Being found when the machine answers

Imagine walking into a bookshop and asking where to find something. You get a shelf number. Ask the bookseller what she makes of an author or a topic, and you get something else entirely: an answer, in her own words, assembled from everything she has read.

Now watch which books she reaches for. Not just a book written about the topic or by the author, but books written around the topic or about the author. She might explain how often they agree with one another and how quotable they are, going beyond whose name is on the spine. And she won’t show how she arrived at her explanation unless you specifically ask her to explain.

That bookseller is the AI assistant, and it now sits between your company and the people trying to find out about it.

Changing communications in the world of AI

In February, @Serra Balls argued in Communications in the world of AI that AI is homogenising journalism, because outlets increasingly rework material from the same shrinking pool of original sources. And that when content becomes infinite, “in this world of a lack of trust, credibility is the only agency we now have.” If the machines draw from a shared pool, then whether a firm sits in that pool decides whether it exists in the answer at all.

But this type of search has been mainstream only for about a year now and is moving very fast. It feels like 2003 redux, when search had no settled metrics, no agreed vocabulary, a lot of confident people making things up. We have a handful of practices that will look obviously wrong by this time next year.

Search optimisation is still important

AI has not killed search, and getting search right still matters, because Google still handles over five trillion searches a year and held 91.27% of global search in June 2026, according to Statcounter. Google's own 2025 data showed usage increasing by over 10% on the query classes where AI Overviews appear. Gartner's much-quoted 25% decline is a forecast, not an observed outcome.

According to BrightEdge research from February 2026, only 11% of the sources cited in AI Overviews for finance queries also appear in Google’s organic top ten. Across all sectors it's around 17%. A firm's page-one rankings are buying them very little in the answer layer. Search rank and citation rank have come apart, and that gap is the whole opportunity.

In order to understand the gap, we need to distinguish between three terms which are often used interchangeably but shouldn’t be:

●       Search Engine Optimisation or SEO makes a page discoverable and competitive in conventional search results. The question posed here is “can we be found?”

●       Answer Engine Optimisation or AEO improves the likelihood of a brand appearing in answer-led surfaces such as AI summaries, assistants and voice enquiries, answering the query “are we the answer?”

●       Generative Engine Optimisation or GEO makes a source more likely to be cited as well as cited prominently, inside a generated answer and this answers “are we quoted when the machine talks?”

SEO helps with being findable. AEO and GEO make a brand quotable. You need all three.

Emerging opportunity for firms

Muck Rack analysed more than 25 million links across ChatGPT, Claude and Gemini in May 2026 and found that 84% of AI citations come from earned media. Journalism alone accounts for 27% while paid and advertorial content accounts for 0.3%.

In a study of 75,000 brands, Ahrefs found branded web mentions and video presence were better predictors of AI visibility than domain authority. The number of pages on a firm’s website made almost no difference at all. Firms cannot rely solely on search optimising their own content to shape how they appear in AI answers.

Other sources are already describing the category, comparing providers and defining who matters within it. In Conductor's 2026 financials benchmark, comparison publishers such as NerdWallet and Bankrate outperform many regulated financial institutions on citation share.

What’s the question?

Every prospect is asking some version of the same question: is this firm any good, and can I trust it? AI assistants are increasingly one of the places they ask it. The exact question varies by sector, from whether a crypto exchange is safe and regulated, to which cross-border payments provider is best, to an asset manager’s track record or a company’s credibility on transition. 6sense found in 2025 that 94% of B2B buyers now use LLMs somewhere in the buying journey.

These questions are not answered by a company’s website alone. AI systems draw on the wider body of coverage, commentary and comparison around it. The commercial impact varies: for businesses where buyers are comparing providers, visibility determines whether a firm makes the shortlist at all; in relationship-led businesses, the bigger issue is accuracy — what an assistant says when someone checks a name they already know.

Boards should also be worried about liability. The FCA requires financial promotions to be fair, clear and not misleading, with enforcement increasing, while ESMA has confirmed that existing MiFID II obligations still apply when firms use AI in investment services. The SEC has also acted against misleading AI claims.

The harder issue is what happens when an assistant independently describes a regulated firm. Liability is still unclear, but the practical risk is simpler: most firms are not measuring what AI systems say about them, so inaccurate descriptions can circulate without appearing in any ranking report, media alert or cuttings file.

There is no settled industry metric yet, and answers vary between runs, so the sensible approach is to track patterns over time rather than individual results. Most firms have never measured this at all. Ask each of the major assistants the four or five questions a real prospect would ask, repeat the exercise several times, and note who gets named, what gets said and which sources shape the answer. It takes an afternoon and the results are usually revealing. But waiting for perfect measurement risks waiting too long. If AI systems are drawing from a shrinking pool of credible original sources, then being present in that pool matters more, not less.

If you’d rather we ran the full audit across every major assistant and tracked it over time, get in touch with us.Nobody can tell the bookseller what to say, but we can help make sure there's something worth reading.