# How to Track AI Traffic in GA4: ChatGPT, Perplexity, Gemini, Claude https://www.winstondigitalmarketing.com/playbooks/how-to-measure-ai-search-traffic-ga4/ // playbook / geo # How to track and measure AI search traffic in GA4. ChatGPT, Claude, Perplexity and Gemini are sending you visitors right now. GA4 labels some of them AI Assistant, files others under Referral, and loses the rest in Direct. Here is how to see all of it: the hostnames to check, the custom channel that catches what Google's misses, why the count will always run lower than the real influence, and how to tie AI citations to the conversions they produce. // reading time 15 minutes · // last updated October 16, 2026 · // author John Morabito GA4 has had a native AI Assistant channel since 13 May 2026, rolled out broadly by 7 June, so most AI sessions no longer sit in Referral by default. It undercounts, and Google has not published the full list of referrers it recognises. Google's channel definitions name ChatGPT, Gemini, DeepSeek, Copilot and Grok, and the launch announcement also named Claude. Perplexity appears in neither, so check where it lands in your own property before you trust the total. Google's own AI Overviews and AI Mode are excluded from the channel entirely: those clicks land in Organic Search when they carry a google.com referrer and in Direct when they do not. Every other no-referrer session lands in Direct too. Read the built-in channel as a floor, add a custom channel for whatever it misses in your data, and pair all of it with citation-share tracking for the visibility that never produces a click. ## Why GA4 hides AI search traffic by default. GA4's default channel grouping was built for a web where traffic came from search engines, social platforms, email, and direct visits. The AI assistants do not fit any of those buckets cleanly. Until May 2026, a click from a ChatGPT answer arrived as a referral from chatgpt.com, so GA4 quietly dropped it into the generic Referral channel next to forum links and newsletter footers. A click from a Gemini answer could land in Organic Search because the host is a google.com subdomain. The traffic was there, smeared across two channels with no label that said "an AI sent this person." Any assistant GA4 still does not recognise behaves the same way today. Google shipped that native channel on 13 May 2026 and it reached most properties by 7 June. It is on by default, it needs no configuration, and it tags qualifying sessions with a medium of `ai-assistant`. It is also incomplete, which is why a custom channel group is still worth building next to it rather than instead of it. The gap that cannot be closed by any channel rule is Google's own AI Overviews, because those clicks arrive either as ordinary organic or with no referrer at all, and nothing distinguishes them from a normal Google click. ## The AI referral hostnames to match. This is the core set as of mid-2026. Keep it in one place and review it quarterly, because the assistants rename and add domains constantly. | Engine | Referral host(s) | Where it lands in GA4 | |---|---|---| | ChatGPT | chatgpt.com, chat.openai.com | AI Assistant (named in Google's channel definitions) | | Gemini | gemini.google.com | AI Assistant (named in Google's channel definitions) | | Microsoft Copilot | copilot.microsoft.com | AI Assistant (named in Google's channel definitions) | | Claude | claude.ai | AI Assistant per Google's May 2026 announcement; confirm in your property | | Perplexity | perplexity.ai | Not named by Google; check source / medium, catch it with a custom channel | | Google AI Overviews and AI Mode | (none distinct) | Organic Search or Direct; excluded from the AI channel | AI Overviews are the one you cannot solve in GA4. A click from inside an AI Overview arrives under google with the ordinary organic medium and carries no hostname that separates it from a normal blue-link click. GA4 alone cannot split AI Overview traffic out, which is one more reason the referral count runs low. ## Build the custom channel group. In GA4, this lives under Admin, then Data display, then Channel groups. Create a new group (cloning the default is the safe way to keep everything else intact) and add one channel above the existing rules so it claims the traffic first. - **Name the channel** AI (all assistants). - **Set two conditions joined by OR.** Medium exactly matches ai-assistant, which picks up everything Google already classifies. Source matches regex chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai , which catches any of those hosts the native list misses in your property. - **Order it first,** above Organic Search and Referral, so a Gemini or Perplexity session gets claimed as AI before another rule can grab it. - **Save, then wait.** Channel groups apply going forward and to most historical reports, but give it 24 to 48 hours before you trust the comparison. If you would rather not touch the channel group, the lighter version is an Exploration with a session-source filter for the same hosts, or a Looker Studio page that does the regex match. The channel group is cleaner because every standard report inherits it. **// the regex caveat** Escape the dots in your pattern. chatgpt.com without the backslash matches "chatgptXcom" too, which is harmless here but a bad habit that bites you on broader rules. And test the pattern against a few weeks of Referral data before you rely on it, because a host you typed wrong fails silently and you will think the AI traffic is zero. ## How to check Perplexity, Gemini and Claude traffic in GA4. Open Reports, then Acquisition, then Traffic acquisition, and switch the primary dimension to Session source / medium. Search the table for perplexity and claude. If a row reads perplexity.ai / referral, the native channel is not catching it in your property and the custom channel above is doing real work. If claude.ai shows the ai-assistant medium, Google is classifying it and the native channel covers that one. Gemini needs its own look. Its host, gemini.google.com, is a Google subdomain, and before May 2026 those sessions could be filed under Organic Search. Google names Gemini in its AI Assistant definitions, so in a healthy property the row reads gemini.google.com / ai-assistant. If it shows up as organic or referral instead, the custom channel above claims it. Copilot works the same way under copilot.microsoft.com. Perplexity is the one most teams ask about, because it shows its sources in every answer and readers click them. To track it over time, save the source / medium view as an Exploration filtered to perplexity, with Landing page and Key events added. That gives you a running Perplexity report without waiting for Google to add it to the native list. ChatGPT has a wrinkle too. Links in ChatGPT search answers often carry `utm_source=chatgpt.com`. The tag sets the source, so those sessions still match the regex. If someone on your team filters by referrer host in another tool, have them search for the UTM as well, or the two reports will disagree. Run the same check for any new assistant you hear about. Your own source / medium report is the cheapest audit of a list Google has not published. Add Landing page as a secondary dimension and you also see which pages each assistant sends people to, which is the input for the next step. ## Can you track AI Overview traffic in GA4? Not directly. A click from an AI Overview or an AI Mode answer arrives as google / organic, the same as a blue-link click, and GA4 has no field that tells them apart. Search Console does not split it out either. Google counts AI Overview and AI Mode impressions and clicks inside the regular Web search type in the Performance report. What you can do is read the pattern. In Search Console, look for queries where impressions hold or rise while click-through rate drops at a steady average position. That shape usually means an answer is sitting above your link. Check those queries by hand, or in a citation tracker that records AI Overview presence, to see whether your page is one of the cited sources. Put the two side by side and you have a defensible read on AI Overview exposure, even though GA4 will never hand you a clean session count for it. ## Picking an AI visibility tool that connects to GA4. A lot of teams go looking for a tracker that puts GA4 and Search Console data next to citation data, so nobody has to stitch spreadsheets every month. Plenty of tools now claim to do this. Before you buy one, ask four questions. - Does it connect to each GA4 property through the API, or does it expect a CSV upload every month? - Does it join on landing page URL? That is the only honest join, because no assistant passes the prompt. - Can it handle several properties in one view? An agency or a company with more than one brand needs that from day one. - Does it pull Search Console too, so AI Overview exposure sits beside the AI Assistant sessions? If nothing on the shortlist fits the budget, a Looker Studio page that blends GA4 with an exported citation sheet does the same job. It takes an afternoon to set up and your team owns it. Our comparison of [the best tools to track AI citations](https://www.winstondigitalmarketing.com/playbooks/best-ai-citation-tracking-tools/) covers the paid options. ## Tie AI visibility to GA4 conversions. Most teams end up asking a harder question than how many AI sessions arrived: is being cited producing sign-ups and revenue? GA4 cannot answer that alone, because no assistant passes the prompt that produced the click. What you can do is join on the landing page. Your citation tracker records which of your URLs each engine cites for each prompt. GA4 records which landing pages AI sessions arrive on and what they convert into. Join the two on URL and week and you get a read you can defend in a budget meeting: the pages that win citations, and the conversions those same pages produce from AI sessions. It is a correlation. Anyone offering prompt-level attribution from GA4 data should be asked where the prompt comes from. For an agency reporting across several clients, or an in-house team reporting to finance, the page-level join is the honest version. It fits on one Looker Studio page fed by GA4 and an export from the tracker. ## Why the number will always undercount. GA4 counts a session only when someone clicks through to your site. The entire purpose of an AI answer is to resolve the question in place, with no click. The engine reads your page, lifts your sentence, names your brand in the answer, and the user leaves satisfied without ever visiting. GA4 records nothing, because nothing was visited. So your AI Referral channel measures the slice of AI influence that converted into a click, while the larger slice (the citations and brand mentions that shaped the answer itself) stays invisible to any clicks-only tool. Watch the referral count as a trend line for the AI traffic that landed. It will not tell you how visible you are in AI search. And even the clicks that do happen often land in Direct rather than your AI channel, because most assistants pass no referrer, one of the reasons [your organic decline is partly a measurement problem](/playbooks/organic-traffic-isnt-dying-your-measurement-is/). Measuring the value that survives when the click does not is the whole subject of our [zero-click search strategy for 2026](/playbooks/zero-click-search-strategy/), and the fuller case for why clicks-based analytics is structurally blind to AI, and what to measure instead, is in [why you can't see AI search in your analytics](/playbooks/why-you-cant-see-ai-traffic-in-analytics/). Paid media has the same offline blind spot, and one clean fix for attributing walk-in and cross-device sales to a campaign is a coded offer, covered in [using promo codes to fix paid media attribution](/playbooks/promo-codes-paid-media-attribution/). ## What to pair it with. Because GA4 only sees clicks, the referral channel is half a dashboard. The other half is [citation share](/playbooks/citation-share-replaces-rankings/), the metric that measures how often the engines actually cite your domain for a fixed set of prompts, click or no click. The two read together: GA4 tells you the AI traffic that landed, citation-share tracking tells you the AI visibility that produced it. If the reason you are here is that AI Overviews cut your clicks while impressions held steady, the full diagnosis-and-recovery path is in [recovering traffic after AI Overviews cut your clicks](/playbooks/ai-overviews-traffic-recovery/). To stand the second half up, you run a fixed prompt set against each engine on a schedule and record who gets named. That instrumentation is the whole job in [how to build an AI visibility dashboard](/playbooks/how-to-build-an-ai-visibility-dashboard/), and if you would rather buy than build, the honest tradeoffs between the off-the-shelf options are in [the best tools to track AI citations](/playbooks/best-ai-citation-tracking-tools/). Either way, the GA4 channel and the citation tracker are the two halves of a real generative-engine measurement setup, which is the foundation under everything in our [generative engine optimization service](/services/generative-engine-optimization/). ## The 20-minute setup. - **Clone the default channel group** in GA4 Admin so you keep a safe fallback. - **Add the AI channel** with the medium rule and source regex above, ordered first. - **Validate** in Traffic acquisition by source / medium that Claude, Gemini and Perplexity land where you expect. - **Build one report or Exploration** that trends AI sessions, engaged sessions, and conversions by landing page over time. - **Stand up citation-share tracking** so you see the influence the clicks miss, then join it to GA4 by landing page. - **Revisit the host list quarterly,** because the engines keep changing domains. ## Frequently asked questions. ### How do you measure AI search traffic in GA4? GA4 has had a native AI Assistant channel since 13 May 2026, rolled out broadly by 7 June, so most AI sessions no longer sit in Referral by default. It undercounts, and Google has not published the full list of referrers it recognises. Google's channel definitions name ChatGPT, Gemini, DeepSeek, Copilot and Grok, and the launch announcement also named Claude. Perplexity appears in neither, so check where it lands in your own property before you trust the total. Google's own AI Overviews and AI Mode are excluded from the channel entirely: those clicks land in Organic Search when they carry a google.com referrer and in Direct when they do not. Every other no-referrer session lands in Direct too. Read the built-in channel as a floor, add a custom channel for whatever it misses in your data, and pair all of it with citation-share tracking for the visibility that never produces a click. ### What are the AI search referral hostnames to track in GA4? The core set as of mid-2026 is chatgpt.com and chat.openai.com for ChatGPT, perplexity.ai for Perplexity, gemini.google.com for Gemini, copilot.microsoft.com for Microsoft Copilot, and claude.ai for Claude. GA4's native AI Assistant channel already claims the assistants Google names, so the hosts worth watching by hand are the ones it may not, Perplexity first. AI Overviews are the hard case: those clicks arrive under google with the normal Organic Search medium and carry no hostname that separates them from a blue-link click, so GA4 alone cannot isolate AI Overview traffic. Keep the host list in one place and review it quarterly, because the assistants add and rename domains often. ### Why does GA4 undercount AI search traffic? GA4 only counts a session when someone clicks through to your site, and the entire point of an AI answer is to resolve the question in place without a click. The engine can read your page, lift your sentence, name your brand, and send the user away satisfied, and GA4 records nothing because no visit happened. So the referral number is a floor. It tells you the AI traffic that converted into a click, while the larger influence (the citations and mentions that shaped the answer) stays invisible to a clicks-only tool. ### How do I check Claude or Perplexity traffic in GA4? Go to Reports, then Acquisition, then Traffic acquisition, set the dimension to Session source / medium, and search for claude and perplexity. Google's May 2026 launch announcement named Claude as a recognised assistant, so claude.ai sessions should carry the ai-assistant medium. Perplexity is not named in Google's channel definitions, so if it shows as perplexity.ai / referral, add it to a custom channel. Sessions where the assistant passed no referrer land in Direct and cannot be recovered. ### Can you tie AI citations to GA4 conversions? Partly. No assistant passes the prompt to GA4, so you cannot attribute a session to the prompt that produced it. You can join citation data to GA4 on landing page: the tracker shows which of your URLs each engine cites, and GA4 shows which landing pages AI sessions reach and what they convert into. Read by URL and week, that tells you whether the pages winning citations are also producing sign-ups. It is a correlation, and it is the most defensible version available. ### Can you track AI Overview traffic in GA4? No. Clicks from AI Overviews and AI Mode arrive as google / organic, identical to a normal Google click, and Search Console reports AI Overview impressions and clicks inside the regular Web search type. The workable proxy is pattern-reading: find queries where Search Console impressions hold while click-through rate falls at a steady position, then check them against a citation tracker that records whether your page is cited in the AI Overview. ### How do I check Gemini traffic in GA4? Open Reports, then Acquisition, then Traffic acquisition, set the dimension to Session source / medium and search for gemini. Gemini is named in Google's AI Assistant channel definitions, so sessions from gemini.google.com should show the ai-assistant medium. If they show as organic or referral, add gemini\.google\.com to the source regex in a custom AI channel ordered above Organic Search. ## Reporting AI search to leadership? We set this up for in-house marketing teams: the GA4 channel for the AI clicks that land, citation-share tracking for the visibility they miss, and one report that joins the two by page. Your team keeps running it after we leave. Twenty minutes on a call is enough to see what your property is missing. [Book a 20-minute call →](/contact/#book-a-call) [See the GEO service →](/services/generative-engine-optimization/)