AI Overview Traffic Loss: What Search Console Can and Can't Tell You
Google now tells you which of your pages turn up inside AI Overviews. It won't tell you whether anybody clicked. The Generative AI performance report in Search Console carries impressions and nothing else, no clicks and no click-through rate, which means the one report built for this question can't answer the version of it that matters.
That gap is why the AI Overview conversation has gone circular. Everybody has the same two numbers, impressions holding and clicks falling, and everybody reads them as the same story. Sometimes they are. Often they're a ranking loss wearing a disguise, or a consent banner change, or plain seasonality, and the remedy for each one is different enough that guessing wrong costs you a quarter.
So this is the separation procedure: how to tell a real AI Overview loss from the three things that look identical to it, using reports you already have. We run website SEO for ecommerce brands, and this read is now the first thing we do on a new engagement rather than the third, because a wrong diagnosis sends a content budget somewhere it physically can't work. Book a call if you'd rather we ran it on your property.
What the Generative AI Report Actually Gives You
Google launched AI features reporting in Search Console on 3 June 2026 and finished rolling it out to all properties on 31 August 2026. It sits under Performance, with separate Search and Discover views, and the Search view covers AI Overviews and AI Mode together.
Two limits decide how you can use it. The report carries impressions only, so there's no click column and no CTR to calculate. And the clicks that do arrive from AI features aren't broken out anywhere: they're counted as ordinary clicks in the main Performance report under the Web search type, mixed in with everything else.
Taken together, the report answers "am I appearing in AI features, and on which pages" and refuses to answer "what is that costing me." If you can't see the report at all, Google's stated reason is usually too few impressions in AI features, which is itself a useful answer.
Treat the impression data as an exposure map rather than a loss statement. It tells you which pages are in the game. Everything past that has to be inferred, which is what the rest of this is for.
One more wrinkle worth knowing before you quote the report to anyone: AI Overviews and AI Mode are counted together in the Search view, so a page whose impressions are climbing there could be gaining ground in two quite different surfaces. That matters because AI Mode is a conversational surface where a brand owner is three questions deep, and an AI Overview sits on a conventional results page. The behaviour either side of those impressions isn't the same, and the report doesn't let you split them.
The Four Shapes a Traffic Decline Comes In
Before attributing anything, work out which of these four you're actually looking at. The columns are what you read off Search Console for the queries that lost clicks.
| Clicks | Impressions | Average position | Most likely cause |
|---|---|---|---|
| Down | Flat or up | Flat | Something on the results page is absorbing the click. An AI Overview is one candidate. So is a new ad block, an expanded local pack, or a competitor winning a rich result. |
| Down | Down | Down | An ordinary ranking loss. Nothing to do with AI Overviews, and the fix is the one you already know. |
| Down | Up | Down | You're ranking for more queries and worse ones. Usually follows a content or internal-link change on your side. |
| Down in GA4 much further than in Search Console | Flat | Flat | A measurement problem. Consent changes, tagging, bot filtering. Nothing on Google's side moved. |
Only the first row makes an AI Overview a candidate, and it's still just a candidate. An expanded ad block does precisely the same thing to your CTR at the same position.
The fourth row is the one that gets skipped, and across our client portfolio it catches more cases than anyone expects. Compare Search Console clicks against GA4 organic sessions over the same window before you blame Google for anything. If Search Console says down 12% and GA4 says down 40%, the extra 28 points are yours, and they're usually a consent or tagging change. Attributing ChatGPT and Perplexity traffic in GA4 covers the related problem of where those sessions get filed in the first place.
Which of Your Queries Are Even Exposed
This is the step that turns the argument into arithmetic, and hardly anybody runs it. AI Overviews don't appear evenly across query types. They cluster hard at one end, and the end they cluster at is not where most ecommerce revenue sits.
Seer Interactive's 2026 update, built on a multi-million-query dataset, puts AI Overview presence by query class roughly here:
| Query class | Example | AI Overview present |
|---|---|---|
| Comparison | agency vs in-house | About 95% |
| Informational | what is fba storage fee | About 36% |
| Commercial | amazon agency pricing | About 8% |
| Transactional | buy kids art supplies | About 5% |
Sort the queries that lost clicks into those four buckets. The decline is usually concentrated in one or two of them, and which ones tells you most of what you need. A brand whose lost clicks are mostly commercial and transactional does not have an AI Overview problem, whatever the aggregate chart suggests, because those results pages mostly don't carry one.
The honest caveat on that table, and it cuts both ways: the research behind these figures skews heavily toward informational and educational queries, because that's where the academic and vendor interest has been. Applying an informational CTR loss to a commercial query set overstates the damage, sometimes badly. Applying it to comparison queries probably understates it.
The Procedure, Step by Step
- Confirm the loss is real. Search Console clicks against GA4 organic sessions, same window, same filters. Settle any gap between them before you go further, because everything downstream assumes the decline happened on Google's side.
- Pick a clean comparison window. Year over year beats month over month on anything seasonal. Search Console retains 16 months, so you get exactly one shot at a yearly comparison and no second one.
- Export by query, not by page. Page-level data hides this entirely. A page total can sit flat while half its queries collapse and the other half grow.
- Build the candidate set. Queries where average position moved less than one place and clicks fell more than 20%. That's row one of the table above, isolated.
- Bucket the candidates by query shape using the four classes. This is the five minutes that does the most work.
- Cross-reference the Generative AI report. A page with real AI feature impressions, a flat position and falling clicks is as close to a confirmed case as Search Console will ever hand you.
- Check ten by hand. Search your top candidates in a clean browser session and note whether an Overview appears and who it cites. Repeat monthly, because Overviews come and go on the same query without warning.
- Write down the residual. Queries that fell with no AI feature impressions, no position change and no tracking gap are unexplained, and naming them as unexplained is far more useful than forcing them into the AI Overview column to make the report tidy.
Step seven is the one that gets dropped because it can't be automated, and it produces the clearest evidence in the whole exercise. Search Console will not flag which queries carried an Overview. Looking is still the measurement.
Why the Published Percentages Aren't Your Number
Four figures get quoted constantly and they disagree with each other. Ahrefs measured a 34.5% lower average CTR for the top-ranking page on keywords carrying an AI Overview, against comparable informational keywords without one, across roughly 300,000 keywords. Its 2026 update, run against a December 2023 baseline with a control set of keywords that had no Overview, reported something closer to 58%. Seer reported a 61% fall in organic CTR on Overview queries. Pew found people clicked a conventional result on 8% of searches carrying an AI summary against 15% of those without.
Those aren't really contradictions. They're different questions asked of different query sets against different baseline years. Which is the point: none of them is your number, and the spread between 34.5% and 61% is wider than most decisions you'd make on the back of it.
Watch for the oldest move in this literature too, which is comparing a pre-rollout year against now and handing the entire gap to one feature. Results pages got busier for everyone over that stretch. Some of the decline in every one of these datasets is the page changing shape in general, and no format change on your side recovers that part. We'd rather tell a brand that half of a decline is structural than sell them a rebuild aimed at all of it.
What to Do About the Part That's Real
For the queries that survive the procedure, the options are narrower than the volume of advice suggests.
Where an Overview answers the query completely, a title rewrite doesn't bring the click back, and accepting that early is cheaper than two quarters of testing it. Those pages still earn their keep. They hold topical authority and they can route a reader who does click toward something that converts. Treat them as internal-link assets with a click forecast of zero, and stop grading them on traffic they're structurally not going to get.
Where the decline sits in commercial and transactional queries, you have an ordinary CTR problem and the ordinary fix works, because those results pages mostly don't carry an Overview. If a click isn't happening at position six on a query where somebody is actively shopping, the title and the description are the likeliest reason and both are cheap to change. That's the most commonly misdiagnosed case we see, and it's the one where the work actually pays.
Where the exposure is structural, the answer is channel mix rather than copy. An organic result is rented. The OOly engagement compounded at more than 70% year over year for three consecutive years on a channel we ran end to end, and that came from controlling the inputs rather than hoping a results page kept its shape. If one rented channel carries most of your demand, the concentration is a bigger problem than the diagnosis.
If what you actually want is to be cited inside AI answers rather than to diagnose a loss, that's a different job with different mechanics, and it starts further upstream than content. Crawler access comes first on those surfaces, because a page an assistant can't fetch doesn't exist to it.
How Often to Re-Run This
Quarterly for the full procedure, monthly for step seven. Overview presence on a given query isn't stable, and we've watched the same query gain and lose one twice in a quarter. A diagnosis from March is a historical document by June, which is an argument for keeping the read cheap and repeatable rather than commissioning an audit once a year and treating it as settled.
Don't authorise a content rebuild off an aggregate chart. Run the four buckets, look at ten results pages, and name the residual honestly. The recoverable half is usually bigger than a brand has been told, and it's usually sitting in the commercial queries nobody was worried about. We'll run the read with you.
More on Website SEO
How to Get Traffic From Perplexity: Crawler Access First, Content Second
Most advice on getting traffic from Perplexity is advice about content. Content matters third. Two crawlers with two different rule sets, plus a CDN default nobody chose, decide whether you are eligible at all.
Attributing ChatGPT and Perplexity Traffic in GA4: What You Can See and What You Can't
Direct traffic grew and nobody ran a campaign. That is usually AI referrals arriving with no referrer. What GA4 can attribute to ChatGPT and Perplexity, what it quietly files as Direct, and how to tell the difference.
Ecommerce Content Strategy in 2026: The Topic Cluster Model That Builds Topical Authority
Most brands publish ten shallow blog clusters and wonder why nothing ranks. The brands winning organic build fewer, deeper topic clusters that earn topical authority and route link equity to money pages. Here's the content system that compounds.