AI Search

The behavioural shift is conversational: people describe situations ("my AC died and it's 105 out, who's good near me?") instead of typing keyword fragments, and they ask follow-ups. Answers reward the businesses whose evidence is complete, current, and specific enough to be safely repeated.

The structural shift is narrower results. A page of ten blue links lets a searcher scan several plausible options and pick. An AI answer names two to five businesses and moves on. There is no second page, no scrolling past the fold to find a better fit, and no obvious way for a searcher to know who was left out. Being the fourth-best roofer in a city used to mean steady work from a first-page listing; in this channel it frequently means nothing at all.

The shortlists are also narrower than most owners expect once they see the data. Across the 540 answers in our study, assistants named about 47 different businesses per market — but roughly 60% of those were named exactly once, in a single answer, and never again. The durable shortlist in most markets is a handful of names, and the long tail is noise that will not survive the next re-run.

The channels differ in mechanics but share one property: they are measurable only by asking. There is no results page to inspect, no rank to look up, and no console reporting your impressions. You ask the questions your buyers ask, record what comes back, and repeat it often enough that non-determinism averages out.

Do people actually use ChatGPT to find local businesses? covers the adoption evidence honestly, including the parts that argue for patience rather than panic.