An agency reports the good news: first result for the main term, top three for a dozen others, traffic up. It is all true. Then the owner asks ChatGPT which firm to use in their city and reads three competitors' names, none of them theirs.
This combination is common enough to be unremarkable, and it is not a contradiction. The two systems are being asked to do different jobs.
Different jobs, different winners
A search engine's task is to choose which pages deserve a click. It compares pages against a query and against each other, and its judgments are positional: first, second, third. The unit of competition is a page, and the reward is attention directed at that page.
An assistant is not allocating attention. It is composing a paragraph — a short, confident, standalone answer that will be read instead of any page. To do that it needs claims it can state without hedging: specific enough to be useful, attributable enough to be safe, and short enough to fit.
That difference reshapes what counts as good material.
Length works against you. The comprehensive 3,000-word guide is an SEO asset. For an assistant it is a haystack, and if the retrievable fact is in paragraph forty, it may never be extracted.
Persuasion works against you. "Trusted by businesses across the region for over two decades" is a fine line on a landing page. It contains no fact, applies equally to every competitor, and cannot be lifted into an answer, because lifting it would say nothing.
Self-description carries less weight. In search, your page is a legitimate authority on you. To an assistant deciding whom to recommend, a business describing itself is a weaker input than a third party describing several businesses side by side. The pages that decide recommendations are mostly not yours.
Pages that rank poorly can still dominate. A directory nobody visits directly, or an old roundup buried on page three, can be quoted in a great many answers, because its format is exactly the format an assistant needs: names, category, place, reasons.
Where the two do overlap
It would be wrong to describe these as unrelated. There is real shared ground, and it is worth being precise about where.
Assistants with live search query a search index. Pages that rank are more likely to be retrieved, so search visibility does feed AI visibility — but only for the fraction of answers that use live search, and only via whichever pages get opened, which is often three or four rather than ten.
Technical health matters in both. A page that cannot be crawled cannot be cited. Clear structured facts — what you do, where, hours, prices, credentials — help both systems, and the schema markup you already added for rich results makes those facts unambiguous to anything reading the page.
And a strong link profile correlates with the kind of third-party presence assistants lean on, because both tend to result from other people writing about you. The correlation is real. The mechanism is not identical, which is why the two can come apart.
Why the divergence is largest for local businesses
For a local business, the search contest is narrow: a handful of competitors, a defined area, a well-understood set of ranking factors, and an agency that has been optimising against them for years. Winning it is achievable.
The AI contest has a wider catchment. An assistant answering best physiotherapist in Bristol is not restricted to the pages that rank for that phrase; it draws on everything it absorbed about physiotherapy in Bristol from anywhere, of any age. A practice can be first on Google today and absent from the assistants because the material they learned from was written before it opened, and nothing since has corrected the impression.
Time works differently too. Search rankings respond to changes in weeks. A model's absorbed knowledge may not shift until the next training run, whatever you publish in the meantime.
The practical position
Keep doing SEO. It brings customers, and it makes your pages retrievable.
But treat AI recommendation as a second, adjacent channel with its own scoreboard — because none of your existing reporting measures it. Rank trackers do not watch it. Analytics shows a customer who arrived after an assistant named you as direct or organic traffic, indistinguishable from anyone else. The channel is invisible in exactly the tools you would expect to see it in, which is why it tends to be discovered late and by accident.
The first useful step is not a strategy. It is a measurement: what do all four assistants actually say when a customer asks the questions that lead to a purchase in your category — and if you are not named, who is, and which source put them there.