Documentation
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
Contents
- 01 Data collection: 12 AI providers, daily
- 02 Brand mention detection
- 03 Competitor discovery: how it works
- 04 How position in the response is calculated
- 05 Sentiment analysis
- 06 Competitive gaps: where you're missing and competitors aren't
- 07 Share of Voice: the main AI-visibility metric
- 08 Citation Share: your domain's share of AI trust
- 09 How recommendations appear in the command center
- 10 Transparency
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 8 of the 12 engines the answer is captured straight from the live product — the interface a real user sees: ChatGPT, Perplexity, Copilot, DeepSeek, GigaChat, Google AI Mode, Google AI Overview, and Alice AI. This matters because API answers often diverge from the app (model version, system prompt, web-search and citation behavior). Claude, Gemini, Grok, and Yandex Search with Alice are collected through their official APIs
- Each prompt is sent to each provider separately — responses aren't cached or reused
- Monitoring runs daily on its own. You can also trigger a manual run at any time
- The full text of every response is stored — read the original and verify the analysis whenever you want
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.
2.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
Add every spelling of your brand at Settings → Brand. This is critical: if AI writes the brand differently (e.g. "T-Bank" vs "Tinkoff"), the mention will be missed without the variation.
3.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
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.
4.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
5.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
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).
6.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
Start with the most frequent and commercially important gaps. Creating content on those topics is the fastest way to increase Share of Voice.
9.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 produces answer capsules 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. An outlet-centric approach: one outlet = one card
- Prioritization. Each action gets a priority_score based on predicted impact, effort, evidence level, and the frequency/importance of the affected prompts. From there come the "Quick win" badge (low effort, visible impact) and "High impact" (large visibility lift), which you can use to filter the feed
- 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
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 high_impact may be tomorrow's quick_win, and an irrelevant signal may fall off on its own.
10.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
If you work with an agency, give them access to GEO Scout — so both sides work with the same objective data.