The GA4 AI Assistant Channel: Measuring Visits From ChatGPT and Gemini
If you have opened Google Analytics 4 recently and spotted a channel called AI Assistant, it is new. Google's default channel group documentation describes it as the channel for visitors arriving from sources like ChatGPT, Gemini, Deepseek, Copilot or Grok. It gives you a first, partial view of how many people reach your site after asking a chatbot. It does not include Google's own AI Overviews or AI Mode, and it does not catch every chatbot visit, so read it as a floor rather than a total.
How the channel is defined
GA4 puts each session into a channel using rules. For this one, the rule is that the medium exactly matches "ai-assistant". GA4 sets that medium automatically, with the campaign set to "(ai-assistant)", when the referrer matches its list of AI assistants.
In plain terms: when someone clicks a link inside a chatbot's answer and the browser passes on where they came from, GA4 recognises the chatbot and files the visit under AI Assistant. Before this change, those visits generally appeared under Referral, mixed in with links from directories and partner sites.
What it leaves out
Google's own AI features
Visits from AI Overviews and AI Mode in Google Search are excluded from the AI Assistant channel, so this report says nothing about them. If you want to understand the effect of Google's AI answers, the AI Assistant channel will not tell you; Search Console and your organic trends are the better tools. Google's guidance on AI features in Search explains that these features draw on the normal index, so the usual search reporting still applies.
Visits with no referrer
The channel depends on the browser passing on a referrer. Some apps, and some privacy settings, strip that information. Those visits land in Direct, alongside people who typed your address or used a bookmark. If your Direct traffic has crept up without an obvious cause, some of it may be chatbot visits you cannot label.
Older data
The classification applies from when Google introduced it. Earlier chatbot visits stay wherever they were originally filed, usually Referral. Comparing AI Assistant traffic with last year will show a jump that is partly a reporting change, not a real one.
Reading the numbers sensibly
For most small firms, AI Assistant will be a small channel for now. That is not a reason to ignore it. Some questions worth asking:
- Which pages do these visitors land on? Open the Traffic acquisition report, filter to the AI Assistant channel and add Landing page as a secondary dimension. If chatbot visitors arrive on your pricing or service pages, that suggests the bots are recommending you for commercial questions.
- Do they enquire? If you have form submissions or calls set up as key events, compare the rate for this channel with Organic Search. A small channel with a high enquiry rate deserves attention.
- Which assistant sends them? Switch the dimension to Session source to see the split between assistants. A sudden change in one source can mean an answer about your business has changed, for better or worse.
Avoid reading too much into week-to-week movement. With small numbers, one person returning three times can look like a trend.
Checking what the chatbots actually say
Analytics tells you that someone arrived, not what they were told first. Once a quarter, ask two or three popular assistants the sort of questions your customers ask: "who does boiler servicing in [your town]", "is [your business] open on Sundays", "how much does [your service] cost locally". Note whether you are mentioned and whether the details are right.
If an answer is wrong, trace it back. Wrong opening hours usually come from an out-of-date listing somewhere. A vague description often means your own site does not state clearly what you do. The cure is the same groundwork that helps ordinary search, which we covered in local SEO basics, and copy that says plainly who you serve, as in writing website copy that turns visitors into enquiries.
Why every channel undercounts a little
Bear in mind that GA4 only records the visits it is allowed to record. If your cookie banner asks for consent before analytics runs, people who decline will not appear in any channel, chatbot or otherwise. Ad blockers and some browsers have a similar effect. That does not make the reports useless, but it does mean the trend over time is more trustworthy than any single figure, and it is another reason to count enquiries directly as well.
Setting up a simple monthly view
You do not need a dashboard project. A short monthly note is enough:
- Sessions from AI Assistant, Organic Search and Direct, compared with the previous month.
- Top five landing pages for AI Assistant visitors.
- Enquiries or key events from each of those channels.
- Any odd change worth checking, such as a new source appearing.
Keep the note in the same document each month and patterns will show up within a quarter. If you work with an agency, ask them to add the AI Assistant channel to their regular report; it takes them minutes.
Where this fits
The new channel is useful because it turns a vague worry ("are people finding us through AI?") into a number, even an incomplete one. Treat it as one signal among several. The things that drive enquiries have not changed much: clear pages about real services, accurate details everywhere you are listed, and a site that makes contacting you easy. If those are in order, you are well placed whichever way people choose to search. For the separate question of whether to let Google use your pages in its AI answers, see our piece on the CMA's AI Overviews opt-out.