Table of Contents
ToggleShort answer: GA4 AI traffic is not one clean number. Google’s default AI Assistant channel counts visits from the assistants on its list. Google’s AI Overviews and AI Mode are counted under Organic Search, and visits that arrive with no referrer are not counted as AI at all. To get a reliable view, read the channel rule, build a custom channel group above Referral, and check it in an Exploration.
This guide is for marketers, ecommerce owners and agencies who run GA4 and want to know what their GA4 AI traffic report really shows. It does not repeat the basic “how to see ChatGPT in GA4” steps. It answers three questions: what the default channel catches, what it misses by design, and how to build a GA4 AI traffic view you can trust anyway.
Every statement about Google’s behavior below was checked against Google’s help pages on 1 November 2026. Where Google’s pages are silent, the text says so instead of guessing.
Google’s page on traffic-source dimensions defines a channel called “AI Assistant” in the default channel group. The rule says the medium must exactly match “ai-assistant”. A second rule says that when the referrer matches Google’s list of AI assistants, GA4 sets the medium to “ai-assistant” and the campaign to “(ai-assistant)”. The description names ChatGPT, Gemini, Deepseek, Copilot and Grok as sources.
Google’s “What’s new” page has an entry dated 13 May 2026, titled “New AI Assistant traffic measurement”. It says Google Analytics now provides a dedicated way to measure traffic from popular AI assistants, and it names ChatGPT, Gemini and Claude.
These two pages do not match. The definition page does not name Claude, and the “What’s new” entry does. I am reporting that as a difference between Google’s pages, not picking a side. The definition page also does not list the referral domains behind the rule. So Google’s pages alone cannot tell you whether a given assistant is on the list. Your own data can, and fix 1 below shows how.
One more open point: Google does not say whether the channel applies to data from before it existed. A few third-party write-ups, such as lovesdata.com and zenda.com.ar, report that it only applies going forward and that older AI visits stay where they were, mostly in Referral. Google does not confirm that, so treat it as a report, not a fact.
Google’s definition says the AI Assistant channel excludes Google’s AI Overviews and AI Mode. Those visits are counted under Organic Search, which the same page describes as including them. If you want one figure for GA4 AI traffic, visitors who click from AI Overviews or AI Mode are mixed in with ordinary search visitors, and the default channel cannot separate them.
Google defines Direct as visits where the source is “(direct)” and the medium is “(not set)” or “(none)”. A visit that arrives with no referrer gives GA4 no source to match against its AI list, so in practice it lands in Direct. A person can read an answer in an assistant, open your link, and still show up as Direct. Google publishes no share for how often this happens, and I will not invent one. What you can say is that the AI Assistant channel is a floor for AI visits, not a total.
Perplexity is not named on the definition page, and neither is Claude. An assistant that is not on the list is not counted in the AI Assistant channel. It lands in whichever channel its source matches. Some third-party sources report that Perplexity visits end up in Referral, and Google’s pages do not confirm it. Check your own Traffic acquisition report by session source and medium before you assume anything.
Each fix below builds on the one before it. Together they give you a view of your GA4 AI traffic that does not depend on Google’s list being complete.
Before you change anything about your GA4 AI traffic setup, open your Traffic acquisition report and look at session source and medium. Search for “ai-assistant” and for the assistants you expect, such as chatgpt, gemini, copilot, perplexity and claude. Write down which source values are inside the AI Assistant channel, and which sit in Referral or Direct.
That list is your gap list. It is also the source list you will use in fix 2. Do this on a long date range, because small sites may see only a handful of AI visits per month.
Google’s page on custom channel groups says a standard property can have 2 custom channel groups, an Analytics 360 property can have 5, and each group can hold up to 50 channels. It also says custom channel groups can be applied to your reports retroactively. That matters, because it means a new channel can show older data as well.
Create a new group, name it something clear such as “Channels with AI”, and add a channel called “AI Traffic”. Define it with conditions on session source, using the source values you found in fix 1. As an example, a pattern like the one below matches several common names. Replace it with the values you actually saw, not with this list:
chatgpt|openai|perplexity|claude|gemini|copilot|deepseek|grokKeep the pattern tight. A loose pattern can pull in unrelated sources, and the channel is only as good as its rule.
Google’s page says traffic goes into the first channel whose definition it matches, based on the current order of channels in the group. So order decides the result. If Referral comes first, an AI visit that arrives as a normal referral is caught there, and your AI channel never sees it.
Move your AI channel above Referral and check the order again after you save. Google’s page also says a new group starts from a copy of the default group, so the default AI Assistant channel will be in your list too. Keep your own channel ahead of Referral, and look at where the copied AI Assistant channel sits so the two do not fight over the same visits.
Google’s “What’s new” page has an entry dated 11 June 2026, titled “New Source Group field”. It says a Source Group dimension consolidates source values into one value per platform. For AI sources it says the field includes built-in grouping for ChatGPT (OpenAI) and Perplexity. It also says the field is populated retroactively, so you can analyze historical data with it.
Here is the part Google does not say: where the field is available. The entry does not name reports, Explorations or the API. Third-party write-ups disagree. Some say it is in standard reports and Explorations, and others say it only appears in the Advertising workspace, in the Conversion performance report. I could not confirm either from Google’s pages.
So the practical step is simple. Look for “Source group” in the Advertising workspace and in the dimension list of an Exploration in your own property. If you find it, use it next to your custom channel group as a second view that includes Perplexity and covers old data. If you do not find it, skip this fix, because fixes 2, 3 and 5 do not depend on it.
A channel rule that has never been checked is a guess. Build a free-form Exploration with session source and medium as rows, and your custom channel group as another dimension. Use sessions as the metric.
If your property lets you pick a custom channel group as a dimension in an Exploration, this works as written. Google’s page lists the places custom groups do not work, which are covered in the limits section below.
Visits only matter if they do something. Once your GA4 AI traffic channel is clean, compare it with Organic Search and Direct on sessions, key events and key event rate, using your custom channel group in the report or Exploration. Look at the same date range for each channel.
Keep the sample size in mind. If the AI channel has a few dozen sessions, a single sale changes the rate a lot. Read the direction over several weeks, not one week. Also remember that your AI visits are a floor, since some arrive as Direct or Organic Search, so the real number of people who came from an assistant is higher than your channel shows.
No setup makes GA4 AI traffic perfect. These are the limits to state in any report you share.
The BigQuery limit has a practical effect. If your reporting runs on BigQuery, your channel logic will not come along. In practice you would rebuild the same source rules in your own query, using the source and medium values from fix 1.
Yes. Google’s help page names ChatGPT as a source for the AI Assistant channel, which counts visits when the referrer matches Google’s list of AI assistants. Visits that arrive with no referrer are not counted there.
Google defines Direct as visits with the source “(direct)” and the medium “(not set)” or “(none)”. A visit with no referrer gives GA4 nothing to match against its AI list, so in practice it lands in Direct. Google publishes no figure for how often that happens.
Not in the AI Assistant channel. Google’s definition says that channel excludes AI Overviews and AI Mode, and the Organic Search description includes them.
Google’s pages differ on Claude. The definition page does not name it, and the 13 May 2026 “What’s new” entry does. Perplexity is not named on the definition page. Google’s Source Group entry names ChatGPT (OpenAI) and Perplexity for built-in grouping.
Google does not say whether the channel applies to older data. Some third-party write-ups report that it only applies going forward. A custom channel group, by contrast, can be applied to your reports retroactively, according to Google.
No. Google’s page says custom channel groups are not available in the BigQuery Export Schema, and that they cannot be used in the Key events paths report. If you work in BigQuery, rebuild the source rules in your own query.
Sources: Google Analytics Help, Traffic-source dimensions, Google Analytics Help, What’s new in Google Analytics and Google Analytics Help, Custom channel groups.
I audit and fix GA4 tracking setups, and you can read more about my work on alfahadhridoy.com.