GA4’s AI Assistant channel counts the visits whose referrer matches a list of AI assistants that Google keeps, so it shows the lowest figure for your ChatGPT and other AI traffic, not the total. Google’s AI Overviews and AI Mode are left out by design, an AI visit that arrives with no referrer and no campaign tags falls under Direct, and one analyst found chatgpt.com visits filed in three channels in the same property.

In this article · 7 sections
- 01How a visit gets into the AI Assistant row
- 02Check 1: read the floor and the day it starts
- 03Check 2: find AI sources filed under other channels
- 04Check 3: rebuild the before-and-after with a source-based channel group
- 05Check 4: name the blind spots you cannot measure
- 06Writing the AI line for a report
- 07Follow-up questions on AI traffic in GA4
Google added the channel to the default channel group in an update that its What’s New in Google Analytics page dates to 13 May 2026. Any channel comparison that crosses that date needs rebuilding before it goes into a report.
How a visit gets into the AI Assistant row
Google’s Default channel group help page, checked on 20 November 2026, sets the rule in two steps. “The medium is set to “ai-assistant” and the campaign is set to “(ai-assistant)” if the referrer matches a list of AI Assistants.” The channel then takes every visit whose medium is exactly ai-assistant.
So for GA4 to set that medium itself, a visit needs a referrer, and that referrer has to be on the list. Google does not publish the list. The same page describes the channel as covering visits from “sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok”, which reads as a set of examples. The May announcement named ChatGPT, Gemini and Claude; the current help page does not mention Claude. Neither page tells you whether Claude or Perplexity visits match today. Your own data does, in check 2.
The same help page says the channel “excludes Google’s AI Overviews and AI Mode.”
The launch date causes a different problem. GA4 assigns the medium when a visit’s referrer matches, and Google documents retroactive application for custom channel groups, not for this channel. Taken together, those facts point to the most likely result: AI visits recorded before the channel existed kept the classification they had then, usually Referral. Google does not say this in as many words, but it is the safe reading for any year-on-year view: the AI Assistant row starts from nothing, and Referral gives up some visits at the same point. That shift moves visits between rows. It adds none to the site.
Check 1: read the floor and the day it starts
Open your traffic acquisition report by default channel group and set the widest date range the property holds. Write down two things about the AI Assistant row: its sessions for the period you report on, and the first day it shows any visits.
The sessions are your floor, the visits GA4 matched to an assistant it recognises. The first day is your boundary. Google’s date is 13 May 2026, but that is when the update was announced, so don’t assume your property started counting that day. A comparison of this row whose earlier period sits before your boundary measures the arrival of a label. Keep it out of any growth claim.
Check 2: find AI sources filed under other channels
Build a view with session source as a dimension across all channels, without filtering to AI Assistant. Filter the source for the AI domains you expect: chatgpt.com, perplexity.ai, and anything containing gemini, copilot, claude, deepseek or grok. Show each source’s channel next to it.
Itamar Blauer, writing in Search Engine Journal on 15 July 2026, found that chatgpt.com traffic in one property “was sitting in three different channels at once”: AI Assistant, Referral and Unassigned. He also reported that Perplexity “keeps landing in Referral”. That is one analyst looking at one property, so take it as a reason to look at yours.
Every session from an AI source that sits in Referral or Unassigned belongs in your AI number. Sessions already in the AI Assistant row are in the floor from check 1. Add only what sits elsewhere, or you count them twice.
Google’s What’s New page also announced a Source Group field on 11 June 2026, with built-in grouping for ChatGPT and Perplexity. A third-party guide reported the field missing from one live property in July, so check whether your property has it before you rely on it. Session source does the job either way.
Check 3: rebuild the before-and-after with a source-based channel group
The default channel group can’t give you a like-for-like comparison across the launch date. A custom channel group can. Google’s Custom channel groups help page, checked on 20 November 2026, says “Custom channel groups can be applied to your reports retroactively.” The same page says you need the Editor role or above at the property level, and that a standard property gets two custom groups (a 360 property gets five) with up to 50 channels each. The fields custom channel groups support include Source, Medium, Campaign name and Source platform.
Define an AI channel on Source, using the domains you found in check 2 rather than a list copied from someone else’s post. Set up the other channels your reports rely on alongside it, so the rest of the report still reads the way people expect. Because the group applies to your reports retroactively, visits from those sources recorded before the launch date fall into the same AI channel as the ones after it. That is the comparison worth showing a board or a client.
Spend your custom groups deliberately. A standard property has two, and a group someone built to test an idea takes one up just as surely. Agree who owns the group, and give it a name nobody will mistake for the default.
Check 4: name the blind spots you cannot measure
Two kinds of AI-driven visit stay out of reach of every check above.
The first is clicks from AI Overviews and AI Mode. Google Search Central’s AI features and your website page, checked on 20 November 2026, says these features “are included in the overall search traffic in Search Console”, reported in the Performance report within the Web search type. Google describes no separate split. In GA4 those clicks come from Google, and the Organic Search rule on the default channel group page takes visits whose source is on Google’s list of search sites or whose medium is organic. They most likely sit in Organic Search with every other search click, and no Google page separates them there either.
The second is AI visits that arrive with no referrer. Direct takes visits whose source is (direct) and whose medium is (not set) or (none). Google’s page on (direct) / (none) traffic says this channel “represents website traffic that doesn’t have a clear referral source.” An AI visit with no referrer and no campaign tags fits that rule, and nothing left on the visit says it came from an assistant. That page doesn’t mention AI assistants, and no published figure says how much AI traffic sits in Direct.
Name both in the report and leave them unquantified. A guessed percentage inside Organic Search or Direct would look precise and rest on nothing.
Writing the AI line for a report
The line that goes into a monthly or annual report reads: at least X AI-referred sessions (the AI Assistant row plus AI sources recovered from other channels), plus an unknown share inside Organic Search and Direct. Don’t reduce it to one growth percentage. If you compare periods, compare floors built from the custom channel group on both sides.
That report line settles the two arguments this channel tends to start. If the floor is close to zero, look at check 2 and the two unknown rows before anyone writes that AI sends you nothing. If it is rising, the custom group shows whether the rise began before the label did, and that answer decides whether AI search earns more budget or more page updates.
The numbers below are illustrative, not from a client: one hypothetical month in one property.
| Bucket | Where GA4 files it | Sessions | In the floor |
|---|---|---|---|
| Recognised AI assistants | AI Assistant | 420 | Yes |
| AI sources found in Referral | Referral | 150 | Yes |
| AI sources found in Unassigned | Unassigned | 30 | Yes |
| AI Overviews and AI Mode clicks | Organic Search | Unknown | No, named in the report |
| AI visits with no referrer | Direct | Unknown | No, named in the report |
| Floor | AI Assistant plus recovered sources | 600 | Yes |
Here the AI Assistant row alone would report 420 sessions, 70% of the floor. The 180 sessions recovered from Referral and Unassigned are the other 30%, and neither figure tells you anything about the two unknown rows. Copy the table and replace the rows with your own.
Follow-up questions on AI traffic in GA4
Why did AI traffic jump in Google Analytics this year?
Part of a jump that starts around 13 May 2026 can be the AI Assistant channel itself. Google added it to GA4’s default channel group in an update dated that day, and AI visits recorded earlier most likely still sit under Referral, so a channel report across that date compares a new row with an empty one. A custom channel group that matches AI sources, applied retroactively, shows whether the visits themselves grew.
Can I see AI Overviews clicks separately in GA4 or Search Console?
No. GA4’s AI Assistant channel excludes Google’s AI Overviews and AI Mode, and Search Console counts them in overall search traffic within the Web search type. Google’s documentation describes no separate split in either tool, so report them as an unknown share of search traffic.




Leave a comment