Home › AI Search

AI Search Visibility: How to Be Found When Someone Asks an AI

AI search visibility is how likely an AI assistant is to name your business when somebody asks it a question you could answer. It is not the same as ranking on Google. An answer engine reads a handful of sources, decides which it trusts, and quotes them, so the winner is not the page in position one but the page that is clearest and easiest to extract. Four things decide it: whether the engines can reach you, whether your content answers questions directly, whether your entity signals are consistent, and whether the wider web corroborates what you say about yourself.

How much of this is real, and how much is noise

More than half of UK adults, 54 per cent, now use AI tools such as ChatGPT, Copilot or Gemini. Among 16 to 24 year olds it is 79 per cent. Those are Ofcom's own figures, published in April 2026, and the headline number was 31 per cent only a year earlier.

Ofcom also found that around 30 per cent of Google searches now show an AI Overview, and that 53 per cent of UK adults say they see those summaries often. So this is not a coming change. It is already the default experience for most people searching in Britain.

What it does to clicks is measurable. Pew Research Center tracked the browsing of 900 US adults across nearly 69,000 Google searches and found that people who saw an AI summary clicked a traditional search result in 8 per cent of visits, against 15 per cent for those who did not see one. They clicked a link inside the summary itself in just 1 per cent of visits. Closer to home, Authoritas examined 3,500 search terms drawn from UK publisher news stories and found clickthrough rates fell 47.5 per cent on desktop and 37.7 per cent on mobile when an AI Overview was present.

And now the part that usually gets left out

The traffic AI assistants actually send is tiny. Chartbeat data reported by Press Gazette in January 2026 put ChatGPT at 0.02 per cent of total publisher referral traffic, and Perplexity at 0.002 per cent. The Reuters Institute's 2026 Digital News Report found that just 1 per cent of UK respondents say they always or often click through to the underlying sources from an AI answer, the lowest figure of any market it surveyed.

So if somebody tells you AI search is about to replace your Google traffic, they are overselling it. Barry Adams of Polemic Digital, who tracks this more carefully than most, is blunt about it.

“This Google Zero narrative is entirely false, and extremely dangerous.”

Barry Adams, Polemic Digital, March 2026

He is right, and it matters. Google is still where the overwhelming majority of your customers are, and the Competition and Markets Authority put that at more than 90 per cent of all UK searches when it granted Google strategic market status in October 2025. The reason to do this work is not referral traffic. It is that being named in an answer has become part of how a business gets on to a shortlist at all, and that effect shows up as brand search rather than as a referral. It is harder to see, which is exactly why it gets underestimated.

“When a business starts getting named in AI answers, the first thing that moves is not referrals, it is brand search volume. We see it in the Google Ads brand campaign before it shows up anywhere else. It is the cheapest early warning we have.”

Mink, PPC, Phoenix Marketing

The four things that decide whether an AI names you

One. Whether the engines can reach you at all

This is the cheapest and most commonly botched part. AI companies run more than one crawler and they do different jobs. OpenAI's own documentation describes GPTBot as being used to crawl content that may train their foundation models, and OAI-SearchBot as being used to surface websites in ChatGPT's search features. Block the first and you stay in ChatGPT's search results. Block the second and you do not. Anthropic and Perplexity draw the same distinction between their training crawlers and their search crawlers.

A great many UK sites have blocked both, usually in one line, usually without realising the second one was in there.

Two. Whether your content can be extracted

Cyrus Shepard of Zyppy reviewed 54 published studies, experiments and patents and scored 23 possible AI citation factors on how well each was corroborated. The two that scored highest were plain URL accessibility and ordinary search rank. He also found that 38 per cent of AI Overviews citations come from the top ten Google results, which is a useful corrective to the idea that AI search is a separate game with separate rules. Most of what wins here is ordinary search engine optimisation done carefully.

What is genuinely different is the shape of the writing. Will Critchlow of SearchPilot puts the mechanism plainly.

“LLMs do not rank anything. There is no internal ranking system in the way SEOs mean that phrase.”

Will Critchlow, SearchPilot, May 2026

An engine is looking for a passage it can lift with confidence. That rewards answering the question in the first sentence, keeping each idea in a self contained unit, and being explicit about dates, sources and scope. It punishes the slow build up that traditional brochure copy is made of.

Three. Whether your entity signals are consistent

Being an unambiguous entity means an engine never has to guess which business you are: one name, one address, one phone number, one description, repeated consistently everywhere you appear. Structured data helps by stating those facts in a form nothing has to infer from prose.

One honest caveat, because this is oversold constantly. Google's own optimisation guide says that structured data is not required for generative AI search and that there is no special schema markup you need to add for it. We still recommend it, because it earns rich results in ordinary search and because it removes ambiguity, but nobody should be selling it to you as an AI ranking factor.

Four. Whether the wider web corroborates you

This is the one most businesses have least control over and most underestimate. Jono Alderson, who writes some of the clearest technical analysis in the UK on this, describes what is actually happening.

“AI systems don't simply repeat the story you tell about yourself. They can test it against whatever documentation, reviews, competitors, support forums, pricing, specifications and expert opinion their retrieval systems happen to surface.”

Jono Alderson, July 2026

He adds that being mentioned in ten places is not inherently useful if those mentions are inconsistent, shallow, outdated or disconnected. Which is to say that consistency beats volume, and that a review profile, a trade association listing and a chamber of commerce page are doing work here that a blog post cannot do.

It is also worth knowing where answers actually come from. In the Pew study, the three most frequently cited sources, across both AI summaries and standard results, were Wikipedia, YouTube and Reddit.

“What surprised me is how often the answer comes back citing a Reddit thread or a YouTube video rather than the company's own website. It has changed how we think about where a client's content should live. Your site is not always the thing being read.”

Tommy, Social Media, Phoenix Marketing

The four surfaces do not behave the same way

Treating ChatGPT, Google, Perplexity and Copilot as one audience is the most common strategic error we see. If the terminology is new to you, start with our plain English guide to what answer engine optimisation actually means. Research by Semrush with Kevin Indig of Growth Memo, covering 3,981 domain appearances across 115 prompts and four engines, found almost no overlap in which brands each engine named for the same prompt. Indig's summary of it is worth repeating.

“There's almost no overlap between which brands ChatGPT cites and which ones Gemini names for the same prompt. These are different behavioral systems. Treat them that way.”

Kevin Indig, Growth Memo, June 2026

The same study found something else that ought to change how agencies report on this. Almost 62 per cent of the citations they measured were what they called ghost citations: the engine used the page as a source link, but the brand name never appeared in the answer the person actually read. Being cited and being named are two different outcomes, and only one of them wins you the customer.

• ChatGPTAnswers from training and from live search. The live route, served by OAI-SearchBot, is the realistic one for most businesses.
• Google AI Overviews and AI ModeBuilt on Google's existing index, so ordinary SEO does most of the work. Both are now formally in scope of the CMA's strategic market status designation.
• PerplexityCites its sources visibly, which makes it the easiest surface to test on and the best place to see who is being chosen instead of you.
• Microsoft Copilot and ClaudeLower volume, disproportionately used inside businesses, and cheap to get right. For B2B this is not the afterthought it looks like.

Three things changed in 2026 that most advice has not caught up with

Google now reports on it, and lets you opt out

Since June 2026 Search Console has carried a generative AI performance report showing impressions from AI Overviews and AI Mode, alongside a separate control that removes a site from generative AI features altogether. Google states that this control is not used as a ranking or inclusion signal affecting other parts of Search. If your agency is still building you a workaround segment in analytics and calling it AI reporting, it is behind.

Google-Extended does not do what people think it does

Google-Extended governs whether your content may be used to train and ground Gemini models. Google's documentation says in terms that it does not impact a site's inclusion in Google Search and is not used as a ranking signal. It is not the AI Overviews switch. That is the separate Search Console control above. We have seen this conflated in three separate audits this year, including one sold to a client of ours.

Google ignores llms.txt

The llms.txt proposal has been sold hard this year. Google's own optimisation guide now addresses it directly, saying that creating such files will neither harm nor help your visibility or rankings in Google Search, because Google Search ignores them. We still publish one, because it takes an hour and other systems may come to use it, but we will not tell you it is an AI search tactic. If somebody quotes you for one, ask them to show you who reads it.

What Phoenix Marketing does about it

We start with an audit, hand checked rather than generated by a tool, scored out of 100 across five areas: whether AI crawlers can reach you, whether your content can be extracted, whether your structured data is valid, whether your business is an unambiguous entity, and whether the basics of AI era readiness are in place. We publish the weightings so you can argue with them. It sits alongside the rest of our marketing services rather than replacing any of them.

It is hand checked for a reason. When we ran an automated pass across 19 sites earlier this year, two of them scored between 25 and 30 on the first attempt and 85 and 90 on a retry, because bot protection had served an interstitial that looks identical to an empty page. A tool would have told those two businesses their websites were invisible to AI. They were not.

After the audit, the work itself is unglamorous and continuous: crawler access, answer first rewrites on the pages that matter, schema and entity clean up, and the corroboration work that sits closer to public relations than to SEO. Entity details drift, schema breaks silently when a plugin updates, and competitors publish. This is not a project you finish.

“We ran this audit on ourselves before we ran it on anybody else, and we scored 58 out of 100. No structured data on the home page, a two word heading, and a slogan in the place where the answer should have been. We would rather show you that than a case study with the awkward bits taken out.”

Stuart Humphrey, Phoenix Marketing

How we measure it, and what the numbers will not tell you

Four things, each imperfect on its own. The Search Console generative AI report for impressions in AI Overviews and AI Mode. An analytics segment grouping referrals from ChatGPT, Perplexity, Gemini and Copilot, which will be a small number and should be. Brand search volume, which is where the real effect usually appears first. And manual prompt testing, run the same way each month, because there is still no substitute for asking the question a customer would ask and reading what comes back.

What none of that captures is the person who read an AI answer naming you, did not click, and searched for you by name two days later. That is the bulk of the value and it is the hardest thing to attribute. Anyone presenting you with a clean, complete AI attribution model is showing you a model, not a measurement.

Frequently asked questions