THE SHORT ANSWER
An AI Overview pushes the first organic link down the page, and the measured effect depends almost entirely on query type. Informational queries lose the most, with credible studies reporting drops between roughly twenty and sixty percent. Transactional, local and branded queries hold up much better, because a summary cannot book a survey or produce a quote. The average across all queries is a meaningless number; the split by intent is what you should measure on your own data.
The question turns up in every board meeting now, usually phrased as a fear rather than a measurement. Google put a generated summary above the results, the first organic link moved down, and traffic went somewhere. What almost nobody does before panicking is separate the queries where an Overview actually appears from the queries where it does not, which is the only comparison that tells you anything useful.
Published estimates disagree wildly, and they should, because each one measures a different query mix. A sample weighted towards definitional questions will show a collapse. A sample weighted towards local and buying-stage questions will show almost nothing. Neither study is wrong about its own data, and neither transfers cleanly to your site. Treat the external range as an expectation and then go and measure yourself.
The numbers, at a glance
Informational queries: these absorb the largest losses, with reported declines commonly falling between twenty and sixty percent where an Overview is present
Transactional and local queries: far more resilient, because a generated paragraph cannot survey a roof, quote a heat pump or take a booking
Overview coverage by market: consistently lower in Dutch, French and Nordic results than in English ones, so European exposure lags the numbers published in the United States
The diagnostic signature: impressions flat while clicks fall is an Overview effect; impressions falling with clicks is an ordinary ranking loss
Why the published numbers disagree so violently
Google does not publish click-through data for queries where an Overview is shown, and Search Console folds Overview impressions into the ordinary web results report. Every public figure is therefore an inference drawn from one of three imperfect methods, each with a different bias baked in.
Clickstream panels. Third-party browser panels observe real sessions, but skew towards consumer devices and English-language browsing, which over-weights exactly the informational queries most exposed to summarisation.
Search Console cohort studies. A site compares its own queries before and after Overview rollout. Cleaner, but confounded by every algorithm update in the same window.
Controlled scraping. Researchers query a fixed keyword list and record whether an Overview appears. This measures coverage accurately and says nothing about clicks.
None of the three is dishonest. They simply answer different questions, which is why a headline claiming one universal percentage should be read as marketing rather than measurement.
The intent split that explains most of the variance
Sort your queries into four buckets and the picture stops being confusing. Exposure runs almost perfectly along the axis of how completely a paragraph can satisfy the searcher.
Definitional. What is a heat pump COP, how does a solar inverter work. Fully answerable in a paragraph, so the click was always fragile and is now largely gone.
Comparative. Air-to-water versus air-to-air, EPDM versus bitumen. Partly answerable, but readers still open sources to check the trade-offs against their own situation.
Costed. What does a flat roof cost per square metre. Summaries here are frequently stale or wrong, and readers who care about accuracy click through to check the date.
Transactional and local. Roofer near me, request a quote. Almost untouched, because the outcome the searcher wants is a human, a visit or a price.
For a contractor or a lead supplier, the commercially valuable queries sit in the bottom two buckets. That is the reason the trade press panic has not matched the revenue impact for most service businesses.
How to isolate the effect in your own Search Console data
Export queries at query level for two comparable periods, ideally the same eight weeks a year apart, so seasonality cancels. Then filter to queries whose average position moved by less than half a place. Anything that moved more is a ranking story, not an Overview story, and mixing the two is how teams reach the wrong conclusion.
In the stable-position set, look for the signature: impressions steady or rising, clicks down, average position unchanged. That pattern has very few other explanations. Score each affected query by intent bucket, and you will usually find the losses concentrate in one or two buckets rather than spreading evenly, which tells you exactly which content type is exposed.
One caveat worth stating plainly: this method cannot prove causation, because it cannot see which impressions carried an Overview. It narrows the field enough to make a decision, and in 2026 that is the best available instrument short of running your own scrape of the results pages you care about.
What is actually worth doing about the queries you lose
Stop defending definitional pages. A page whose only job was to define a term has lost its economic basis, and rewriting it longer will not bring the click back. Retire it into a section of a broader page, or keep it purely as a citation asset that feeds your brand into the Overview itself.
Then move the effort one step down the funnel. The searcher who read the definition still has the next question, and that next question is usually costed or comparative. Pages that answer the second question inherit the traffic the first question used to earn, and they convert better because the reader arrives already briefed.
Finally, treat being cited inside the Overview as a legitimate goal rather than a consolation prize. Overviews link out, those links appear above every organic result, and the click rate on a citation from a summary is respectable precisely because the reader chose to verify.
A four-week diagnosis using data you already have
Export query-level Search Console data for the same eight-week window this year and last year.
Discard every query whose average position shifted by more than half a place.
Tag the remainder as definitional, comparative, costed or transactional.
Calculate the click change per tag rather than one blended figure for the site.
Reallocate content effort away from whichever tag lost most and towards the tag directly below it in the funnel.
Want leads like this in your pipeline?
Flock runs the campaigns, screens the enquiries and hands you only the ones that match your service area, job size and capacity. You pay per lead, not per month.
Book a 15-minute fit check | See lead package pricing
Related answers
Frequently asked questions
Do AI Overviews appear on every query?
No. Coverage is heavily skewed towards informational and how-to phrasing, and it is deliberately restrained on health, finance and some legal topics. Coverage in Dutch, French, German and the Nordic languages remains lower than in English, so European sites typically see less exposure than United States benchmarks suggest.
Is the traffic loss permanent?
The loss on purely definitional queries should be treated as permanent, since nothing about that reader ever required a click. Losses on comparative and costed queries have partly recovered on sites that added dated figures and specific conditions, which summaries handle badly.
Can I opt out of AI Overviews?
Only by using the nosnippet or max-snippet directives, which also remove your ordinary rich snippet and usually cost more traffic than they save. Google-Extended controls training and Gemini grounding, not Overview inclusion, so it is the wrong lever for this problem.
Does being cited in an Overview send meaningful traffic?
Less than a top organic position used to, but the visitors are unusually well qualified because they clicked to verify a claim rather than to start researching. Several service businesses report higher form-completion rates from that path than from ordinary organic entry.
NEED A CLEARER PLAN?
Let’s turn your next move into momentum.
Talk to us →