyond simple transliteration: the bank VTB may show up in AI answers as "ВТБ", "ВТБ24" (its former retail sub-brand, merged into VTB in 2018), and "VTB Bank" (the English form) — three mentions that are technically the same bank. After a rebrand, the old and new names often appear side by side in the same answer. GEO Scout lets you merge such variations into a single competitor — monitoring history is preserved, and analytics stop fragmenting on duplicates.

  • Go to Monitoring → Competitors and use the checkboxes to select competitors that are the same brand (at least 2)
  • Click the merge button — a dialog opens where you pick the primary name variation that will hold the data
  • All monitoring data (mentions, alerts, responses) is transferred to the selected primary variation, and the duplicates are removed
  • Subsequent monitoring runs respect the merge — new mentions of the duplicates are automatically attached to the primary competitor
Tip

Merging is irreversible — carefully verify the selected competitors really are the same brand before confirming.

14.3

Reporting a response processing error

If you spot a response that was processed incorrectly — for example, brand or competitor mentions were misidentified — report it from the interface.

  • Open a response card in Monitoring → Responses and click the response you want to review
  • Use the rating buttons: thumbs up (accurate), thumbs down (inaccurate), or flag (serious error, needs review)
  • When you flag a response, our team gets a notification and recalculates it manually
Tip

Your feedback improves the processing algorithms. Every flagged response is reviewed by the team and recalculated when needed.

14.4

In-app feedback

GEO Scout has a built-in feedback system — report a bug, suggest an improvement, or ask a question right from your account.

  • Click the feedback button in the sidebar to open the form
  • Pick a category: bug report, improvement suggestion, or other
  • Describe the issue or suggestion and optionally attach up to 3 screenshots (PNG, JPEG, WebP, up to 5 MB each)
  • Every submission is tracked — see the status (new, under review, in progress, done) and your conversation with the support team
14.5

Support contacts

Got questions or hit an issue? Reach out through any channel that works for you.

  • Telegram: @geoscout_support — message us on Telegram for a quick response
  • Email: support@geoscout.pro — for detailed questions and formal inquiries
  • The in-app feedback form — the fastest way to report an issue linked to your account
15 · 11 sections

Methodology

How GEO Scout collects data, detects mentions, calculates sentiment, position, competitive gaps, Share of Voice, and Citation Share — and how those data points become recommendations in the command center

15.1

Data collection: 12 AI providers, daily

GEO Scout sends your prompts to 12 AI providers every day and saves the full responses for analysis. For most engines we read the answer straight from the live product — the same interface a real user sees — rather than a developer API, because API answers routinely differ from the app. Each provider receives identical queries — that's what makes the results comparable.

  • 12 AI providers: ChatGPT (OpenAI), Claude (Anthropic), DeepSeek, Gemini (Google), Google AI Mode, Google AI Overview, Grok (xAI), Perplexity, Yandex Search with Alice, Alice AI, GigaChat, and Microsoft Copilot
  • For 11 of the 12 engines the answer is captured straight from the live product — the interface a real user sees: ChatGPT, Perplexity, Copilot, DeepSeek, GigaChat, Gemini, Google AI Mode, Google AI Overview, Grok, Alice AI, and Yandex Search with Alice. This matters because API answers often diverge from the app (model version, system prompt, web-search and citation behavior). Only Claude is collected through an official API
  • Each prompt is sent to each provider separately — responses aren't cached or reused
  • Monitoring runs daily on its own
  • The full text of every response is stored — read the original and verify the analysis whenever you want
  • Request geography is taken into account according to the AI provider’s capabilities. See “How request location is determined” for details.
  • If during data collection a model answers from memory or returns an empty response, we send a follow-up request asking it to look up current information — so the answer reflects fresh data rather than the model's stale knowledge
  • The follow-up is off by default and applies to Alice, ChatGPT, DeepSeek, Gemini, GigaChat, and Perplexity — other providers are unaffected. You can turn it on in Settings → Brand → "Data collection" → "Follow-up when an answer has no sources": we then send a follow-up whenever an answer comes from memory and keep the second answer if it cites sources
Tip

On lower plans, focus on niche prompts — queries that describe your service or product as precisely as possible. On higher plans, add broad prompts for a market overview — that's how you see the full competitive landscape in AI.

15.2

How request location is determined

Monitoring geography comes from your brand settings: country, region or city. If a prompt cluster has its own geography, those settings are used for its requests. This lets you monitor different markets within a single brand.

  • The monitoring region and the country of connection may differ. Depending on the AI provider’s capabilities, we pass the selected location through search settings, browser coordinates and the connection region. Each service determines how to use these signals in its answer.
  • For ChatGPT monitoring of Russia and Belarus, the connection is routed through the Netherlands. This reproduces the scenario of accessing the service through a foreign VPN. The brand or cluster geography is preserved in the monitoring settings.
  • For local queries, include the city directly in the prompt. For example, “Which dental clinics are recommended in Kazan?” specifies the region more precisely than “Which dental clinics are recommended nearby?”. In the latter case, the answer depends more heavily on the location determined by the AI service itself.
Tip

Monitoring results may differ from answers in your personal account: they are influenced by the country of connection, the geolocation available to the service, conversation history and personalization.

15.3

Brand mention detection

Once a response comes back from AI, the system analyzes the text and identifies which brands and competitors are mentioned. We use fuzzy matching so no mention is missed — even when AI misspells the name or uses a different transliteration.

  • Every brand-name variation is checked in every response: the primary name, transliterations, abbreviations, and alternative spellings
  • Brand domains are also tracked — if AI mentions your site with a link, it counts as a citation
  • Competitors are detected automatically: if AI mentions a company in your industry, it's added to the competitor list
  • Duplicate competitors can be merged, and if your brand lands among competitors — mark it as "our brand" with one click
  • You don't need to track spelling variations by hand — the system regularly checks every brand and competitor mention, finds different spellings, and merges duplicates in the background
Tip

Add every spelling of your brand at Settings → Brand. This is critical: if AI writes the brand differently — e.g. "T-Bank" instead of "Tinkoff", or "SberBusiness" instead of "Sber" for a separate product line — the mention will be missed without the variation.

15.4

Competitor discovery: how it works

GEO Scout discovers competitors automatically from AI responses. Every brand AI recommends as a solution to the user's task is added to your competitive landscape. We deliberately use a broad approach: the system captures everyone AI places alongside your brand — that's how AI search engines see competition.

  • The system extracts every mentioned brand from each AI response and classifies its role: solution, tool, source, infrastructure, example, etc.
  • Only brands with the "solution" role become competitors — the ones AI recommends as an answer to the user's query. Tools, sources, and infrastructure are filtered out automatically
  • Broad prompts = a broad competitive landscape. If your prompt is "best AI tool for marketing", AI may mention dozens of services — from CRMs to content generators. That isn't an error but a real picture of who you compete with for AI attention
  • With few prompts (lower plans), focus on narrow, niche queries: instead of "best CRM", try "CRM for a dental clinic with WhatsApp integration". On higher plans (20+ prompts), add both broad ones for a market overview and niche ones for precise competitive intelligence
  • Group prompts into clusters (business directions) — this structures the competitive landscape. Don't worry about being exact: the weekly report and recommendations correctly classify all competitors from a GEO perspective regardless
  • The system regularly runs a deduplication pipeline: it checks different spelling variations of competitors and your brand, automatically merging duplicates. Manual intervention usually isn't needed — cleanup runs in the background
  • Deactivate irrelevant competitors by hand and merge duplicates via aliases. Aliases also handle rebrands: if a competitor changed its name, add the old name as an alias — mention history stays intact. If your brand or one of its names (old brand after a rebrand, abbreviation, spelling in another language) lands among competitors, mark it as "our brand" with one click — all mentions transfer to your brand and the system excludes them from competitive analysis
Tip

Competitive landscape quality depends directly on prompt quality. Generic prompts give a broad market overview; niche ones surface precise competitors. Combine both types and group them into clusters — even rough grouping gives the system context for more accurate recommendations.

15.5

How position in the response is calculated

When AI lists several brands, order matters. Position 1 means the brand is mentioned first, signaling AI priority. GEO Scout determines the ordinal position of every brand in every response automatically.

  • Position is the ordinal number of the brand among all mentioned companies: 1 = first, 2 = second, and so on
  • Average position is the arithmetic mean of the brand's positions across all responses for the period. Closer to 1.0 is better
  • Position is tracked per provider — see where your brand is stronger (e.g. first in ChatGPT, third in Gemini)
  • Position dynamics show the trend: is your brand rising in AI priorities or falling
15.6

Sentiment analysis

Being mentioned isn't enough — what matters is how AI talks about your brand. GEO Scout analyzes the sentiment of every mention and classifies it as positive, neutral, or negative.

  • Positive: AI praises the brand, recommends it, highlights advantages. Example: "One of the best services in the category"
  • Neutral: AI mentions the brand without evaluation. Example: "Available options include Brand A, Brand B, Brand C"
  • Negative: AI criticizes, points out drawbacks, warns. Example: "This service has had issues with…"
  • Recommendation — a special kind of positive mention: AI directly advises choosing your brand. This is the most valuable type
Tip

A high share of positive mentions and recommendations is a sign of strong expert reputation. If sentiment is negative, check what's said about you in the sources AI uses (Wikipedia, reviews, forums).

15.7

Competitive gaps: where you're missing and competitors aren't

Competitive gaps are the key competitive-intelligence metric. These are prompts where AI mentions your competitors but not your brand. Every gap is a missed opportunity that can become a growth zone.

  • Gaps are detected automatically: if at least one competitor is mentioned in the response and your brand isn't, the prompt becomes a competitive gap
  • For each gap you see the specific competitors mentioned instead of you, and on which providers
  • Gap analysis feeds content recommendations: which articles to create so AI starts mentioning you on these queries
  • Gaps are tracked over time — see whether they closed after you published content or whether new ones appeared
Tip

Start with the most frequent and commercially important gaps. Creating content on those topics is the fastest way to increase Share of Voice.

15.8

Share of Voice: the main AI-visibility metric

Share of Voice (SoV) is the share of your brand's mentions among all brand mentions in AI responses. If competitors are mentioned 100 times in total and your brand 15, your SoV = 15%. It's the primary KPI for AI brand visibility.

  • Formula: SoV = your brand mentions ÷ total mentions of all brands × 100%. All responses for the selected period are counted
  • SoV is computed separately per provider, intent type, and prompt — you see the breakdown at any level
  • SoV trends show the change over time: is your share growing or are competitors capturing AI attention
  • A 1% SoV increase can mean dozens of additional brand recommendations in AI responses every day
15.9

Citation Share: your domain's share of AI trust

Citation Share is the share of citations of your brand's domain among all citations in AI responses (including competitors and any third-party sources). Where Share of Voice shows how often AI mentions your brand, Citation Share shows how often AI uses your site as the source — compared to every other domain AI links to.

  • Formula: Citation Share = brand domain citations ÷ all domain citations in AI responses × 100%. The denominator counts every cited domain (competitors, reference sites, third-party articles — any domain AI links to), not just tracked competitors
  • A growing Citation Share means AI increasingly relies on your site as a factual source, not just mentions the brand by name
  • Read Citation Share alongside Share of Voice: many mentions but a low citation share means the brand is visible but your content isn't yet an authoritative source for AI
  • Citation Share answers "are you beating competitors on domain authority" directly. If your share is higher than direct competitors', AI treats your site as a more authoritative niche source than theirs. If lower, competitors enjoy more AI trust on their domains — and the ranking shows exactly whom to take share from
Tip

Read Share of Voice and Citation Share together. SoV high but Citation Share low — you're mentioned but AI links to competitor sites: invest in domain authority (pillar pages, schema, external mentions of your own site). Citation Share growing faster than SoV — the brand isn't on everyone's lips yet, but your site is already winning the battle for AI trust.

15.10

How recommendations appear in the command center

Recommendations in the command center aren't generic advice — they're a consequence of observations on your prompts. Signals are extracted from monitoring data, turned into actions across two surfaces — site changes and external publications — given a priority, and verified by the next monitoring cycle and your marketing campaigns.

  • Signals from monitoring. After every cycle of AI responses, the system records concrete observations: "a page showed up in answers but wasn't cited", "the page has no FAQPage markup", "an AI bot is blocked in robots.txt", "competitors have a vc.ru publication on topic X, the brand doesn't". Each signal is tied to real prompts, providers, and competitors with an evidence level: a specific domain, only the source class, or a candidate from the catalog of analogs
  • Materialization — site changes. Signals become pinpoint actions: improve an uncited page (the AI agent builds answer blocks for specific queries), generate missing Schema markup, create an llms.txt, open access to AI bots. The goal is to turn an appearance into a citation where the engines already show interest
  • Materialization — external publications. Here actions are tied to outlets AI already cites (media, communities, catalogs, reviews): land a mention in the right topic and, from a content plan, generate an article in the format that already works, with verified facts from the knowledge base. One outlet — one action card
  • Publication plan. External publications roll up into a separate plan tab that's built from coverage gaps. The system looks at 'query × provider' pairs where the brand doesn't yet appear or isn't cited and proposes briefs for them — ideas for future pieces with a topic, a format, and a rough article outline. Each brief is its own topic, not a rewrite of an old text. Aggregator catalogs don't make it into the plan: they're handled by separate actions, while the plan focuses on articles and mentions that actually move visibility
  • Prioritization. The system orders the feed on its own so the most urgent and impactful steps sit on top. Priority reflects how often — and on how important the prompts — a signal surfaces, how well-evidenced it is (a specific domain, a source class, or a candidate from the catalog), and the expected visibility lift. You don't need to read any scores manually — just work the feed top to bottom
  • Validation and refresh. An action stays in the feed while the signal is relevant; after "Done", the next monitoring cycle checks the effect, and a publication's URL can be added to a marketing campaign to measure citations directly. If AI starts citing the page or mentioning the brand on a previously-losing query, the signal is confirmed; stale signals move to archive. For a page optimization, the recommendation returns only if the page turns up uncited again for a different prompt
Tip

The priority of the same recommendation can shift from cycle to cycle: a competitor publishes a new piece — priority rises; the gap narrows — priority drops. Don't try to close everything at once: today's top-of-feed action may slip tomorrow, and an irrelevant signal may fall off on its own.

15.11

Transparency

GEO Scout is built on full transparency. Every metric can be verified, every AI response can be read in the original. We don't hide algorithms or manipulate data.

  • The full text of every AI response is available in Monitoring → Responses. Read the original and verify the analysis
  • Every metric is computed with open formulas described in this section. No black boxes
  • Data history is preserved — see how metrics changed over time and where each value came from
  • Reports are generated from the same data you see in the dashboard. Agencies can't show you a prettier picture — only the real numbers
Tip

If you work with an agency, give them access to GEO Scout — so both sides work with the same objective data.

16 · 5 sections

Referral program

How to earn 15% on every GEO Scout payment from clients you refer

16.1

What the referral program is

The GEO Scout referral program lets you earn commission by inviting new users. You get 15% of every payment from a referred customer — for as long as they stay subscribed.

  • 15% commission on every payment from a referred customer — lifetime, as long as the customer is subscribed
  • Free to join — any registered GEO Scout user can participate
  • Referral link works for 90 days — if a customer follows your link and signs up within 90 days, they're attributed to you
  • Transparent stats — track sign-ups, payments, and earnings in real time from your account
16.2

How to join

Joining takes a couple of clicks right from your account.

  • Go to Billing → Referral program in the sidebar
  • Click "Join referral program" — a unique referral code is generated automatically
  • Copy your personal referral link, geoscout.pro/?ref=YOURCODE
  • Share the link with colleagues, customers, or followers — on social media, in messengers, by email, or on your site
Tip

Your referral code is generated from your account name plus a unique suffix. The code can't be changed, but it's easy to remember.

16.3

How commission is credited

Commission is credited automatically every time a customer you referred pays for a GEO Scout subscription.

  • When a customer follows your link, the referral code is stored in a cookie for 90 days
  • After sign-up and email confirmation, the customer is permanently attached to your partner account
  • Every subscription payment from a referred customer credits you 15% of the amount
  • Commission is credited indefinitely — as long as the customer keeps paying, you keep earning
Tip

Commission applies to every payment, including renewals and plan upgrades.

16.4

Payouts and payment details

Once your balance reaches the minimum threshold, you can request a payout. Fill in payment details first.

  • Minimum payout is 3,000 ₽. Once your balance hits that amount, the "Request payout" button becomes available
  • Fill in payment details under "Payout details": legal entity type (self-employed, sole proprietor, LLC, or individual), full name, tax ID, and a Telegram for contact
  • Available payout methods: bank card transfer (for self-employed, sole proprietors, and individuals) or bank wire by full details (required for LLCs)
  • After a payout request, funds move to processing. We'll reach out on Telegram to confirm and complete the transfer
Tip

Verify your payment details before requesting a payout — they're captured at the moment the request is created.

16.5

FAQ

Answers to the most common questions about the GEO Scout referral program.

  • Can I refer myself? — No, the system automatically blocks self-referrals
  • What if the customer is already registered? — Referral codes only attach to new users. Each user can be referred by one partner only
  • How many referrals can I bring in? — No limit. The more customers you refer, the more you earn
  • How long does the referral attribution last? — Permanently. The customer stays attached to you from the moment of sign-up, and you earn commission on every payment they make
  • Do I owe taxes on referral income? — Yes, referral income is taxable. We recommend registering as self-employed (NPD) for a 6% rate