Why AI traffic is invisible
GA4 has no native 'AI' channel. The traffic sent by ChatGPT, Perplexity, or Gemini is scattered across 'Referral', 'Organic Search', and above all 'Direct', with no grouping or dedicated label. The result: you do not see it.
The cause is mechanical. GA4 attributes a source based on the HTTP referer header passed by the browser, then applies channel classification rules. But AI assistants blur that signal in two ways. First, some propagate a referer (perplexity.ai, gemini.google.com) that GA4 files under 'Referral', buried among all your other referrers. Second, many clicks come from the native app or a context that sends no referer: GA4 then has nothing to attribute and shifts the visit to 'Direct'.
The problem is strategic. GEO is now a fully fledged acquisition channel — ChatGPT exceeds 900 million weekly users, and more than 50% of Google queries trigger an AI Overview. Without dedicated measurement, you cannot connect your optimizations to clicks, leads, or revenue. You are steering blind a channel that carries more and more weight.
GA4 has no AI channel. You have to create it manually through a custom channel group that bundles the referring domains of the assistants. Without it, your GEO traffic is diluted into 'Referral' and 'Direct', and your AI visibility strategy stays unmeasurable.
The list of AI referrals to track
To isolate AI traffic, you first need to know the referring domains the assistants pass along. These are what will serve as the grouping rule in GA4.
Here are the domains to watch, organized by assistant. This list evolves: vendors add subdomains and variants across versions. Check your referral reports every quarter to keep it up to date.
| Assistant | Referring domains | Default GA4 behavior |
|---|---|---|
| ChatGPT | chatgpt.com, chat.openai.com | Referral or Direct depending on context |
| Perplexity | perplexity.ai, www.perplexity.ai | Referral (propagated referer) |
| Gemini | gemini.google.com | Referral, sometimes Organic Google |
| Copilot | copilot.microsoft.com, bing.com/chat | Referral or Organic Bing |
| Claude | claude.ai | Referral (propagated referer) |
| Others | you.com, poe.com, grok.com | Referral |
Note the important nuance: these domains only cover clicked traffic. AI citations that generate no click — zero-click answers, mentions in an AI Overview with no visit — will never appear in GA4. To measure that visibility, traffic tracking is not enough: you need citation tracking and AI share of voice monitoring in parallel.
Creating the custom GA4 channel
The grouping is created through a custom channel group in GA4's admin. The operation takes a few minutes and is fully reversible.
In GA4, click Admin (the gear at the bottom left), then, under the Property column, click 'Channel groups'. There you see the default group, which is not editable. Click 'Create channel group'.
Name the group 'Acquisition with AI', for example. Inside it, add a new channel named 'AI Assistants' and place it at the top of the list: GA4 applies rules in order, and the first matching channel wins.
For the 'AI Assistants' channel, set the condition: 'Source' matches one of the regular expressions bundling your domains, for example chatgpt.com|openai.com|perplexity.ai|gemini.google.com|copilot.microsoft.com|claude.ai. Use the 'matches regex' operator.
Save. Verify that 'AI Assistants' is indeed ranked before 'Referral' and 'Organic Search', otherwise those channels will capture the traffic first. The priority order conditions the entire measurement.
The custom group only applies to data collected from its creation onward, not retroactively across all reports. Allow 24 to 48 hours before seeing the first consolidated volumes in the new channel.
Once this channel is in place, your standard acquisition reports will display 'AI Assistants' as soon as you select the custom channel group in the dimension picker. That is the baseline. For fine-grained analysis, move on to exploration.
Building the exploration report
The free-form exploration report lets you cross AI traffic with landing pages and conversions, something standard reports do not do well. This is where the data becomes actionable.
Go to 'Explore' then 'Free form'. Import the following dimensions and metrics, then build the table.
| Element | What to use | Why |
|---|---|---|
| Primary dimension | Default channel group (custom) | Isolates the 'AI Assistants' row |
| Secondary dimension | Source / Landing page | Identifies which assistant and which page convert |
| Metrics | Sessions, Users, Engagement | Volume and quality of AI traffic |
| Key metric | Conversions, Revenue | Links GEO to revenue |
| Filter | Channel group = AI Assistants | Excludes all other traffic |
Apply a filter on the custom channel group to keep only 'AI Assistants', add 'Source' and 'Landing page' as rows, and place sessions, engagement rate, and conversions as values. You get a table that answers three questions: which assistants send you traffic, on which pages, and with what business result.
92% of AI Overview citations come from the top 10, but 47% concern positions 5 to 10. Your exploration report cross-referenced with Search Console reveals which pages, even non-first ones, capture AI traffic — a valuable signal for prioritizing your optimizations.
Save this exploration and share it with your team. Compared month over month, it shows the real evolution of your AI channel. To go further and benchmark your performance against your competitors, cross these volumes with an LLM competitive benchmark.
The blind spots of measurement
No GA4 tracking captures 100% of AI traffic. You must interpret your numbers while knowing three structural limits, or risk false conclusions.
Masked Direct traffic
When an assistant does not propagate a referer — a click from the native app, an answer with no tagged outbound link — GA4 classifies the visit as 'Direct'. An irreducible share of your AI traffic therefore stays invisible. An unexplained jump in 'Direct' coinciding with a GEO campaign is often disguised AI traffic: cross-check it against your landing pages with high citation potential.
Zero-click
Most of GEO's value is not measured in clicks. A brand cited in a ChatGPT answer gains awareness and authority without any visit being recorded. Recall that off-site mentions correlate at 0.737 with AI citations, versus 0.266 for Domain Rating (Ahrefs analysis of 200,000 domains, Dec. 2025): the lever is presence, not only the click.
Low source overlap
Only 11% of domains are cited by both ChatGPT and AI Overviews. Tracking a single assistant gives a partial view. Your report must segment by source to reveal these gaps and guide platform-specific optimizations.
GA4 traffic tracking measures clicks, not visibility. Always combine it with citation and share-of-voice tracking to steer GEO in full. 'Direct' and zero-click hide a value your dashboard will never show.
To know where you stand before even configuring GA4, test your presence on the assistants with the AI Visibility Score: it tells you whether you are already cited, and on which queries.
Our free GEO audit reveals where your AI traffic comes from, what you are leaving on the table, and the priority optimizations to become a cited source.
Questions fréquentes
Why does ChatGPT traffic show up as Direct in GA4?+
Because ChatGPT does not always pass a referer header. When the user clicks from the native app or from an answer with no propagated referer, GA4 has no source to attribute and classifies the visit as 'Direct'. A portion of AI traffic therefore stays structurally invisible, even with a well-configured custom channel.
Should I create the channel in a new GA4 property?+
No. The custom channel group is created in an existing property, via Admin then Channel groups. It does not affect the default group and remains reversible. Note: custom groups only apply to future data and to certain reports, not retroactively everywhere.
Which referring domains should I include for AI traffic?+
At a minimum chatgpt.com, chat.openai.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, and claude.ai. Add the mobile variants and subdomains. The list evolves: check your referral reports every quarter to capture the assistants' new domains.
Does UTM tracking work for AI traffic?+
Not directly, because you do not control the links generated by the AIs. You cannot tag a ChatGPT answer with UTM parameters. The only reliable method remains grouping by referring domain in a custom channel, complemented by an analysis of landing pages with strong AI signal.
