The State of AI Search in the UK, 2026

Every few years someone publishes a chart that quietly resets how the internet works, and for most of 2025 and 2026 the charts have all pointed the same way. People are starting to ask AI assistants the questions they used to type into a search box, and the assistants are answering directly, naming a handful of businesses and moving on. The rest of the market never gets seen. That is not a prediction any more. It is showing up in third-party data from Similarweb, Pew Research, Ahrefs and academic researchers, and this article pulls that data into one honest picture for UK business owners.

A word on what this is and is not, because it matters for how much you should trust it. Aeora has not run a survey or a study. We do not have a proprietary dataset, and anyone in this field who claims a definitive one is usually overreaching. What follows is analysis of genuinely published, third-party figures, each attributed inline and listed in full at the end so you can check the methodology yourself. Where the studies disagree, or where a number is shakier than it first looks, I say so plainly. The point is not to frighten you into buying anything. It is to give you a sober, sourced read of where AI search actually stands, and what a measured response looks like.

How big is AI search really, and how fast is it growing

Start with scale, because the loudest mistake in this whole conversation is at one of two extremes: either AI search is already everything, or it is a rounding error you can ignore. The data supports neither. It supports something more useful and more interesting: a small share of traffic that is compounding at a rate traditional channels never manage.

The clearest single figure comes from Similarweb, reported by TechCrunch in July 2025. AI platforms sent 1.13 billion referral visits to the top 1,000 websites in June 2025, up 357% on the same month a year earlier. That is real people clicking through to real sites from tools like ChatGPT, Gemini, Perplexity and Copilot.

Now the context that keeps you honest. In the very same month, Google Search sent those same top sites around 191 billion referral visits (Similarweb via TechCrunch, 2025). So AI referrals are running at well under one percent of what Google Search still delivers. If you took that comparison alone, you might conclude AI search is not worth your attention for years. That would be a mistake for a different reason: the growth rate. A channel growing at triple digits year on year does not stay a rounding error for long, and the base it is growing from is already in the billions.

The assistants doing the sending are not niche either. OpenAI's Sam Altman said in October 2025 that ChatGPT had reached 800 million weekly active users (OpenAI via TechCrunch, 2025). Google confirmed in mid-2025 that its Gemini app had passed 450 million monthly active users (Google, 2025). These are among the fastest-adopted consumer products in history, and in the UK specifically, Ofcom's research found that a majority of UK adults now say they regularly encounter AI-generated summaries in their search results (Ofcom, 2025). Reach is not the constraint. What matters is whether these systems name your business when someone asks, and that is a separate question the reach numbers cannot answer.

What this means for you. Do not size this opportunity by today's referral traffic, because that number is small and will mislead you into waiting. Size it by the growth curve and the reach of the platforms, both of which are large. The businesses that treat AI search as a channel to build presence in now, while it is early, are the ones that will already be established when the traffic catches up to the reach.

What AI Overviews are doing to clicks

If AI referrals are the traffic flowing in, the harder story is the traffic that stops flowing out of Google. When Google places an AI-written summary above the blue links, fewer people click through to any website at all. The direction of that effect is now well established across independent studies. The exact size of it is genuinely contested, and I want to be straight with you about both.

The most carefully constructed study is from Pew Research Center, published in July 2025. Pew tracked the actual browsing of 900 US adults across nearly 69,000 Google searches in early 2025. When a search produced an AI summary, people clicked a traditional result in just 8% of visits. When no summary appeared, they clicked in 15% of visits, nearly twice as often. Clicking a link inside the summary itself was rarer still, happening in only about 1% of visits.

Sitting alongside Pew is a large market-research study from Ahrefs, updated in 2026, which compared click-through rates across 300,000 keywords with and without an AI Overview. Ahrefs found the top organic result loses 58% of its clicks on average when an AI Overview is present, up from a 34.5% drop the same team had measured eight months earlier. Different method, different population, but the same conclusion: the answer box takes a large bite out of clicks, and the bite is getting bigger.

Here is the honesty this topic demands. Google publicly disputes the Pew study, arguing the methodology and query set are not representative of how people really search (Pew Research Center, 2025). Google has an obvious interest in that position, but the objection is not baseless, and studies like these do measure averages across large samples that will not match your particular sector. So do not anchor on any single percentage. Anchor on the fact that four separate lines of evidence, from a respected non-partisan research centre and from independent industry analysts, all point the same way. Clicks fall when an AI answer appears. The argument is about how much, not whether.

What this means for you. If your Google Search Console shows impressions holding steady while clicks drift down, you are very likely watching this effect rather than a ranking loss or a penalty. The instinct to chase back the lost clicks is the wrong one, because they are structural, not recoverable. The better response is to make sure your business is named inside the answer that is replacing them, which is where the rest of this piece goes.

The overlap problem: ranking is not being cited

This is the finding I would most want a UK business owner to take away, because it quietly overturns twenty years of received wisdom. The comfortable assumption is that AI just reads Google's top results and repeats them, so if you rank well you are safe. The data says that is not how it works.

When Ahrefs ran 15,000 long-tail queries through ChatGPT, Perplexity and Copilot in 2025 and checked which pages those assistants cited against Google's and Bing's rankings, the overlap was small. Only about 12% of the URLs the assistants cited also ranked in the top ten for the same query. Put plainly, roughly seven or eight of every ten pages an AI names as a source are not sitting in Google's top ten at all. The assistants are drawing from a wider and different pool than the one businesses have spent years optimising for.

Now the nuance, because this number is noisier than a single figure suggests, and a flagship piece has to say so. A separate, larger Ahrefs study looking specifically at Google's own AI Overviews found a higher overlap, around 38% of AI Overview citations coming from pages that rank in the top ten, though that was itself down from about 76% seven months earlier (Ahrefs, 2026). Ahrefs is careful to add an important caveat: its citation-parsing has improved over time, so part of that apparent fall reflects better measurement of citations it previously missed, not purely a change in what Google selects.

So which number is right, 12% or 38%? Both, for different systems. Google's AI Overviews are built on top of Google's own index and lean more on its rankings, which is why the overlap there is higher. Standalone assistants like ChatGPT and Perplexity roam more widely, which is why the overlap there is lower. The table below lays the two studies side by side so the difference is clear rather than confusing.

The two figures are not in conflict once you see what each one measures. The shared lesson survives either way: a strong Google ranking helps your chances of being cited, especially inside AI Overviews, but it is nowhere near sufficient on its own, and for the standalone assistants it barely correlates. Ranking and being recommended have become two related but distinct contests.

What this means for you. Auditing your AI presence separately from your rankings is no longer optional. You can be page one on Google and still be absent from every AI answer your customers see, and you would never know it from your rankings dashboard. The only way to find out is to ask the assistants the questions your customers ask and record what they say.

What the research says actually moves AI answers

It would be easy to end on the problem and sell you a solution on faith. There is better ground than faith. There is now peer-reviewed academic work on what content choices measurably change whether a generative engine cites you, and it is worth knowing exactly what it found and what it did not.

The foundational study is titled GEO: Generative Engine Optimization, by Pranjal Aggarwal and colleagues at Princeton, Georgia Tech, IIT Delhi and the Allen Institute for AI, published at the KDD 2024 conference. The researchers built a benchmark they call GEO-bench, made up of 10,000 real queries across eight domains, each paired with the web sources a generative engine would pull from to answer. They then tested nine different ways of modifying a source's content and measured how each one changed that source's visibility in the generated answer.

The changes that helped most were not tricks. Adding relevant citations to back up claims, adding direct quotations, adding credible statistics, and writing in a clearer, more authoritative and fluent style all lifted a source's visibility in generated answers, with the strongest tactics producing improvements of up to roughly 40% in relative terms (Aggarwal et al., 2024). Crucially, the researchers found these gains varied a great deal by domain and query type, so 40% is a ceiling seen in favourable cases, not a number you should expect everywhere. They also validated the strongest tactics on a live system to check the findings held outside the benchmark.

Two things about this study make it trustworthy rather than just convenient. First, it is independent academic work with a published method and benchmark, not agency marketing. Second, and this is the part that should reassure you about the whole discipline, the tactics it found effective are the same things that make content genuinely good: cite your sources, quote credible voices, support claims with real evidence, and write clearly. The research is not describing a way to game the machines. It is describing why trustworthy, well-evidenced content gets cited, which is a far more durable thing to build on.

What this means for you. The levers that improve your standing in AI answers are legitimate and largely within your control: verifiable evidence, quotable answers, credible sources and clear structure. That is the honest opposite of the two hyped shortcuts doing the rounds. An llms.txt file has no confirmed effect on how you appear in AI answers, and schema markup alone does not buy citations. The work that the actual research supports is slower and less glamorous, and it is the work that lasts.

What all of this means for a UK business

Step back from the individual numbers and the shape of the year is clear. The audience is moving into AI answers far faster than most websites have registered. Google is answering more questions itself and passing on fewer clicks. And the businesses that get named in AI answers are being chosen through a process that only partly overlaps with the Google rankings you may have spent years and money building. None of that means traditional search is finished. It means the definition of being found has widened, and a business that only optimises for rankings is now competing in one half of a two-half game.

The measured response is not to panic, and it is certainly not to buy a guarantee, because nobody can honestly promise a specific placement in a system they do not control. The response is to treat AI visibility as a thing you measure and improve deliberately: find out what the assistants say about your business today, make your key pages answer the questions customers actually ask, ensure AI crawlers can reach your site, build a consistent presence in the third-party sources these systems read, and re-check every month because the answers move. We set out that whole method in our guide to AI search optimisation in the UK, put the wider shift in context in The Age of Answers, and dig into the click data specifically in the UK action plan on AI Overviews and falling traffic.

The one advantage that is genuinely time-limited is being early. Models learn who to trust from evidence that accumulates over time, so a consistent, well-evidenced presence built this year compounds while late arrivals start from behind. That is the sober conclusion the data actually supports. Not that the sky is falling, and not that you need to buy certainty from anyone, but that the ground under search has shifted, the shift is measurable, and the businesses that respond to it calmly and early will be the names the machines learn first.

Questions people ask

These are the questions UK business owners put to us most often about the state of AI search, with short, sourced answers you can act on or drop into a board update. The fuller working sits in the sections above and in the sources below.

Is AI search actually big enough in the UK to matter yet?

It is growing fast from a base that is still small next to Google. Similarweb counted 1.13 billion AI referrals to the top 1,000 sites worldwide in June 2025, up 357% year on year, while Google Search sent those same sites around 191 billion in the same month (Similarweb via TechCrunch, 2025). So AI referrals are a fraction of search traffic today, but a fast-compounding one, and the assistants doing the sending are enormous: ChatGPT reported 800 million weekly users in October 2025 (OpenAI via TechCrunch, 2025) and the Gemini app passed 450 million monthly users in mid-2025 (Google, 2025). The honest read is that this is early, not fringe.

Do AI Overviews really reduce clicks to websites?

The independent evidence points the same way, though the exact figure is disputed. Pew Research tracked the browsing of 900 US adults and found that when an AI summary appeared, people clicked a traditional search result in 8% of visits, against 15% when no summary was shown, and clicked a link inside the summary itself only 1% of the time (Pew Research Center, 2025). Separately, market research from Ahrefs measured a 58% fall in click-through for the top organic result when an AI Overview is present (Ahrefs, 2026). Google disputes the Pew methodology, so treat any single number with care, but multiple independent studies agree that clicks fall when an AI answer appears.

If I already rank on Google, am I automatically in AI answers?

No, and this is the most important thing for UK businesses to grasp. When Ahrefs ran 15,000 prompts through ChatGPT, Perplexity and Copilot and checked the citations against rankings, only about 12% of the URLs the assistants cited also ranked in Google's top ten for the same query (Ahrefs, 2025). A separate Ahrefs study of Google's own AI Overviews found a higher but still partial overlap of around 38%, down from 76% seven months earlier, though Ahrefs cautions that better citation parsing explains part of that fall (Ahrefs, 2026). Either way, ranking and being cited are related but far from the same thing, so you have to earn both.

Is there real research on how to influence what AI answers say?

Yes. The foundational study is GEO: Generative Engine Optimization by researchers at Princeton, Georgia Tech, IIT Delhi and the Allen Institute for AI, published at KDD 2024. They built a benchmark of 10,000 queries across eight domains and tested nine content changes, finding that adding relevant citations, quotations and statistics, and writing in a clear authoritative style, lifted a source's visibility in generated answers by up to roughly 40% in relative terms, with gains that varied a lot by query type (Aggarwal et al., 2024). It is peer-reviewed evidence that content choices measurably affect citation, not a guarantee of any specific placement.

Sources

Every statistic in this article comes from published third-party research, linked below so you can check the method yourself. None of these figures describe Aeora's own clients or a study we ran. Bear in mind that most of these studies measure averages across large samples, and several use US or global panels, so the numbers for your particular UK sector, query types and time of year will differ.

• Similarweb via TechCrunch, AI referrals to top websites were up 357% year over year in June, reaching 1.13B (2025)
• Pew Research Center, Google users are less likely to click on links when an AI summary appears in the results (2025)
• Ahrefs, Update: AI Overviews Reduce Clicks by 58% (2026)
• Ahrefs, Only 12% of AI-Cited URLs Rank in Google's Top 10 for the Original Prompt (2025)
• Ahrefs, Update: 38% of AI Overview Citations Pull From the Top 10 (2026)
• Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande (Princeton, Georgia Tech, IIT Delhi, Allen Institute for AI), GEO: Generative Engine Optimization, KDD 2024
• OpenAI via TechCrunch, Sam Altman says ChatGPT has hit 800M weekly active users (2025)
• Google via Statista, Google Gemini: global monthly active users 2025 (2025)
• Ofcom, From apps to AI search: how the UK goes online (2025)

Want to see where your own business stands against this picture, rather than the market average? Our fixed-price search and AI audit records exactly what ChatGPT, Gemini, Google's AI answers and the others say about you today, checks it for accuracy, and hands you a prioritised plan. You can get started here.