Google's AI answers are taking clicks from informational pages, not from the local map

What the click data actually shows

In March 2025 the Pew Research Center watched what 900 American adults did on Google. The panel members had agreed to let an app record every page they visited, which produced a dataset of 68,879 unique searches and a record of what each person did next ^[1]. About 18 per cent of those searches returned an AI summary ^[1]. On the results pages where a summary appeared, a standard result link was clicked on 8 per cent of visits ^[1]. On the pages without one, the figure was 15 per cent ^[1].

Two further numbers from the same dataset matter more than the headline. Users clicked a link inside the AI summary itself on only 1 per cent of visits to pages that had one ^[1]. They were also more likely to stop browsing altogether after a page with a summary, on 26 per cent of visits against 16 per cent of visits to pages without one ^[1]. The summary did not move the click somewhere else. It removed the occasion for one.

The largest keyword level study points the same way. Ahrefs compared 300,000 keywords in aggregated Search Console data, half of which carried an AI Overview in December 2025 and half of which were informational keywords without one, and measured average desktop click-through rate against the same month in 2023 ^[2]. For the top-ranking page on a keyword that had gained an AI Overview, the measured rate fell from 0.073 to 0.016, where the forecast without an AI Overview was 0.037 ^[2]. The authors put the reduction at about 58 per cent for the first position, easing to roughly 19 per cent by the tenth ^[2].

Both studies describe informational searching. Pew's panel ran general Google searches, and Ahrefs built its comparison set explicitly from informational intent keywords ^[1][2]. Neither was designed to answer the question a plumber, a solicitor or a clinic owner is actually asking, which is whether the searches that produce enquiries have changed.

Why Google's account and the independent studies do not match

Google's position was set out by Liz Reid, its vice president and head of Search, in August 2025, and it is a claim about the aggregate ^[3].

Overall, total organic click volume from Google Search to websites has been relatively stable year-over-year.

The same post argues that average click quality has risen, defining a quality click as one where users do not quickly click back, and that people are searching more and asking longer and more complex questions than before ^[3].

The reaction from people who read Search Console data for a living was blunt. Laurence O'Toole of Authoritas said it was about time Google shared data but noted that no figures were given. David Buttle, previously platform strategy director at the Financial Times, said the post contained "no data whatsoever". Glenn Gabe urged Google to "stop the BS" and acknowledge that AI Overviews and AI Mode are reducing traffic for many sites ^[4].

Google's own trading does not suggest a business under strain. Search and other revenues reached $63.27 billion in the second quarter of 2026, up 17 per cent on the year, although that was the first slowdown after four quarters of acceleration ^[5]. Sundar Pichai attributed query growth to the company's AI features ^[5]. Google has also said it handles more than 5 trillion searches a year, and that in an experiment run in the United States and India, queries that showed AI Overviews produced an increase in Google usage of more than 10 per cent ^[6].

These accounts are less incompatible than the argument around them suggests. A stable total is consistent with a sharp redistribution underneath it, and Google's post concedes that some sites are receiving less traffic while others gain ^[3]. The real disagreement is about who sits in which group, and Google has published no figures that would let anyone outside the company check ^[4].

What has happened in the United Kingdom

The clearest British measurement comes from the Association of Online Publishers, whose impact study analysed 10.8 billion pageviews across eight UK publishing groups ^[7]. Organic Google referrals to the participants fell 7.1 per cent between the first and second quarters of 2025 ^[7]. That average hides a split, with news publishers losing about 5 per cent in the quarter and consumer and business titles 18 per cent ^[7]. If the trajectory holds, the study expects Google pageviews to those publishers to be cut in half by the third quarter of 2027 ^[7]. The analysis was written up by the AOP's managing director, and premium publishers are not representative of smaller sites ^[7].

Wider figures are worse. Chartbeat data reported in the Reuters Institute's 2026 Digital News Report found Google organic search traffic to more than 2,500 sites down 33 per cent globally between November 2024 and November 2025 ^[8].

What is striking is how little of that the British public's own use of AI assistants would explain. The same report puts weekly use of AI chatbots for news in the United Kingdom at 4 per cent, the lowest of its 48 markets, with trust in chatbot answers at 6 per cent, also the lowest ^[8]. Ofcom's news research, published in September 2026, found a larger figure on a looser measure, with 21 per cent of UK adults having used an AI app or service for news in the past month and ChatGPT the most common at 13 per cent ^[9]. Ofcom also found that 56 per cent of adults are less trusting of a news story written by AI than one written by a person, and it notes that this year's survey used an updated methodology ^[9].

A monthly measure and a weekly one will always differ, and a question about news is not a question about choosing a dentist. The defensible reading is narrower than either headline. Assistant use in Britain is growing from a small base and trusted by few, while search referrals are falling faster than that growth on its own would account for. The pattern that fits both facts is the one Pew measured, which is fewer clicks on the results pages that carry an answer ^[1].

The searches where AI answers appear, and the searches where they do not

The most useful study for a local business is also the smallest. In May 2025 Whitespark ran 540 searches by hand across three American cities and six trades, those being plumbers, personal injury lawyers, dentists, optometrists, medical clinics and real estate agents, and sorted every query by intent ^[10].

Across the whole set, AI Overviews appeared on an average of 68 per cent of queries ^[10]. Split by intent, the picture changes completely. On local intent queries, the sort that name a trade and a town, an AI Overview appeared 15 per cent of the time while the local pack appeared 93 per cent of the time ^[10]. On informational queries the position reversed, with AI Overviews on 92 per cent and the local pack on 6 per cent ^[10]. Hybrid queries, which attach a question to a place, drew an AI Overview 97 per cent of the time ^[10].

This is a small manual sample from three American cities, collected before several rounds of change to both surfaces, and its author presented the numbers as initial takeaways from a limited dataset ^[10]. No equivalent British study has been published. It is best treated as strong evidence of a pattern rather than a measurement of today's prevalence.

The pattern is at least coherent with the way Google describes the two surfaces. Google says local results are based mainly on relevance, distance and prominence, and defines prominence as how well known a business is ^[21]. That is a different ranking problem from the one a generated answer solves, and it draws on a different input, the Business Profile rather than the page. A business whose enquiries come from queries naming its trade and its town has been reading headlines about a surface that mostly answers a different kind of question.

What AI Mode changes about the way people ask

The displacement story is about AI Overviews, which sit on a conventional results page. AI Mode is a different product and the more consequential one for a local business. Google says it has passed a billion monthly active users globally, that its queries have more than doubled every quarter since launch, and that the average AI Mode search is three times the length of a traditional query ^[11]. Planning queries grew 80 per cent faster than AI Mode queries overall in the six months to May 2026, and searches beginning with "where to", "where should I" and "ideas for" are growing ^[11].

In the United Kingdom that has already become a transaction rather than an answer. Since April 2026, AI Mode in Search has been able to take a request such as a table for two at a dog-friendly Italian restaurant in Shoreditch and carry it through to a booking with partners including TheFork, ResDiary, OpenTable and Foodhub ^[12].

The feature with the sharpest implication is the least discussed. Google documents that when a user asks about local availability or pricing and "businesses have limited information available online", it will telephone the business on that user's behalf ^[13]. The service runs in English for signed-in users over 18, a user can type "don't call" to restrict the answer to online sources, and Google warns that AI responses may include mistakes ^[13].

That is best read as a statement about incentives rather than a product announcement. Where the price, the availability and the service list are published, a system composing an answer can use them. Where they are not, the business is relying on whoever answers the phone, or on a summary of a page that does not contain the answer. This is judgement rather than measurement, but it points the same way as Google's published advice on its generative features, which is that the useful work is unique, specific, crawlable information and not special formatting ^[20].

Whether anyone is actually choosing a local business through an assistant

The best available evidence is a survey, and an American one. BrightLocal asked 1,002 US adults in March 2026 how they find local businesses, and 45 per cent said they had used an AI tool for a local business recommendation in the previous year, against 6 per cent the year before ^[14]. Among those who had, ChatGPT was the most used at 31 per cent and Google's AI Mode second at 23 per cent ^[14].

The qualifying detail is that very few people take the answer at face value. BrightLocal reports that 88 per cent of its AI users check the sources behind a recommendation, and that 97 per cent at least sometimes verify a recommendation against reviews on a review platform ^[14]. Use of Google reviews for local recommendations still fell over the year, from 83 per cent to 71 per cent ^[14].

It is a self-reported survey from a company that sells local search tools, and the gap between stated and observed behaviour is usually wide. Set against the British figures on assistant use for news, which are the lowest in the Reuters Institute's set of markets ^[8], a UK owner should assume that assistant led discovery here sits behind the American figure and is growing. It is also a reason to treat reviews and profile accuracy as one project rather than two.

What the regulator has decided, and what it does not settle

Britain is now the jurisdiction where this is being regulated most directly. On 10 October 2025 the Competition and Markets Authority designated Google as having strategic market status in general search and search advertising, finding substantial and entrenched market power and noting that more than 90 per cent of searches in the UK take place on Google's platform ^[15]. AI Overviews and AI Mode fall inside the designation, while the standalone Gemini assistant does not, although the CMA said it would keep that under review ^[15].

On 3 June 2026 the authority imposed a conduct requirement on Google's search services. Google must let publishers opt out of having their content used to power AI features such as AI Overviews, must let them opt out of its use for fine-tuning models, must attribute publisher content properly using clear links in AI generated results, and must submit and publish compliance reports, with nine months to implement all the changes ^[16].

Today, we have introduced a world-first requirement on Google's search services in the UK.

Google announced the matching control on the same day: a Search Console toggle that decides whether a site appears in generative AI Search features, tested first with a subset of UK website owners before a worldwide rollout ^[17]. Its wording is the part worth reading twice.

Sites that opt out will not receive traffic or impressions from our generative AI features.

Google adds that the choice will not be used as a ranking signal for search results outside those generative features ^[17]. For a publisher with a licensing strategy, that toggle is bargaining power, which is what the CMA said it was for ^[16]. For a local business it is a way of removing itself from a surface where it currently appears at no cost, and there is no evident reason to use it.

What a business can now measure, and what it cannot

Measurement improved sharply during 2026. Search Console now carries a generative AI performance report covering AI Overviews and AI Mode, which shows how many times links to a site were shown in those features, which pages appear, and the breakdown by device and country ^[18]. Google completed the rollout to all websites worldwide on 31 August 2026 ^[18]. Search Labs experiments are excluded, the most recent days are preliminary and may change within hours, and the report covers the Web search type only ^[18].

Two limits matter. The report counts impressions, which are not visits, and Google says it will introduce additional metrics over time rather than presenting the current set as complete ^[17]. It is also a report about Google Search. Clicks arriving from AI features are folded into the ordinary Search performance report under the Web search type ^[19], so they cannot be separated out there, while a visit from ChatGPT, Gemini or Perplexity never enters Search Console at all and has to be found as referral traffic in the site's own analytics. That last point is inference, not a documented statement.

For a business with a few hundred enquiries a year, none of this is the primary instrument. Calls, forms, booking clicks and the source attached to each of them are the measurement that decides whether the work paid for itself, and that measurement sits in the site's own analytics and its phone records rather than in a platform report.

What it means for a local business

The practical conclusion is less dramatic than the headlines on either side, and it redirects effort rather than adding to it.

1. Keep the Business Profile complete and accurate, because the surface that answers local intent searches is the map. Google states that businesses with complete and accurate information are more likely to show up in local search results, and that there is no way to request or pay for a better local ranking ^[21].
2. Publish the facts that a generated answer would otherwise have to ask for. Google documents calling a business for things like the earliest appointment or current stock when that information is not available online ^[13], and the same logic covers prices, areas served and who is qualified to do what.
3. Judge informational pages on a different basis. The evidence that they lose clicks where an AI answer appears is now strong ^[1][2], so a guide earns its place by convincing the person who reads it, or by being the material an answer is built from, rather than by its session count.
4. Do not build AI specific plumbing. Google says Search ignores llms.txt files, that no new machine readable files, markup or Markdown are needed, and that structured data is not required for its generative features, though it remains worth having for rich results ^[20].
5. Do not opt out of generative AI features without a licensing reason, because Google's own wording is that opting out removes traffic and impressions with it ^[17].
6. Measure enquiries by source rather than sessions, and treat generative AI impressions as an indicator of visibility rather than as an outcome ^[18].

Nothing in the published evidence supports a promise about placement in an AI answer, and Google's own documentation states that indexing and serving are not guaranteed even for pages that follow its guidance ^[20].

What remains uncertain

The direction is clear and the magnitude is not, particularly for the searches a local business depends on.

• No published study isolates click-through rate for local commercial queries, in the UK or anywhere else. The local prevalence figures come from 540 manual American searches in 2025 ^[10], and the click-through figures from informational keyword sets ^[1][2].
• The Ahrefs result is a correlation measured against a forecast of what click-through would have been, on desktop only, and its authors say as much ^[2].
• Google has published no absolute click figures that would allow its claim of stability to be tested, which is the substance of the criticism it received ^[3][4].
• The two British surveys of AI use for news differ by a wide margin because they measure different periods in different ways ^[8][9], so neither is a safe basis for sizing assistant led discovery.
• The CMA's requirements run over nine months, and their effect on attribution, on links and on traffic is not yet observable ^[16].
• Whether an appearance in a generated answer produces an enquiry at all is still unmeasured. Impressions in the new Search Console report are the first step towards knowing ^[18], and that is the step before the one that matters.

A fair summary is that the informational half of search is being absorbed into the answer, while the local half still runs through the map and the profile, and nobody yet has the data to say how long that division will hold. The response that survives either outcome is the same. Make the business easy to understand, easy to verify and easy to contact, and count the enquiries rather than the impressions.