ChatGPT is sending you traffic. Can you see what happens next?

Can GA4 track ChatGPT traffic?

Yes. GA4 can identify website visits from recognised AI assistants when usable referral information reaches the site.

Google Analytics now includes an AI Assistants channel for identifiable traffic from services such as ChatGPT, Gemini, DeepSeek, Copilot and Grok. That makes basic AI-traffic reporting considerably easier than it was a year ago.

But there is an important limitation: GA4 can only classify the traffic signals it actually receives. It cannot see the conversation that happened before the click. It cannot tell you the prompt the person used. And it cannot see an AI interaction if the eventual website visit arrives through another channel.

So the correct interpretation is: GA4 can measure identifiable AI referral traffic. Not: GA4 can measure all AI influence.

Start with the native AI Assistants channel

In GA4, start with your acquisition reporting. Look for AI Assistants alongside channels such as Organic Search, Paid Search, Direct, Referral and Organic Social.

Then analyse it like any other acquisition source. You want to understand:

That last question is where the interesting work starts.

Go one level deeper than the channel

The aggregate channel is useful. The individual source is more useful. Look at Session source / medium. You may begin seeing individual AI-originating sources separately. That allows questions such as: Does ChatGPT send the most traffic? Does Perplexity send fewer but better visitors? Which AI source sends people to commercial pages? Which source produces leads? Which source produces qualified leads?

Do not assume the service producing the most sessions is the one producing the most commercial value. It often isn't.

Build a dedicated AI traffic report

I would create a simple exploration or report specifically for AI acquisition.

The landing-page report is particularly valuable

AI traffic behaves differently from a traditional search result. A user may already have described their problem, compared alternatives, narrowed the shortlist, understood the basics and formed an initial preference before clicking. That means the landing page can sit much further down the buyer journey than a normal informational search visit.

Look for patterns. Is ChatGPT sending people to your homepage, a commercial service page, an article, a comparison page, or a case study? That tells you something about which information the AI considered useful enough to surface.

Measure conversion quality, not only conversion volume

Suppose ChatGPT sends 120 sessions and Google Organic sends 6,000 sessions. At first glance, ChatGPT looks irrelevant. But now suppose ChatGPT produces 9 qualified enquiries while Google Organic produces 75. That changes the picture. The smaller traffic source may be disproportionately valuable.

Useful calculations include:

Traffic volume alone tells you almost nothing about economics.

Carry the source past GA4

This is where most implementations stop too early. GA4 knows: ChatGPT sent the visitor. The website knows: the visitor completed the form. The CRM receives: new enquiry. And the source disappears. Now the business can report AI traffic. But it cannot report AI customers.

Where your stack allows it, preserve acquisition information through the form submission. Useful fields can include: original source, original medium, original channel, landing page, first identifiable source, latest identifiable source, lead creation date. Then store those values against the CRM record.

That is much more valuable than a traffic graph. This work sits inside a wider Revenue Measurement Architecture, and the deeper method is covered in AI Search Attribution.

Do not forget first-user vs session acquisition

GA4 can answer different acquisition questions. A session-level view can tell you what brought this visit to the site. A first-user view can help answer what identifiable source first acquired this user. Those are not the same question.

Imagine this journey:

Depending on the report you open, the journey can look different. That does not necessarily mean the data is wrong. It means the report is using a different scope. Understand the question before arguing about the answer.

What about Google AI Overviews and AI Mode?

This needs its own treatment. Google's current GA4 AI Assistants channel explicitly excludes AI Overviews and AI Mode.

Google Search Console is increasingly important here. In 2026, Google began rolling out dedicated generative-AI performance reporting in Search Console, including visibility data for generative AI features in Search. That can include information such as impressions, pages surfaced, countries, devices and performance over time.

So think of the measurement stack as complementary. GA4 helps answer: what did identifiable website visitors do? Search Console helps answer: where is Google surfacing the site? Different questions. Both matter. For the discovery side of this, see How to Measure AEO Performance.

Can GA4 tell me which ChatGPT prompt generated the visit?

No. Do not invent that capability. The referral tells you where the user came from. It does not hand you the conversation that happened inside the AI product. That means you cannot reliably say "this customer searched this exact prompt in ChatGPT" unless you have another defensible data source telling you that. Prompt visibility measurement and referral analytics are different datasets. They can complement one another. Do not pretend they are the same dataset.

AI traffic can still disappear into Direct

Attribution is never as clean as the diagram. Referral information can be absent. A user may also discover through ChatGPT then later Google the brand or type the domain directly. GA4 may legitimately report that later session under another source. That is why measured AI referral traffic is the observable floor, not necessarily the total AI influence. That line matters.

What I would put on the AI traffic dashboard

Keep it simple.

Then compare those numbers with Organic Search, Paid Search, Direct and Referral. AI should not get a special reporting universe simply because it is new.

The measurement standard

I do not want a dashboard whose conclusion is: "ChatGPT sent 312 visits this month." I want:

"ChatGPT sent 312 identifiable visits. Twenty-two became leads, nine were qualified, four reached opportunity stage and one has closed."

Now we can make a decision. That is the point of measurement. When you want to take this all the way through to return on investment, see AI Search ROI, and for the strategy behind the visibility, see AEO & GEO Consulting.