The impact of AI Overviews on SEO is the gap that has opened between the number of times a site is shown and the number of times it is clicked. Google answers above the results, the user reads, the visit never happens. Organic positions stay where they are, traffic does not.
Since 31 August 2026 that gap has its own report in Search Console. It shows half the story, and the less useful half. How to read it alongside your other sources is covered in the guide to SEO analytics.
Search Console now counts AI impressions. Not clicks
Google announced the Generative AI reports on 3 June 2026, starting with the United Kingdom, and completed the rollout at the end of August. The documentation puts it like this: "As of August 31, 2026, we've rolled out these insights to all websites worldwide".
The report contains your pages' impressions in AI Overviews and AI Mode, broken down by page, country, device and date. The numbers come from the "Web" search type in the Performance report, so they are not extra traffic: they were already in the totals, mixed in with everything else.
Clicks, CTR, position and queries are left out. The CMA, the UK competition regulator that imposed the new obligations on Google, explicitly asks in its interpretative notes for click-throughs and click-through rate separated from organic results. They are not there. The controls for excluding content from AI features currently work only at domain level, and page-level controls are expected by March 2027.
In practice: you know how many times you appeared inside a generated answer, but not whether it did you any good. If you are setting up tracking now, the starting point is still the Search Console Performance report, with the new AI panel alongside it.
How much is really lost, and where it was measured
Most articles on the subject quote the same two numbers, Pew's and Ahrefs', without saying that both come from US samples. The difference matters, because AI Overviews reached the UK in August 2024, about three months after the US launch, and the mix of queries is not the same. Here are the main studies, with the market each one measured.
| Source | Sample | Period | Market | Result |
|---|---|---|---|---|
| Pew Research Center | 900 adults, 68,879 searches | March 2025 | United States | 8% clicked a link with an AI summary, 15% without. 1% clicked inside the summary |
| Ahrefs | 300,000 informational keywords | March 2024 vs March 2025 | United States, desktop | Position one CTR 34.5% lower |
| Amsive | 700,000 keywords, 10 sites | 2024-2025 | United States | Average CTR -15.49%, non-brand -19.98%, brand +18.68% |
| Seer Interactive | about 10,000 informational keywords | January 2024 - January 2025 | United States | Organic CTR from 1.41% to 0.64% where an AI Overview appears |
| Sistrix | 51 million keywords | September 2024 - March 2025 | United Kingdom | AI Overview on over 18% of keywords, up from 4%; featured snippets down from about 10% to about 5% |
| Studio 36 Digital | 1,000 queries in 8 sectors, 100+ monthly searches each | April 2024 - October 2025 | United Kingdom | AI Overview on roughly 42% of searches |
One Pew figure that hardly anyone quotes: with an AI summary on the page, 26% of sessions end there, against 16% when there is no summary. It is not only the click that goes missing, it is the browsing that stops. It is the same dynamic that drives zero-click searches, which existed long before AI Overviews.
The two UK numbers do not match, and the reason matters
18% and 42% look contradictory. They measure different things. Sistrix looks at coverage across a database of 51 million keywords, most of them long-tail and low-volume. Studio 36 Digital looks at 1,000 queries with at least 100 monthly searches, picked across eight sectors, over a later and longer period. Before you put a percentage on a slide, it is worth knowing which of the two you are quoting.
On sectors the sources agree on one thing: health is where it breaks. In a Sistrix analysis of UK news publishers published by Press Gazette in May 2026 (January to April 2026 data), health keywords showed an AI Overview 72% of the time, tech 49%, travel 39%, money 35%, football 5%. Studio 36 Digital also puts health and wellbeing at the top, with finance and local services almost untouched. In the March 2025 Sistrix data the retail tracker stood at 14.98%.
The number that should worry you most comes from Italy, from a SEOZoom study of over 30 million keywords published in December 2025: of 168 million visible domains, Italian AI Overviews select about 175,000, or 0.1%. The top sources are YouTube and Wikipedia, followed by health publishers. And about 40% of the links cited do not appear in the top ten organic results, which undermines the idea that ranking first is enough to be cited. We have covered how the selection works in a separate guide to AI Overviews ranking factors.
Ten UK B2B queries, checked one by one
All the available studies cover health, recipes, travel, retail, news. For industrial B2B, which is the market of many of our clients, there is no public measurement. We ran a small one: ten real queries, Google UK on desktop, run on 26 September 2026 through the DataForSEO SERP API. It is a spot check, not a study, and should be read as such.
| Query | Search type | AI Overview |
|---|---|---|
| what is cnc turning | definition | Yes, at the top of the page |
| difference between milling and turning | comparison | Yes, at the top of the page |
| what is predictive maintenance | definition | Yes, at the top of the page |
| how to choose a metal fabrication supplier | supplier evaluation | Yes, at the top of the page |
| laser cutting quote online | transactional | Yes, but at the bottom of the page, below nine organic results |
| subcontract cnc turning | supplier search | Yes, at the top of the page |
| metal fabrication services | supplier search | No, the page opens with a local pack |
| plastic injection moulding services | supplier search | Yes, below the local pack |
| laser cutting sheet metal price | price | Yes, at the top, above the product carousels |
| b2b seo agency | supplier search | Yes, at the top, with a list of named agencies |
Nine out of ten. When we ran the equivalent Italian queries on Google Italy on 11 September 2026, the result was five out of ten, and the line fell neatly between questions (all with an AI Overview) and supplier or price searches (none). On Google UK that line has gone. The only query without a generated answer is "metal fabrication services", where Google leads with a local pack. What changes on supplier searches is where the answer sits and what it does. On "laser cutting quote online" it is at the very bottom, under nine instant-quote services. On "plastic injection moulding services" it comes after the three businesses in the map. On "b2b seo agency" it sits at the top and names four London agencies: the answer has become a shortlist. That is why search intent analysis matters more than before, not less: the same intent can now be served by a list of links, a map or a summary that picks the suppliers for the buyer.
Two details you only see by opening the summaries. For "difference between milling and turning" the AI Overview cites Standish Engineering and Tarvin Precision, two UK engineering firms, alongside JLCCNC, BFW, Admati, three YouTube videos and a Facebook video. Standish Engineering was also the first organic result, so in that case the citation came on top of the ranking rather than replacing it. For "what is predictive maintenance", on the other hand, the summary draws mostly on software vendors: IBM, SAP, Tractian and Sensorfact, plus Wikipedia and two YouTube videos. The only engineering body in the list is TWI.
The takeaway is simple, and for a manufacturer it is good news. Where a topic has a global software vendor that has written a page about it, Google takes that page. Where the topic is a machining process, it takes the sites of the people who do the work. In manufacturing this means that a technical page written by someone who works on the shop floor still has a structural advantage, as we saw working on SEO for manufacturing, where the Italian LED wall maker Macropix went from position 88 to 2 for "ledwall" on Google Italy and now holds 25.55% share of voice on the cluster, ahead of Amazon at 18.08% (Semrush, July 2026).
Why the chart goes up and down at the same time
The typical shape of a Search Console chart over the last two years is impressions rising and clicks falling. It is not a measurement error, it is how Google counts. The documentation says that "an AI Overview occupies a single position in search results, and all links in the AI Overview are assigned that same position", and that the standard rules apply to impressions.
So if one of your pages is cited in the summary, it collects an impression at an average position that is often better than the organic one it had. The click, on the other hand, depends on how complete the answer already is. Pew measured 1% of clicks on links inside the summary. Google, for its part, continues to say that no special optimisation is needed: "There are no additional requirements to appear in AI Overviews or AI Mode".
How to tell whether the drop is caused by AI Overviews
In most of the cases we see, a drop blamed on AI is partly down to something else. This is the sequence we use to separate the causes.
- Look at average position, impressions and clicks over the same period. Stable position, stable or rising impressions, falling clicks: it is a CTR problem, and AI Overviews are the prime suspect. Worse position: the problem is ranking, not AI.
- Segment the queries. Put the ones starting with what is, how, difference, why on one side, and the ones with price, supplier, subcontract or your brand name on the other. If the drop is concentrated in the first group, the diagnosis holds.
- Open the Generative AI report and sort pages by impressions. The ones at the top are the most exposed. Cross-check them with the pages losing clicks.
- Rule out everything else. A core update in the same period, seasonality compared year on year, a migration, a drop in indexing. Any one of these can explain half the curve on its own.
- Change the unit of measurement. On a B2B site the useful question is not how many visits the page has lost, it is how many quote requests it generates now compared with a year ago. We have collected the SEO KPIs that still make sense.
Step 2 is the one that almost always gets skipped, and it is the one that changes the conclusion. A site that loses 30% of clicks on dictionary queries and none on quote queries does not have a traffic problem. It has an internal reporting problem.
The opt-out switch is a shot in the dark
Since June 2026 Search Console has had a setting that excludes a domain's content from AI Overviews, AI Mode and AI Overviews in Discover. Google states that using it does not affect rankings in standard results, and the CMA has made both that option and the obligation to attribute sources with a link binding.
The problem is the order in which things arrived. The switch exists, the data to decide whether to use it does not: without clicks you do not know how much traffic you are getting from the citations you would be switching off. It works only at domain level, so it is all or nothing, and page-level control will not arrive before March 2027. For a site that sells, switching it off now also means giving up brand visibility inside the answers, which is the part that is becoming most valuable. If you are thinking about how to work with it rather than how to get out, the starting point is SEO for AI search, and for the mechanics the guides on what AI Overviews are and on AI Mode are still useful.
What is being lost is not traffic
There is a consequence that gets little attention, and it concerns exactly the pages that seem least important. Dictionary queries brought visits that rarely turned into customers, which is why so many people are writing them off without much regret. But they were where a company got itself known before anyone needed anything.
Amsive measured something interesting: on brand queries CTR went up by 18.68%. Brand searches are holding. But brand searches come from somewhere, and in B2B they often come from that first technical piece someone read six months before they had a need. If AI answers in your place for two years, the pool those queries come from shrinks quietly, and you will see it in the numbers when it is too late to act. The defence is not writing more definitions. It is putting into your pages what a summary cannot condense: your own data, a measurement taken on the shop floor, a case with names and dates. It is the same principle that applies to B2B SEO and that you can see in the projects we work on.
Frequently asked questions
No. An AI Overview takes its own position above the results and organic rankings stay where they are. What changes is click-through rate: Ahrefs measured a 34.5% lower CTR in position one, on a US sample of 300,000 informational keywords.
No. The Generative AI report in Search Console, available for all sites since 31 August 2026, shows only impressions broken down by page, country, device and date. Clicks are still counted within the Web search type, with no way of isolating them.
Questions, definitions and comparisons. Pew Research Center found an AI summary on 8% of one- or two-word searches and on 53% of searches of ten words or more, on a US sample from March 2025. In our check of ten UK B2B queries on 26 September 2026, nine had an AI Overview, including most supplier searches; the only one without was a supplier search where Google showed a local pack first.
The setting has been in Search Console since June 2026 and Google states that it does not lead to penalties in standard results. However, it works only at domain level and there is no click data to tell you how much traffic you would be giving up. Page-level controls are expected by March 2027.
No. According to a SEOZoom study of over 30 million Italian keywords, about 40% of the links cited do not appear in the traditional top ten organic results. In our UK check, the milling versus turning summary cited the first organic result but also sources outside the top ten.
