How do you see ChatGPT and Perplexity visits in GA4?

AI visits are the sessions that start on your site after a reader clicks a link in the answer of a chat engine such as ChatGPT, Gemini, Perplexity, Copilot or Claude. They record the click, not the whole effect of the answer.

Three things never count as one. A reader who read your name in the answer and stopped there leaves no visit, which is what people call zero-click search. A reader who copied your name and then searched for it on Google reaches you from search, not from the engine. And Google counts clicks from Google AI Overviews and AI Mode as organic search, like any other search result.

That is why we treat visits as a supporting indicator, as in how to measure the success of optimising for AI answers. They tell you an answer sent you a reader. They do not tell you how often answers named you, on which questions, or who was cited instead of you.

In a channel Google added to the default channel group on 13 May 2026, called AI Assistant (in Google's Arabic interface, «المساعد المستند إلى الذكاء الاصطناعي»). Google says it is the channel by which users arrive from sources like ChatGPT, Gemini, Deepseek, Copilot or Grok, and that it excludes AI Overviews and AI Mode.

Google puts a visit in this channel when its referrer matches a list of AI assistants it keeps, setting the session medium to ai-assistant and the campaign to (ai-assistant). It does not publish the full list. These are the steps to read it:

1. Open Google Analytics 4 and go to Reports, then Acquisition, then Traffic acquisition.
2. Read the AI Assistant row in the channel column: it is the total of your visits from chat engines over the period you chose.
3. Switch the dimension to Session source / medium to see each engine by name, such as chatgpt.com or gemini.google.com, with ai-assistant beside it as the medium.
4. Read the Key events column beside each source: how often those visitors completed something you defined as a key event, such as a quote request or a booking.
5. Compare a month with the month before, not a day with a day: small numbers swing a lot from one day to the next.

By source. Every visit with a known source carries the name of the site it came from, and you can type part of it into the search bar above the report's table, as Google itself suggests: chatgpt, perplexity or copilot, and only matching rows remain.

Google does not list Perplexity among its examples for the AI Assistant channel, although on 11 June 2026 it named Perplexity among the sources grouped by a new Source Group field. If you cannot find perplexity.ai in the channel, look for it under Referral, the channel for visits from links on other sites.

If you see utm_source=chatgpt.com in your landing-page URLs, know that the utm_source tag tells Google Analytics 4 the name of the source, as Google explains on its page about custom campaign URLs. The source is recorded as chatgpt.com; it is not a campaign you launched.

Google does not say that the new channel reclassifies visits from before 13 May 2026. If you compare with earlier months, look for the engines by their source names under Referral, and do not read a jump in the channel in May as sudden growth: it may simply be visits moving from one column to another.

Because Google Analytics 4 relies on the referrer, the address of the page the visitor came from, and that address does not always arrive. Google counts a visit as Direct when the source is (direct) and the medium is empty, as when a visitor types your address or opens a saved link. The same can happen to a link copied from an answer and sent in a message, or opened from an app that does not pass its source on.

Google AI Overviews and AI Mode are a different case. Google places their clicks in Organic Search with the rest of its results, and its channel page says plainly that Organic Search includes AI Overviews and AI Mode. You will not find them in the AI Assistant channel however hard you look.

So the number in the channel is a floor, not the whole effect. It leaves out visits from apps that do not pass their source on, and journeys that started with an answer and ended with a search for your name.

They add what Google Analytics 4 cannot see: the answers that carried your link and got no click. The first counts impressions in Google's AI surfaces; the second counts citations in Microsoft's answers.

• The generative AI report in Google Search Console, announced by Google on 3 June 2026 and available to every site since 31 August 2026, counts impressions of your site's links in AI Overviews, AI Mode and Discover, by page, country, device and date. It does not count clicks, and Google says these surfaces are also counted within the Web search type in the overall Performance report. The short definition is in what is the generative AI report?
• Bing's AI Performance report counts how often your site is cited in Copilot answers and Bing's AI summaries, which pages are cited, and a sample of the phrases the AI searched with. The detail is in how does Copilot pick sources in Bing, and cite you?

We use our own site, tasadr.ai, as the example, because it is new to Google Search and its numbers are small. Small numbers sometimes teach more than big ones: they show you when a report starts counting, and from which page.

Because a visit starts with the reader's decision, not the engine's. An engine can cite you in a hundred answers and nobody clicks; ten visits can come from one answer a reader passed on to colleagues. The number of visits does not tell you how many questions cite you, or who leads the questions you are absent from.

And when visits rise after you edit a page, you hold three things and no fourth: timing, meaning both happened in the same period; comparison, meaning the page you edited moved more than pages you left alone; and estimate, meaning a rough figure for the value it brought. That the edit was the cause, none of the three reports can prove, because the engine and your competitor can change in the same month.

That is why at Tasadr (تصدَّر) we measure the side these reports cannot see: the answer itself. We ask the engines your customers' questions and record whether your site was cited, mentioned without a link or absent, and keep the excerpt of the answer that names you or your competitor. What we work out from that is explained in how we measure the Tasadr score.

Five small habits protect you from common mistakes.

• Compare like with like: month with month, or a month with the same month last year when seasons are involved, since Ramadan or the back-to-school season can shift all your visits, not only those from engines.
• Read the landing page before the total: a hundred visits to an old article are not ten visits to the service page your enquiries come from.
• Ask each new customer how they found you, and add "an AI engine" as an option on your contact form: some readers who start in an answer arrive directly or through search, and no report knows it.
• Never merge different measures into one number: impressions in Google Search Console are not visits, and citations in Bing are not clicks. Each measures something else.
• Write down the date of every edit, so you read the numbers after it knowing what changed on your site, and what changed in the market.

They cannot tell you how many people asked an engine your question, because the engines do not publish it. They cannot tell you what an engine wrote about you, or whom it recommended instead. And Google Analytics 4 does not separate clicks from AI Overviews and AI Mode from the rest of Google's results.

What this guide says about Google Analytics 4 comes from Google's help pages as we read them on 17 October 2026. Google changes its channels and their names from time to time; if anything changes, we correct it with a date. You can start measuring the answers themselves with the 7-day free trial, with no credit card, on 10 questions in ChatGPT and Gemini.