AI visibility: how to get your business mentioned by ChatGPT, Gemini and Google AI

More people now ask ChatGPT, Gemini, Perplexity or Google's AI Overview for a recommendation before they visit a website. When they do, the AI tool answers from what it has read about you, and very little of that comes from your own site. This article sets out what the research found, what it means for a business or a board in Dubai or the UK, and a test you can run yourself in an afternoon.

What the research found

McKinsey's State of the Consumer 2026, published on 22 June, studied roughly 2.6 million citations given by ChatGPT, Gemini and Perplexity in answers about 25 consumer brands. The study covered 89,119 websites in the UK and US between October 2025 and May 2026. Only 1 per cent of the sources the AI tools cited were the brands' own websites. The ten most-cited websites accounted for about 22 per cent of all citations, and even among those ten, brand websites were at most 10 per cent.

The same report shows how quickly habits are changing. Visits to websites through direct traffic and search clicks are down 8 per cent since 2023. Among Gen Z, 60 per cent regularly use the AI overview at the top of a Google search and 28 per cent already use AI tools when they shop. For baby boomers the figures are 29 per cent and 16 per cent.

Two cautions. The citation study covered consumer goods only, in two countries, using one measurement tool. A law firm, a clinic or a property developer may see a different split. The figures on behaviour come from what people said in a survey, and those answers are less precise than measured behaviour. The overall finding is reliable. The exact percentages will vary by sector.

Where AI tools get their information

If only 1 per cent of citations come from company websites, the useful question is where the other 99 per cent come from. McKinsey gives the breakdown. Competitor and other brand websites accounted for 30 per cent of citations on ChatGPT, 40 per cent on Gemini and 29 per cent on Perplexity. The rest came from retailers, publishers, review sites and forums.

The BoF-McKinsey State of Fashion 2026 reaches the same conclusion. Reviews, blogs and affiliate posts make up around 80 per cent of the sources AI tools use. Its interview with the founder of Profound, a company that tracks AI answer engines, adds a technical point that many site owners miss: content built in JavaScript is invisible to many AI crawlers, so a well-designed page can be unreadable to the tool you want to read it.

Google's own guidance is short and worth reading, because it contradicts most of the sales pitches you will receive. Google Search Central states that there are no additional requirements to appear in AI Overviews or AI Mode, no AI text files, no special markup and no special structured data. A page must be indexed and eligible to show with a normal snippet. Google recommends the same things it has recommended for years: pages that can be crawled, useful text, internal links and a good page experience.

In plain terms, AI tools learn about your business mainly from what other people publish about you. Your own pages need to meet ordinary technical standards, and nothing more than that. I made the same point in brand trust when the buyer is a machine: the evidence that matters sits on other people's websites.

Why the answers change so easily

Search rankings move slowly enough to plan around. AI answers change much faster, and the research now shows by how much.

Oguz Acar of King's Business School and David Schweidel of Emory University, writing in Harvard Business Review in March 2026, cite research from Carnegie Mellon University in which changing a single word in a question moved the likelihood of a consumer being pointed to a particular brand by as much as 78.3 per cent. Their example is Chivas Brothers, the Pernod Ricard whisky business, which found that one AI tool had placed Ballantine's in the wrong category altogether, and only knew because it checked.

McKinsey's June 2026 report on advertising and AI surveyed 182 US advertising leaders. More than half of Google searches now show an AI overview. 42 per cent of the advertisers surveyed said that depending on automated systems they cannot see inside is a key risk to their media spending. The sample is small and US only, and the concern is a fair one.

It is also worth keeping the scale in view. AI tools accounted for 2 to 4 per cent of global search in the first half of 2025, according to the State of Fashion 2026. Bing alone accounts for 4 per cent. AI summaries inside normal Google searches were 16 to 23 per cent. Traditional search was still 74 to 79 per cent. The change is happening, it is still at an early stage, and there is time to prepare for it in an organised way.

What not to pay for

Three services are being sold heavily at the moment. I would decline all three until the work in the next section is done.

• An "AI SEO" or "GEO" retainer that promises you a position. If one word in a question can change a recommendation by 78 per cent, nobody can promise a position. Ask what they will measure, how, and whether they will show you the actual AI answers each month.
• A rewrite of your website "for AI". Google has said in writing that no special files or markup are needed. If your pages are indexed, readable as text and answer real questions, they qualify. Rewriting them in a mechanical style makes them worse for the people who still make the buying decision.
• A dashboard that gives you an "AI visibility score" without showing the sources. The useful part is the list of websites the AI tools cite instead of you. A score on its own gives you nothing to work on.

The approach is the same one I use in every marketing audit: find the evidence, name the source, decide what to change, then measure again.

Three things worth spending on

These three keep their value however the AI tools develop, because a human buyer checks the same things.

1. Evidence on other people's websites. Reviews on the platforms your customers use, trade press coverage, directory listings with the same name, address and description everywhere, expert citations, guest articles, and a Google Business Profile that is complete and current. This is what the AI tools read, and it is what a careful buyer reads too. Most of it costs time rather than money.

2. Clear facts on your own website. Plain pages that state who you are, what you do, for whom, at what price, and where. An FAQ that answers the questions customers actually ask, in full sentences, on a page that displays as text. The same company details on your site as on your profiles. Structured data is not required by Google for AI features, and it still helps any machine confirm that the company on LinkedIn, the one in the trade press and the one on your site are the same business. It costs little and it is worth having.

3. A monthly check. The same set of customer questions, run through the same three or four AI tools, recorded in a spreadsheet, so that in six months you know whether anything has changed. Without this, every other decision is a guess. The test is below.

None of this costs more than the agency retainer many SMEs already pay for links that nobody clicks. If you want to check where else AI belongs in the business first, the AI audit for SMEs is the place to start.

The ten-question test

This is the method I use with clients. It takes an afternoon the first time and about an hour a month after that.

• Write ten questions. Six without your name in them, the way a buyer would ask before they know you exist: "who are the best X in Dubai", "recommend an X for a growing UAE business", "I need X without hiring someone full time, who should I consider". Four with your name: "who is [your name]", "tell me about [your website]", "is [your company] any good", "what does [your company] charge".
• Ask them in four tools. ChatGPT logged in with search switched on (logged out, it answers from memory and the results are not representative), Perplexity (it shows its sources openly, which makes it the clearest evidence), Gemini logged in, and Google itself, noting whether an AI overview appears for the unbranded questions and whether you are in it.
• Record five things for each question in each tool. Whether you were mentioned. Whether you were linked, and to which page. Every source the tool used. For the branded questions, whether the description is accurate and which facts it used. And the competitor names that keep appearing, because they are your real competitors in AI answers, whatever your pitch deck says.
• Score it. Your visibility score is the number of tools in which you appear in an answer to an unbranded question. Being recognised when someone uses your name is a good sign and it does not count towards the score.
• Turn the list of sources into a work plan. Every website the AI tools cite instead of you is somewhere you need to be present: a review to earn, a directory listing to correct, a publication to pitch, a forum to answer in. Rank the list by how often each source appears. That is the content and PR work for the next three months, and almost none of it is on your own website.

Question	Tool	Mentioned	Linked (page)	Sources used	Accurate	Competitors named
Who are the best interior design firms in Dubai?	Perplexity	No	No	Two directory sites, a lifestyle magazine, Google reviews	n/a	Five firms named
Who are the best interior design firms in Dubai?	ChatGPT	Yes	No	Directory site, Instagram, a magazine feature	n/a	Four firms named
Tell me about [firm name]	Gemini	Yes	Yes (home page)	Company website, LinkedIn, Google Business Profile	Partly (old address)	None

The example above is a made-up Dubai interior design firm. Three rows are enough to show how the log works: the firm is recognised when named, it appears in one tool's unbranded answer with no link, one directory site and one magazine keep appearing as sources, and Gemini has an old address. Each of those is a task. Copy the layout into a spreadsheet and complete it for your own business.

A note for UAE businesses

Three things matter more here than in the UK.

Consistent details across directories. UAE businesses collect listings quickly, on free zone portals, chamber directories, Google, aggregator sites and marketplaces, and the company name, licence description and address rarely match across all of them. An AI tool that reads five inconsistent listings has less confidence in you than one that reads five identical ones. Correcting this is administrative work, and it makes a measurable difference. It is the same problem I describe in competitive analysis in the UAE: the visible market is wider and less tidy than the real one.

Arabic and English. If your customers ask in Arabic, run the branded questions in Arabic as well and see whether the tools find your Arabic pages, or translate your English ones badly. Bilingual FAQ pages, written properly rather than machine-translated, are an advantage very few SMEs here have taken.

Google Business Profile. For any business with a physical location, the profile is one of the strongest signals available to every AI tool, and in my experience it is the most neglected asset in the region. Categories, opening hours, photos, services, posts and reviews all feed the tools.

What to expect, and what to do next

The normal starting position for a specialist firm or an SME with a good website and a short record of third-party coverage is this: the AI tools recognise you when someone uses your name, and they do not recommend you when someone asks about your category. That is why the work plan starts on other people's websites. Mentions in unbranded answers follow the same pattern as organic search rankings and arrive months after the work is done, so the sensible expectation is steady progress over two or three quarters rather than a result next week. The monthly check matters for that reason: without it, you cannot tell the difference between work that is taking effect and work that is not working.

For a business owner, this means a small monthly budget, an hour a month, and a list of places to be present. For a board, it means one question for whoever runs marketing: where does our business appear when a buyer asks an AI tool instead of Google, and how do we know. If the answer is a score with no sources behind it, ask again. A fractional CMO can own this work, and it starts with the same test described above.

Sources. McKinsey, State of the Consumer 2026: When tech acceleration and cost pressures collide, 22 June 2026. McKinsey, The agentic advertising economy: From attention to action, June 2026 (survey of 182 US advertising leaders, February 2026). The Business of Fashion and McKinsey, The State of Fashion 2026, November 2025. Oguz Acar and David Schweidel, "Preparing Your Brand for Agentic AI", Harvard Business Review, March to April 2026. Google Search Central, "AI features and your website", accessed 19 August 2026.