How do I tell whether an AI Overview took my clicks or demand just changed?
Compare a fixed set of queries across two periods instead of looking at sitewide totals. If impressions held steady and clicks fell while your position stayed put, something above the results is absorbing the click. If impressions fell alongside clicks, demand moved. If position slipped too, it is an ordinary ranking problem wearing a new costume. The three patterns look identical in a traffic chart and require completely different responses.
Three patterns, three different problems
Almost every 'AI is killing our traffic' conversation resolves into one of three signatures once you pull query-level data.
- Impressions flat, position flat, clicks down. The result was shown as often as before and chosen less often. Summary absorption is the leading candidate.
- Impressions down, clicks down, position flat. Fewer people searched. That is a demand or seasonality story and has nothing to do with AI.
- Position down, impressions down, clicks down. You lost ranking. The AI layer is a bystander, and treating it as the cause will send you fixing the wrong thing for a quarter.
Do the comparison on a frozen query cohort
Sitewide averages hide all of this, because your query mix changes month to month and new queries enter constantly. Take the queries that existed in both periods, freeze that list, and compare only those. Match day-of-week counts too — a 28-day window against a 31-day window will invent a trend on its own.
This is the same discipline that makes any marketing reporting defensible: fix the population before you compare the metric.
Segment by intent before you conclude anything
Click absorption is not uniform. Informational queries — how something works, why a part fails, what a term means — lose clicks first, because a summary can genuinely finish the job. Transactional and local queries hold up better, because a person who wants someone at their house tomorrow still has to choose a company and contact them.
If your losses are concentrated in informational queries and your booked work is unchanged, you lost low-intent sessions. That is worth knowing, not worth panicking about.
What each diagnosis actually calls for
Summary absorption on informational queries: shift the measurement to conversions per session and downstream revenue rather than raw sessions, and make sure the pages that still earn clicks are the ones nearest the buying decision. Demand decline: check the calendar and the category, not your site. Ranking loss: run the ordinary technical and content diagnosis you would have run three years ago.
The expensive mistake is skipping the diagnosis. A rewrite aimed at AI visibility will not recover clicks that were lost to a competitor outranking you, and no amount of ranking work recovers a query nobody types anymore.
Topics: AI Overviews · diagnostics · search console · click loss
Have a version of this question about your own business?
The useful answer usually depends on which systems you run and how they're connected. That's a conversation, not a blog post.