AI Search Is Changing How Brands Get Found

For twenty years the mechanics of being found were stable: publish a page, earn links, climb the ranking, collect the click. That chain is now breaking in the middle. Search engines increasingly answer the question on the results page itself, and a growing share of buyers begin with a chatbot rather than a search box. The user receives a synthesised answer with two or three sources attached — and often has no reason to visit any of them.

The numbers behind this are still contested, but the direction is not. Publishers across news, travel and software have reported double-digit declines in organic traffic to informational pages since AI-generated summaries became standard, even where their ranking positions held steady. Commercial pages have held up better, because people still click when they intend to buy. What has evaporated is the traffic that used to arrive from questions.

What replaces the click is the citation. When a model composes an answer, it draws on a handful of sources it judges relevant and trustworthy, and it names some of them. Being one of those named sources is the new front page. The awkward part is that the selection is probabilistic: ask the same question twice and the sources may differ, so visibility is no longer a position you hold but a probability you influence.

Early practitioners — the field has already acquired the ugly name generative engine optimisation — report a consistent pattern in what gets cited. Content that states a clear answer near the top, uses concrete numbers, carries a named author and a recent date, and is structured in a way a machine can parse tends to surface more often. So do brands mentioned frequently across third-party sources: reviews, forums, comparison sites, trade press.

That last point deserves attention, because it inverts a decade of practice. Optimising your own site matters less than being talked about elsewhere, since a model synthesises consensus rather than reading your homepage. A brand with no reviews and no independent coverage is, to an answer engine, close to invisible — however polished its own content. Public relations, in other words, has quietly become a technical discipline again.

Measurement is the unsolved part. Analytics tools were built to count visits, and there is no reliable equivalent of a rank tracker for a system that generates a different answer every time. The workarounds are crude: run a fixed set of buying questions through the major assistants each month, record whether your brand appears and how it is described, and treat the result as a sample rather than a score.

At work, the useful framing avoids both panic and dismissal. "Our informational traffic is falling while our conversion rate is stable — the visits we lost were probably never going to buy" is a defensible reading of most current data. "We should measure how often we are cited alongside how often we rank" gives your team something concrete to do. Avoid promising a method for ranking inside a model; nobody credible is selling one yet.

None of this means the discipline is dead; it means the unit of visibility has moved. The work is still to be the clearest, most citable source on a question your buyers ask, and to be mentioned by people whose opinion the machine has learned to weigh. What has ended is the comfortable assumption that a good ranking converts into a visitor. Increasingly it converts into a sentence in somebody else's answer.