Track Traffic from ChatGPT, Grok, Perplexity, Claude and Gemini

Questions like "how do I track Grok traffic" or "how do I check traffic from Claude" have the same answer for every analytics tool: you look for the referrer the AI product passes when someone clicks a cited link, and you accept that part of the traffic never carries one. GA4 and Yandex Metrica both classify these visits by referrer and UTM parameters, and neither has a dedicated AI source.

This guide covers the referrer domains of 22 AI platforms, platform-specific quirks, what must not be counted as AI traffic, and a step-by-step setup. The walkthrough uses Yandex Metrica, the default analytics tool for sites in Russia and the CIS; the logic is the same in GA4, and the GA4 specifics are in our separate guide on how to track AI traffic in GA4.

How an AI click reaches your analytics

Yandex Metrica, according to its documentation, determines the source from a tag (such as UTM) in the first pageview of a session and, if there is no tag, from the HTTP referrer. GA4 uses the same two signals. There is no AI category in the list, so one "AI visit" can end up in different buckets:

So AI traffic is not a report you open. It is a segment you build from referrers.

What utm_source=chatgpt.com does

OpenAI's publisher FAQ says ChatGPT automatically includes utm_source=chatgpt.com in referral URLs. Two practical consequences:

• The tag survives when the referrer does not, so a UTM condition catches part of the ChatGPT visits that would otherwise sit in Direct.
• The tag has no utm_medium. In Yandex Metrica, a visit with an unknown utm_source is only classified as advertising when utm_medium is one of cpc, ppc, paidsearch, cpv, cpa, cpp, display, cpm or banner, per the Ad systems report docs. ChatGPT visits therefore stay in referral or direct, and the tag remains visible in the UTM tags report. In GA4 the missing medium can push these sessions into Unassigned; the GA4 guide shows how to fix that with a custom channel group.

Metrica reads tags from the first pageview only, and tag values are case-sensitive. If your site redirects to a clean URL before the counter loads, the tag is lost. Other AI products do not add tags consistently, so for them the referrer is the only signal.

Platform by platform

ChatGPT. Referrers chatgpt.com, chat.openai.com, openai.com, plus the UTM tag. It is the largest identifiable AI source in our data. The app is the main reason ChatGPT traffic is undercounted.

Grok. Referrers grok.com and x.ai. Grok is also built into X, and clicks from there cannot be told apart from normal X traffic by referrer. Across 13 sites and 12 months we have not recorded a single identifiable Grok referral, so a "Grok traffic" report will usually be empty even when Grok cites you. To see whether Grok mentions and cites your brand at all, you need answer monitoring; see how to track brand visibility in Grok.

Perplexity. Referrer perplexity.ai. In our sample its share of AI referrals fell from 13.7% over 12 months to about 3% over the last 90 days.

Claude. Referrer claude.ai. Small but steady in our data: 1.3–1.7% of AI referrals.

Gemini. Referrers gemini.google.com and aistudio.google.com. Do not confuse Gemini with AI Overviews and AI Mode: those live on google.com and send organic-looking clicks.

Microsoft Copilot. Referrers copilot.microsoft.com, edgeservices.bing.com (Copilot in the Edge sidebar), copilot.cloud.microsoft and m365.cloud.microsoft (Microsoft 365 Copilot). Copilot answers inside Bing results arrive as Bing organic.

Alice AI. Yandex's assistant. Only the chat at alice.yandex.ru and alice.ya.ru is identifiable. Yandex reported in its Q2 2026 results that Alice AI answers 42% of queries in Yandex Search, but clicks from those answers carry the same yandex.ru/search referrer as ordinary organic results.

DeepSeek, GigaChat, Qwen, Kimi. Referrers chat.deepseek.com, giga.chat, chat.qwen.ai, kimi.com and their variants below. None of them produced an identifiable visit in our sample over 12 months.

AI referrer domains

This is the catalog the AI Traffic section in GEO Scout uses to recognise AI referrals. The domains are taken from the product code.

Step-by-step setup in Yandex Metrica

The same logic works in GA4 with a custom channel group and an Exploration segment; see the GA4 guide for the exact screens.

1. Check what already arrives. Open Reports → Sources → Sites for the last 12 months. The first level shows the referring domain. Mark the domains from the table above. In our sample a site receives referrals from one to six AI platforms a year, with a median of four.
2. Make sure the site runs on HTTPS. Metrica's documentation states that a visit from an HTTPS page to an HTTP site has no referrer. Every AI platform is on HTTPS, so an HTTP site sees all AI traffic as direct.
3. Build the referrer segment. In the Sites report, hover over the chatgpt.com row and choose Segment by row. Open the condition and add the other AI domains: values inside one condition are combined with OR. Alternatively, switch to manual input and start the value with ~ to use a regular expression (RE2 syntax, applied to the full URL, case-sensitive):

^https?://([^/]+\.)?(chatgpt\.com|openai\.com|alice\.yandex\.ru|alice\.ya\.ru|perplexity\.ai|claude\.ai|gemini\.google\.com|aistudio\.google\.com|copilot\.microsoft\.com|edgeservices\.bing\.com|copilot\.cloud\.microsoft|m365\.cloud\.microsoft|deepseek\.com|giga\.chat|gigachat\.ru|gigachat\.devices\.sberbank\.ru|grok\.com|x\.ai|qwen\.ai|tongyi\.aliyun\.com|kimi\.com|kimi\.moonshot\.cn|mistral\.ai|meta\.ai|doubao\.com|duck\.ai|you\.com|genspark\.ai|felo\.ai|monica\.im|poe\.com|phind\.com)(/|$)

4. Add the UTM condition. Add a UTM Source condition equal to chatgpt.com and set the segment to match when any condition is met. If you tag your own links in articles and directories, keep those tags in a separate segment: that is traffic through your content, not a direct citation.
5. Save the segment with Save as, then apply it to Reports → Sources → Sources, summary to see how much of the identified AI traffic came as link traffic and how much as direct traffic with the ChatGPT tag.
6. Look at landing pages. Reports → Content → Landing pages with the AI segment shows which pages AI assistants actually send people to. Usually it is articles, comparisons, pricing and FAQ pages rather than the home page.
7. Add goals and revenue. Add your counter goals and, if e-commerce is enabled, revenue. AI volumes are small, so compare channels on conversion rate and revenue per visit, not raw sessions. More on audience quality in AI traffic retention vs organic.
8. Keep direct traffic next to it. Growth in direct visits without growth in ads, email or offline activity is an indirect sign of AI visits that lost their referrer. It is not proof, but it is the only signal you get for app traffic.

Why the number is always a floor

A referrer segment only counts clicks where both the AI product and the browser passed the source. Four flows bypass it:

1. Apps. The ChatGPT app and the Alice AI chat often open links without a referrer. The ChatGPT UTM tag rescues part of them; other apps have no tag.
2. AI answers inside search. AI Overviews, AI Mode, Copilot in Bing and Alice AI in Yandex Search send organic-looking clicks.
3. Delayed visits. Someone gets a recommendation, closes the chat and searches for the brand name the next day. Analytics records branded search.
4. Zero-click influence. The assistant names the brand, the user calls or buys elsewhere. No web analytics trace at all; this is the AI dark funnel.

On the same sites and dates as our data below, direct visits outnumbered identified AI referrals by roughly 130 to 1. Nobody can tell how much of that direct traffic actually came from AI: not Metrica, not GA4, not GEO Scout.

What our data shows: 13 sites

• ChatGPT and the Alice AI chat deliver 82 to 94% of identifiable AI referrals, depending on the period.
• Grok, DeepSeek, GigaChat, Qwen and the rest of the catalog produced no identifiable visits in 12 months on any of the 13 sites. That does not prove nobody clicks; the clicks may simply carry no referrer.
• Volumes are small: about 80 identified AI visits per site per year. Judge the channel by goals and revenue.

We collect all Metrica channels for these sites from September 15, 2026, which lets us compute the AI share for September 15–26. In those 12 days the 13 sites had 287,000 visits, 85 of them identified AI referrals:

• 0.03% of all visits and 0.19% of non-paid visits;
• per-site median 0.12%, range 0 to 2.4%;
• about 250 times fewer than search engine visits and about 130 times fewer than direct visits.

If AI traffic looks like a rounding error in your analytics, that is normal for the identifiable part. The useful question is not the size of the number but the link between what AI assistants say about your brand, which pages they cite, and who eventually arrives.

Automating it with GEO Scout

The AI Traffic section of GEO Scout was rebuilt on Yandex Metrica data on September 17–18, 2026.

• Connection. On the Setup tab, connect Yandex Metrica via OAuth with read-only access and pick a counter. No script on the site. 12 months of history are loaded on connection, then data syncs daily; a Refresh button triggers a manual sync.
• Recognition. The 22-platform catalog above, with the same exclusions for google.com, bing.com and yandex.ru. New AI-looking referrer domains are detected weekly and reviewed before they are added.
• Summary and conversions. AI visits, new visitors, AI share of all site visits, top source and growth; reaches of any counter goal, conversion rate as calculated by Metrica, e-commerce revenue from AI when e-commerce is enabled, bounce rate and time on site.
• Landing pages with citations. Which pages receive AI visitors and whether the same pages are cited in AI answers to your tracked prompts. Filters show pages with traffic but no citations, and pages that are cited but get no clicks. A page card shows daily visits with publication markers, sources, citations by AI provider and the prompts that cite it.
• Visibility → traffic. Answer reach on each monitored platform next to actual visits from its domain. Google AI Overviews, AI Mode and Alice AI in Yandex Search are marked as not trackable.
• Audience and blind spots. Devices, countries, new vs returning, plus the direct-traffic volume and trend for the period.

The same data is available to the Agent mode and through the GEO Scout MCP server.

Limits: the section works with Yandex Metrica only; GA4 is not supported, so for GA4 use the manual setup. Visits without a referrer are not guessed; their volume is shown next to AI traffic instead of being attributed to it.

Pricing: the AI Traffic section is available on every plan, including the free one. The visibility side depends on monitoring: the free plan covers up to 5 prompts in 6 AI assistants with a refresh every 7 days, and paid plans with daily monitoring start from $24/mo (pricing).

Checklist

• The site runs on HTTPS.
• All AI referrer domains that actually send traffic are found in the Sites report.
• A saved "AI referrals" segment: catalog referrer OR utm_source=chatgpt.com.
• google.com, bing.com, yandex.ru, ya.ru and duckduckgo.com are not in the segment.
• The counter fires on the tagged URL before any redirect.
• The segment is applied to landing pages, goals and revenue.
• Direct traffic is tracked alongside.
• New AI domains are checked monthly.
• Referral data is read together with answer monitoring and cited sources; to estimate the business case, see how to calculate GEO ROI.