Look for stable impressions against falling clicks on queries that read as questions. Then search those queries yourself and see what sits above the results. There is no report that names AI Overviews as the cause — Search Console counts an impression whether or not anyone scrolled past the generated answer.
Why this needs a manual check
Search Console doesn't segment AI Overviews. If your page is referenced inside a generated answer, or ranks beneath one, you get an impression the same as always. The click-through rate falls; the reason doesn't appear anywhere.
So the diagnosis is two steps: find the pattern in the data, then confirm it by looking at the actual search results.
Step one: find the shape
Open Search Console → Performance. Compare the last three months against the same period last year, with clicks and impressions both displayed.
Then go to the Queries tab and sort by impressions. Add a mental filter: which of these are questions? "How long does probate take." "What is a fiduciary." "When are quarterly taxes due."
Informational queries are where generated answers appear most. Commercial queries — "estate attorney charleston" — are far less affected, because a generated paragraph is a poor substitute for a list of local firms.
If your click-through rate has fallen specifically on the question-shaped queries while holding on the commercial ones, that split is the strongest signal you'll get from your own data.
Step two: look at the actual result
Take your five highest-impression question queries and search each one. Logged out, ideally on a phone, since layout differs.
Record for each:
- Is there a generated answer at the top?
- Does it cite you?
- How far down is the first blue link?
- Does the answer fully resolve the question, or does it leave a reason to click?
That last one is the commercially meaningful distinction. An answer that fully resolves "when are quarterly taxes due" ends the session. An answer that summarises "how long does probate take" and gestures at complexity leaves the reader wanting a specialist.
What this doesn't mean
Two over-corrections are common, and both are expensive.
"Informational content is dead." It isn't. Being cited inside a generated answer is a visibility position — often in front of someone who'd never have scrolled to position four. The click economics changed; the value of being the source didn't disappear.
"We should stop answering questions." The opposite usually applies. The content most likely to be cited is the content that answers clearly and early. Retreating from question-shaped content removes you from the surface where the answer gets assembled.
The adjustment worth making is to the expectation: informational content increasingly earns citation and awareness rather than sessions. If your entire measurement framework is sessions, it will report a problem that is partly a change in what the channel does.
What you can and can't conclude
Directional — That clicks fell against stable impressions on question queries, and that generated answers appear for those queries. Both are directly observable.
Unconfirmed — That the AI Overview caused the decline. Results are personalised, vary by device and location, and change without notice. What you see today isn't necessarily what your prospects saw last quarter. Correlation between the pattern and the presence of generated answers is reasonable evidence; it is not proof, and anyone quoting you a percentage of clicks lost to AI Overviews for your specific site is extrapolating from aggregate studies.
Pending — What it's costing. That needs the click data joined to conversion data over the same window, and most firms don't have the second half at a query level.
The record worth keeping
Screenshot the results for your five queries, with the date. Do it again quarterly. Generated answers change constantly, and a written record is the only way to tell a trend from a bad day.
Clicks disappearing while rankings hold is the first symptom in the full map at how do I tell if my website is losing me clients.