Short answer: Visits from ChatGPT, Gemini or Copilot show up in GA4 under the AI Assistants channel. Google AI Overviews and AI Mode, by contrast, count as ordinary organic search and can only be surfaced in Search Console as impressions, never as clicks.
Why this is harder than it should be
When a customer tells you they found you through ChatGPT, that visit may not appear as such in your analytics at all. The reason is mundane: AI interfaces behave very differently under the hood. Some pass a clean referral, some pass none, and Google’s own AI answers are not a separate channel from an analytics point of view at all. They are part of ordinary search.
That creates an awkward situation. Traffic from AI search is growing, but the tools mid-sized companies use to measure their websites only capture part of it. If you do not know that, you draw the wrong conclusions and pull the wrong lever.
What you can see, and where
| Source | Visible in GA4 | Visible in Search Console | Limitation |
|---|---|---|---|
| ChatGPT, Gemini, Copilot, Deepseek, Grok | yes, default AI Assistants channel | no | only where a referral is passed |
| Perplexity, Claude | only via a custom channel group | no | not listed in Google’s default channel |
| Google AI Overviews | no, counts as organic search | yes, generative AI report | impressions only, no clicks |
| Google AI Mode | no, counts as organic search | yes, generative AI report | follow-up questions count as new queries |
| Classic Google search | yes, organic search | yes, Web search type | AI share cannot be separated out |
Step 1: set GA4 up properly
Google Analytics now has its own default channel for AI assistants. The official definition states that this channel captures users arriving from sources such as ChatGPT, Gemini, Deepseek, Copilot or Grok, and that Google’s own AI Overviews and AI Mode are explicitly excluded. Technically the channel triggers when the medium exactly matches ai-assistant.
For most companies that is already enough. You will find the numbers under Acquisition, in the user acquisition or traffic acquisition reports. If you also want Perplexity or Claude attributed cleanly, create a custom channel group under Admin, Data display, Channel groups. Google documents this in its custom channel groups guide, including an AI assistants example.
One practical note: the example expression printed there is drawn far too widely. Its first branch matches any source containing the letter sequence ai anywhere, including mail domains. Build a narrow list of specific hostnames instead and extend it when new services appear. Custom channel groups apply retroactively, so you lose no history.
ChatGPT makes this particularly easy. OpenAI states that referral URLs automatically carry the parameter utm_source=chatgpt.com, provided the OAI-SearchBot is not blocked in robots.txt. Blocking that bot costs you not only measurability but the chance of visibility itself.
Step 2: read Search Console correctly
For Google’s own AI answers there is no way around Search Console. The Google documentation on AI features says plainly that sites appearing in AI features such as AI Overviews and AI Mode are included in overall search traffic in Search Console, reported in the Performance report within the Web search type.
Since June 2026 there is an additional generative AI performance report. It shows impressions only, broken down by pages, countries, dates and devices. No queries, no clicks, no click-through rate. And according to Google it is being rolled out to a subset of properties first, so check whether it appears for you at all.
An expectation-setting point: separating AI clicks from classic search clicks is still not possible. Anyone promising you an exact click figure for AI Overviews is estimating.
Step 3: keep the gap in view
Some AI traffic inevitably lands under Direct. Google lists missing campaign parameters, redirects, shortened URLs and links from PDF or text documents among the causes of direct traffic. The industry also observes that links from mobile apps and in-app browsers frequently arrive without a referrer. There is no solid vendor source for that observation, so we treat it here as field experience rather than fact.
The workable answer is a control value: ask on your enquiry form, or in the first conversation, how somebody came across you. In a smaller company that single question often yields more insight than three dashboards.
Worked example: what a CTR drop actually means
A trade or service business ranks first for its most important search term and collects 6,000 impressions a month for it. The click-through rate used to be 28 percent, roughly 1,680 clicks.
Now an AI answer appears above the results. An Ahrefs analysis based on aggregated Search Console data across some 300,000 keywords finds a 58 percent lower click-through rate for position one. Applied to our example: 28 percent times 0.42 gives about 11.8 percent, or roughly 706 clicks. The loss is 974 clicks per month.
At an enquiry rate of two percent that is 33.6 versus 14.1 enquiries, so around 19 lost contacts a month. With an average order value of EUR 1,400 and a 20 percent close rate that corresponds to roughly EUR 5,400 in monthly order volume left on the table.
Read this deliberately as a scenario, not a forecast. The 58 percent figure comes from a third-party study by a tool vendor and describes an average across many websites. Google disputes the overall reading: the company states that total organic click volume from Search has stayed relatively stable year over year, while also noting that traffic redistributes between sites. Both can be true at once. For your business the message is simple: the average says nothing about your website. Measure your own.
What to do with the numbers
Measurement on its own changes nothing. It gets interesting when you put three things side by side: impressions in AI features, visits from AI assistants, and actual enquiries. If impressions rise but visits do not, you are being cited without being linked. If visits rise but enquiries do not, the wrong people are arriving, or your landing page does not answer the question they arrived with.
How to show up in AI answers in the first place is covered in GEO and AEO: how to get cited by AI answers. And if visibility is meant to turn into enquiries, AI in marketing covers the next step.
Frequently asked questions
Can I measure clicks from AI Overviews exactly?
No. Google counts them into the Web search type in the Performance report but does not break them out. The generative AI report shows impressions only. Everything else is an estimate.
Why do I barely see any ChatGPT visits in GA4?
Three common causes: the channel exists but you are looking at the wrong report; the OAI-SearchBot is blocked in robots.txt; or you are simply rarely cited. Check in that order.
What is the difference between AI Overviews and AI Mode?
AI Overviews are the summary answer blocks above the classic results, AI Mode is the conversational search surface. Google has merged the two experiences. For measurement nothing changes, both sit under Web and organic search respectively.
Should I block AI bots in robots.txt?
If you want to be found, usually not. What makes sense is a deliberate distinction between bots that serve answers and bots that collect training data. Make that decision consciously and write it down.
How often should I review this?
Monthly is enough. Weekly review mostly produces noise at small sample sizes. Consistency in how you define the channels matters more than frequency.
Conclusion
You can measure a good share of AI traffic today, but not all of it. Set up the AI Assistants channel in GA4, check the generative AI report in Search Console, add a source question to your enquiry form, and accept some residual blur. If you would like help with that, take a look at our SEO and AI search services, start with the AI check or get in touch. Further reading: the Pew Research analysis of click behaviour with AI summaries, the Ahrefs study on falling click-through rates and Google’s counter-position.
