How to Track AI Traffic in GA4 and Find What It Misses

GA4's AI Assistant channel misses part of your AI traffic. Build a custom channel that catches more, see which tools drop the referrer, and check your logs.

GA4 now puts some AI visits in their own AI Assistant channel, but that channel misses part of your AI traffic. To see more of it, build a custom channel group that matches AI sites by source, let it catch ChatGPT's tracking tag too, and accept that visits with no referrer will still land in Direct. Search Console and your server logs fill in two more pieces: AI Overview impressions, and which pages AI assistants fetch for their users.

What GA4 counts as AI traffic by default

According to Search Engine Journal's GA4 walkthrough by Itamar Blauer, Google added an AI Assistant channel to GA4's default channel group on May 13, 2026. Google's default channel group page describes it as visits from sources like ChatGPT, Gemini, DeepSeek, Copilot or Grok. When the referrer matches Google's list of AI assistants, GA4 sets the medium to ai-assistant, and the channel collects every session with that medium.

Two gaps follow from that definition. First, the list is short and it changes. SEJ reports that Claude was named at launch and later dropped, and that Perplexity is not on the list, so its visits keep landing in Referral. Second, Google's own AI Overviews and AI Mode are left out on purpose: the same Google page counts them under Organic Search.

Why one AI source shows up in three channels

GA4 picks a channel from source and medium together, not from source alone. SEJ found that sessions from chatgpt.com split three ways. With the ai-assistant medium they go to AI Assistant. With the referral medium they go to Referral. With no medium, shown as (not set), they go to Unassigned, which is the value Google's page says GA4 uses when no channel rule matches.

So reading the AI Assistant channel on its own undercounts. To see the split on your own property, open the Traffic acquisition report, add Session source / medium, and search for chatgpt.

How to build a custom AI channel in GA4

A custom channel group fixes the split because you write the rule, and you can match on source while ignoring medium. Google's custom channel groups help page says custom channel groups can be applied to your reports retroactively, so the new channel also picks up older sessions that sat in Referral. A standard property allows two custom groups on top of the default one.

Open Admin in GA4. Under Data display, click Channel groups.
Click Create new channel group. GA4 starts you with a copy of the default group, so the usual channels are already there.
Give the group a name you will recognize, such as Channels with AI.
Add a new channel, name it AI, and set the condition to Source matches regex, using the pattern below.
Click Reorder, drag AI above Referral and Organic Search, then click Apply. GA4 puts each session in the first channel whose rule it matches, so the order decides which channel wins.
Save the group, then choose it as the primary dimension in your acquisition reports.

Here is our starting pattern. It lists the source values of the main assistants as full domains:

Three notes on it. In GA4 a regex has to match the whole value by default, as Google's regex help page explains, so each entry is a complete domain, not a fragment. Before you save, look at your own Session source report and add any variant you see there, such as a www. version. And never add a bare word like google, or the channel will swallow ordinary organic search. SEJ gives the same warning and suggests reviewing the pattern every quarter, because assistants come and go and their domains may change.

What the custom channel still cannot see

A channel group can only sort sessions that arrive with a source. A visit with no referrer is filed as Direct, and no rule can move it, because there is nothing to match. Whether a referrer arrives depends on the tool, the app and sometimes the account.

Patrick Stox of Ahrefs tested several assistants by checking the referrer in the browser console. A link in his paid ChatGPT account had a noreferrer attribute, while the same type of link in a free account did not. Perplexity's website seemed to send referrers on everything, and its desktop app on nothing. Copilot on the web passed the referrer, Copilot in Windows did not. Claude seemed to pass it everywhere he tested, Grok did not, and Mistral and Meta AI on the web did. His test was manual, and these tools change, so treat it as a method to repeat, not a ranking of products.

AI browsers behave differently again. MarTech's comparison by Dan Taylor was published in November 2025. It says sessions from Perplexity Comet often appear with source perplexity.ai and medium referral, while links opened in ChatGPT Atlas often lose the referrer and may show up as Direct or occasionally (not set).

ChatGPT does leave one trace you can find. Ahrefs found that links in ChatGPT's Sources list add ?utm_source=chatgpt.com to the URL. GA4 reads that tag as the session source, so a tagged visit shows as chatgpt.com even without a referrer, and the custom channel above catches it. Clickport notes that the tag sets the source only, which is why chatgpt.com / (not set) lands in Unassigned, and that UTM parameters do not appear in GA4's Landing page + query string.

How much AI traffic goes missing

A figure that circulates in SEO content says about 60% of AI traffic lands in GA4 as Direct. The newsletter writer David Rosam traced it to one page: the analytics vendor Loamly's The AI Traffic Attribution Crisis by Marco Di Cesare. A February 2026 update there replaces that estimate. It reports that 14,413 of 20,428 AI visits it identified arrived without referrer headers. That share is 70.6%. The whole dataset held 446,405 visits. Rosam points out that this is a share of AI visits, not of all traffic.

Rosam also shows the scope: the 20,428 AI visits are about 4.6% of the 446,405 total. So the figure describes a small slice that one vendor's method had already flagged as AI. It does not say that most of your Direct traffic is AI.

A second vendor's numbers show how much the rate moves. Clickport's analysis reports that identified AI visits with a source tag but no referrer rose from 35.7% (935 of 2,619) in April to 49.7% (958 of 1,928) in October 2026. Different definition, different customers, different month. Neither is a universal rate, so measure your own.

Check Search Console and your server logs

GA4 sees clicks. Search Console shows a different part of the picture for Google's own AI features. Its generative AI performance report shows impressions in AI Overviews and AI Mode, which Google defines as how many times links to your site were shown in those features, grouped by page, country, device and date. It reports impressions, not clicks. Google's AI features guide says clicks from AI Overviews and AI Mode are included in the overall search traffic in the Performance report, under the Web search type, so they sit inside your normal Google numbers. Our post on what the Search Console AI report can and cannot show goes further.

Server logs show something GA4 never sees: AI tools fetching your pages to answer a user's question. OpenAI, Perplexity and Anthropic each name the agent they use for this. OpenAI's crawler page says ChatGPT may visit a page with the ChatGPT-User agent when users ask it a question. Perplexity's crawler page describes Perplexity-User the same way, and Anthropic's help article on its crawlers says Claude may access websites using a Claude-User agent. Do not mix these up with the training crawlers, such as GPTBot and ClaudeBot, which collect pages for model training.

A request from one of these user agents is not a visit by a person, and it does not mean anyone clicked. It does tell you which pages assistants are reading when people ask about your topic, which is useful next to your GA4 numbers.

A monthly routine

This is our rule of thumb for keeping the numbers honest. Once a month, note the sessions in your custom AI channel by source, the AI impressions in Search Console, and the user-agent fetches in your logs. Also watch Direct visits to deep URLs that no internal page links to: a rise there is a pattern worth noticing, not a measurement. A steady baseline across months tells you more than any single figure.

For related reading, see our posts on tracking parameters from AI Overviews and a four-week log of AI answers, and use the steps in how to trace a backlink statistic to its source on any AI traffic percentage you meet.