How to Measure AI Referral Traffic in GA4 and Search Console
Where AI referral traffic shows up in GA4 and Search Console: the AI Assistant channel, AI Overviews in Organic Search, visits lost to Direct, and which landing pages to fix.
Prefer Maya AI in Google
Highlight our stories in Search, AI Mode & AI Overviews.
AI referral traffic is the visits that reach your site when someone clicks a link in ChatGPT, Gemini, Claude or another AI assistant. GA4 now files them in a built-in AI Assistant channel when it recognises the referrer, but Google counts clicks from its own AI Overviews and AI Mode as Organic Search, and visits without a referrer land in Direct. Maya reads the ChatGPT-source visits from GA4 and puts them next to your AI visibility. The setup works the same in every global market.
Why AI visits are split across three GA4 channels
A marketing manager who opens GA4 after a month of AI visibility work usually wants one number: how many people came from AI. GA4 does not give one number, because AI visits arrive by three different routes.
The first route is a click from an AI assistant that passes its address to your site. Google added an AI Assistant channel to the default channel group on May 13, 2026 (Google Analytics, What's new). When the referrer matches Google's list of AI assistants, GA4 sets the medium to "ai-assistant" and the campaign to "(ai-assistant)". Google's channel description names ChatGPT, Gemini, DeepSeek, Copilot and Grok as examples, and the May announcement adds Claude (Google Analytics, Default channel group). Google does not publish the full list on either page.
The second route is Google Search itself. The same Google page says the AI Assistant channel excludes Google's AI Overviews and AI Mode, and counts clicks from them as Organic Search. Search Console follows the same logic. Google Search Central states that sites appearing in AI Overviews and AI Mode are included in the overall search traffic in the Performance report, within the "Web" search type (Google Search Central).
The third route is a visit that arrives with no referrer and no campaign tag, which GA4 records as Direct. Maya's research on ChatGPT answers found that around October 9, 2026, part of ChatGPT traffic began appearing as Direct (Maya). Did ChatGPT remove UTM attribution? covers that change, so this guide does not repeat it.
AI crawlers are a separate matter. GA4 automatically excludes traffic from known bots and spiders (Google Analytics), so a crawler reading your page never shows up as a visit. Crawler activity lives in server or CDN logs, and the AI bot access audit explains how to check it.
Five places to read AI traffic, compared
Each place answers a different part of the question, and none of them sees every AI-driven visit. The table sets them side by side, ordered by how much setup they need.
| Where you look | What it counts | AI sources it names | What it misses | Setup |
|---|---|---|---|---|
| GA4 AI Assistant channel | Sessions whose referrer matches Google's AI assistant list | ChatGPT, Gemini, Claude, DeepSeek, Copilot, Grok (examples, not the full list) | AI Overviews and AI Mode; visits with no referrer | None, built into the default channel group |
| GA4 custom channel group | Sessions whose source matches a rule you write | Whatever your rule lists; Google's own example includes Perplexity | Visits with no source information | An admin builds it once and keeps the list current |
| Search Console Performance report, Web search type | Clicks and impressions from Google Search, AI features included | AI Overviews and AI Mode | Visits from AI assistants outside Google Search | A verified property |
| Server or CDN logs | Requests from AI crawlers and agents | Bots that identify themselves | Human visits | Access to your CDN or a log shipper |
| Maya Traffic Analytics | AI-referred sessions, purchases and revenue read from GA4, next to visibility | Currently centred on ChatGPT-source traffic | Clicks without a clean referrer; ad attribution | A read-only Google Analytics 4 connection |
Read the first and third rows together to see the AI-driven clicks that GA4 and Search Console can attribute. Everything that arrives without a referrer stays out of view, so any total you report is a floor, not the full count.
How to track LLM traffic in GA4, step by step
Start with what GA4 already gives you, then add your own rules only where the built-in channel falls short. None of the steps below needs code.
Open the AI Assistant channel
Go to Reports, then Acquisition, then Traffic acquisition. Set "Session default channel group" as the main dimension and find the AI Assistant row. Add "Session source" as a secondary dimension to see which assistants sent the visits. Then add the key events your team already uses, such as a demo request or a purchase, so each AI source shows sessions and outcomes on one row.
Check where Perplexity and other sources land
Google's channel page does not name Perplexity among its examples. Switch the dimension to "Session source / medium" and search for the assistants your buyers use, such as perplexity.ai or claude.ai. If one of them appears with the medium "referral" instead of "ai-assistant", the default list does not cover it in your data.
Build a custom AI channel if sources are missing
Google's help page includes a worked example for exactly this case (Google Analytics, Custom channel groups). An admin creates a custom channel group, adds a channel named "AI assistants" with a "matches regex" condition on the source, and moves that channel above Referral. Google says the order matters for correct processing, and that custom channel groups apply to your reports retroactively. Copy the expression from Google's article rather than writing one from memory, and update it when you add an assistant. Standard GA4 properties allow two custom channel groups, according to the same page.
Visits are not the same thing as visibility. A person can read an AI answer that names you and never click. Mentions, citations and share of voice are measured inside the answers, and the 7 AI visibility metrics guide covers them.
Check whether AI Overviews hurt your Google clicks
GA4 cannot show AI Overviews or AI Mode clicks as their own row, because Google counts them as Organic Search. Search Console is where Google reports them, mixed into the Web search type with every other Google click. Whether your brand appears and is cited in those answers is a separate question, which How to measure Google AI Overviews and AI Mode visibility covers.
What you can do is compare the same page across two periods. Open Performance, filter by page, and compare a period before and after the change you are worried about. If impressions hold steady while clicks fall, one possible explanation is that more people get their answer on the results page. Search Console alone does not tell you which result feature appeared for each query, so treat that reading as a hypothesis to test. Google says that when people click from results pages with AI Overviews, those clicks are higher quality, meaning visitors tend to spend more time on the site (Google Search Central). Check that claim against your own engagement and key events for the page before you accept it.
Match AI visits to the landing pages they reach
Totals tell you how much AI traffic arrived. Landing pages tell you what to change. In the Traffic acquisition report, add "Landing page" as a secondary dimension and look only at the AI Assistant row, or build an exploration with the dimensions "Landing page" and "Session default channel group". You now see which pages AI visitors reach first, how long they stay, and whether they complete a key event.
Then compare that list with the pages AI answers cite. A page cited often but rarely visited may be one that answers fully inside the AI response. A page visited often but never cited may be reached through a few links you do not track. The situations below are the ones marketing teams ask about most, with the first thing to check.
| Situation | What to check first | Why |
|---|---|---|
| ChatGPT sessions fell after October 9 while Direct rose on the same pages | Read the AI Assistant and Direct rows for those pages side by side | Part of ChatGPT traffic began appearing as Direct around that date |
| An AI source you recognise shows up under Referral instead of AI Assistant | Add it to a custom AI channel placed above Referral | Google's default list is not published in full |
| AI answers mention your brand, but GA4 shows almost no AI visits, and you want to validate the visibility data against real customer traffic | Compare the trend of mentions with the trend of AI visits, not one number with the other | People read answers without clicking, so the two never match one to one |
| AI visitors land on category pages that were renamed or removed | Check the landing page view for error pages and old URLs | A visit that ends on a missing page is lost before the page can answer anything |
| Google clicks fell on pages that rank for questions where AI Overviews appear | Compare clicks and impressions for those pages in Search Console | Google reports AI Overviews clicks there, not in GA4's AI Assistant channel |
| You run ChatGPT ads and paid visits mix with organic ones in Google Analytics | Tag ad links consistently and keep a separate channel for paid AI | Measuring ChatGPT Ads performance covers the tagging standard |
Every row starts from a page or a channel you can open today. None of them needs a new tool.
Fix the pages AI visitors land on
Measurement only pays off when it changes a page. The three changes below follow from the landing page view. Each one is a test, not a guaranteed result.
Redirect AI visits that land on dead or outdated URLs
AI answers can link to a page your team has since moved, renamed or retired. When the landing page view shows AI visitors reaching an error page or an old address, set a permanent redirect to the page that answers the same question today. The visitor gets an answer on the first click.
Give the landing page the answer and one next step
An AI visitor arrives with a specific question already half answered. Read the page they land on as that visitor would. Does it state the fact the answer pointed to, such as the price, the setup time or the coverage area, in the first paragraph? Is there one clear next step, such as a demo, a pricing page or a contact form? If not, add both. Whether the change lifts key events is something to measure, not to assume.
Leave zero-click pages alone
A page that AI answers cite often but send few visitors to is not a failed page. The answer may be doing its job: your page shaped what the assistant said. Rewriting it only to force clicks can remove the very passage the answer relies on. Keep tracking how often it is cited, and judge it on visibility, not visits.
Illustrative example: an old pricing page in the AI channel
Illustrative example: the company, pages and figures are invented, not Maya data. A B2B software company opens the GA4 landing page view filtered to the AI Assistant channel and sees that a large share of AI visits land on an old pricing URL that now shows an error page.
- Observation: AI visitors reach the old pricing address and leave without a key event. In Maya's Source Analysis, the citation examples show that AI answers to pricing questions still cite that address.
- Possible cause (hypothesis): the answers rely on a copy of the page from before the site moved its pricing.
- Test: add a permanent redirect from the old address to the current pricing page, and state the plan prices in the first paragraph there.
- Verification: compare AI Assistant sessions and demo requests on the current pricing page for the weeks after the redirect, and check in Maya whether answers start citing the new address.
Read the change against three break dates
Write down the date of every redirect, page change or new channel rule. When you compare periods, mark three outside dates as well, so that a change in the platforms does not look like a change in your pages.
The first is May 13, 2026, the day Google introduced the AI Assistant channel (Google Analytics, What's new). For earlier dates, look for the same assistants under Referral. The other two are September 29–30 and October 9, 2026, when Maya's research recorded shifts in the sources ChatGPT shows and in how its traffic was attributed (Maya). Compare the same landing page across the same number of days before and after your change. A single period of growth is a reason to keep watching, not proof that the change worked.
What this guide does not cover
The guide does not cover paid ChatGPT ads, which follow their own tagging and conversion rules. Nor does it describe how each AI assistant passes referrer information, because the assistants do not document it in full. Consent settings, ad blockers and privacy tools can also remove visits from GA4, and those losses are outside the scope here. Google's channel rules and lists can change; the help pages linked in this guide are the current reference.
Frequently asked questions
How can I validate AI visibility data against real customer traffic?
Use trends, not totals. If AI answers mention you more often over several weeks, check whether AI Assistant sessions to the cited pages move in the same direction. Expect a delay, and expect the visit count to stay far below the mention count.
How do I integrate AI visibility tracking with Google Analytics?
Connect your GA4 property to a tool that reads its reporting data and places AI-referred sessions next to mention and citation data. Maya's Google Analytics 4 connection is read-only and starts on the Premium plan, listed at $399 per month on the pricing page.
Can I track citation changes over time and identify new referring sources?
Yes, in two places. In GA4, a new AI assistant appears as a new "Session source" value in the AI Assistant or Referral rows. Changes in which pages AI answers cite are measured inside the answers, for example in Maya's Source Analysis.
Conclusion: AI referral traffic is a floor, so fix the landing pages first
GA4 now shows AI assistant visits in their own channel, Search Console holds the clicks from Google's AI features, and visits without a referrer stay in Direct. No single report adds them up. The useful work is to read the landing pages AI visitors reach, fix the dead ends, and give each page the answer and next step the visitor came for.
Which AI answers mention your brand, and do those mentions turn into visits? Maya's Traffic Analytics reads AI-referred sessions, purchases and revenue from GA4 and places them next to your AI visibility, so you can read both trends in one place.
Sources
- Google Analytics Help, What's new in Google Analytics (AI Assistant traffic measurement): https://support.google.com/analytics/answer/9164320
- Google Analytics Help, Default channel group: https://support.google.com/analytics/answer/9756891
- Google Analytics Help, Custom channel groups: https://support.google.com/analytics/answer/13051316
- Google Analytics Help, Known bot-traffic exclusion: https://support.google.com/analytics/answer/9888366
- Google Search Central, AI features and your website: https://developers.google.com/search/docs/appearance/ai-features
- Maya, 13 Things That Changed in ChatGPT in 2 Months: https://withmaya.ai/blog/13-things-that-changed-in-chatgpt
- Maya, Did ChatGPT Remove UTM Attribution?: https://withmaya.ai/blog/did-chatgpt-remove-utm-attribution
- Maya Help Center, Connecting Google Analytics 4: https://help.withmaya.ai/integrations/google-analytics
- Maya, Integrations: https://withmaya.ai/integrations
GEO researcher at Maya. Focused on measuring AI visibility — the metrics, benchmarks, and methods behind tracking how brands appear in AI answers.
