Every few weeks somebody asks how to see AI Overviews traffic in their analytics. The answer is that you cannot, and it is worth understanding exactly why, because the reason also tells you what you can infer and stops you buying a tool that claims otherwise.
The mechanics
When Google shows an AI Overview above the results and a person clicks a citation inside it, their browser sends a normal request to your server with a normal referrer: google.com.
An ordinary organic click on a blue link sends the identical thing.
There is no parameter, no distinguishing header, no fragment, nothing in the request that differs. The two events are byte-for-byte indistinguishable by the time they reach you.
The same applies to Bing's inline Copilot answers. This is different from standalone assistants like ChatGPT and Perplexity, which have their own hostnames and are perfectly measurable — the problem is specific to AI answers rendered inside a search results page.
Why no tool can fix it
This is worth being blunt about, because products are being sold on the claim.
The information is not obscured, or hard to extract, or behind an API. It was never transmitted. No amount of engineering recovers a signal that does not exist in the request. Any product showing you an "AI Overviews traffic" figure is doing one of three things:
- estimating from Search Console impression patterns and telling you it is a measurement;
- scraping search results for your keywords to see whether an Overview appears, then attributing a share of clicks by assumption;
- counting standalone assistant referrals and labelling them AI Overviews.
The second is a legitimate technique when it is labelled as modelling. The other two are not.
What you can actually see
Search Console, read carefully
Google does not break out Overview impressions. What you can watch is the shape of the relationship between impressions and clicks over time.
The signature of an Overview appearing for a query you rank in is: impressions steady or rising, clicks falling, average position stable. That combination is unusual otherwise. Position dropping explains clicks falling; here position is not dropping.
Track it per query group rather than site-wide, because it happens unevenly — informational queries get Overviews far more than transactional ones.
The query types most affected
Definitional and how-to queries — "what is X", "how do I Y" — get Overviews most and lose the most clicks, because the answer genuinely is complete in the box. Comparative, transactional and navigational queries are much less affected.
If your traffic is concentrated in the first group, your exposure is high and worth planning around. That is a content strategy fact you can act on without measuring a single Overview.
The honest empty state
If your analytics tool has an AI section, it should say plainly that Overviews are not in it. A dashboard showing "AI traffic: 1.2%" without that caveat invites the conclusion that AI is a rounding error, when the largest AI surface is simply missing from the number.
We say this in the product's own empty state, because the alternative is a number that quietly misleads the person reading it.
What to do instead of measuring it
Assume it is happening on informational queries and check whether those pages still earn their place. A page whose whole value was answering a question that is now answered in the results page needs a different reason to exist.
Measure the surfaces you can. Standalone assistants are fully measurable and are the fastest-growing part. So is crawler activity, which tells you what is being read regardless of what is cited.
Write things an Overview cannot replace. Original data, opinion, specific experience, a method with your name on it. The pages most damaged by Overviews are the ones that were competent summaries of general knowledge — which is also the kind of page that gets crawled and never cited. Those two problems have the same solution.
Common questions
Can Google Search Console show AI Overviews separately?
No. Search Console does not break out Overview impressions or clicks as their own dimension. The nearest usable signal is the pattern of steady impressions, falling clicks and stable average position for a query group, which is unusual for any other reason. It is an inference, not a measurement, and should be described that way.
Are tools that claim to measure AI Overviews traffic lying?
Not necessarily, but they are modelling rather than measuring, and the good ones say so. A legitimate approach is to check search results for your keywords, see where an Overview appears, and estimate the click impact — that is real research. What is not legitimate is presenting the output as observed traffic, because the underlying click carries no signal that distinguishes it from any other organic click.
Which pages lose the most traffic to AI Overviews?
Pages answering definitional and how-to queries — the ones where the Overview can be complete and correct in a paragraph. Comparative, transactional and navigational queries are much less affected, because the person needs to reach a specific destination. If your organic traffic is concentrated in short informational answers, your exposure is high regardless of what you can measure.