Your Next Customer Is Asking ChatGPT First
By the alphaa team — we run AI-visibility scans across ChatGPT, Claude, Gemini and Perplexity for local businesses, so we see firsthand which businesses get named in these answers and which get skipped. This is part three of our Death of the Blue Link series.
Short answer: A growing share of buying research now starts inside an AI assistant rather than a search box — 65% of consumers now use AI tools to research products before buying (Clutch, 2026), and ChatGPT alone serves 800M+ weekly active users, handling billions of queries every day (OpenAI, 2025). The practical consequence for your business is that a shortlist can be formed before anyone visits a website, reads a review page, or sees a single ad. If the assistant doesn't name you, you were never in the running — and nothing in your analytics will tell you it happened.
What actually changed about the first question
The old sequence was: search, scan ten links, open four tabs, compare, decide. Every step of that happened on surfaces you could measure — a ranking, a click, a session, a bounce rate.
The new sequence collapses the first three steps into one exchange. Someone types "I need a good family dentist in Tempe, my insurance is Delta Dental, and I'm nervous about dental work" and gets back three names with reasons attached. That is not a list of links; it is a recommendation with rationale, shaped to a context far richer than a keyword. Then the follow-up questions do the comparison work that used to happen across four open tabs — pricing, hours, whether they handle anxious patients.
By the time a website is opened, the decision is often mostly made. The visit is confirmation, not discovery. This is why the shift is easy to underestimate from inside your own dashboard: the traffic you still get looks normal, because the customers you lost never generated a single event.
Why AI answers are trusted differently
A results page presents ten options and openly signals that some paid to be there. The reader does the filtering and knows they are filtering. An assistant presents three, in a confident sentence, with the filtering already done and the reasoning stated. Psychologically these are not the same artifact, even when the underlying sources overlap heavily.
There is real evidence this produces better-qualified visitors: traffic to retail sites from AI assistants grew over 1,200% in under a year, and those visitors convert better than traditional search traffic (Adobe Analytics, 2025–26). That figure is retail, not local services, so don't import it wholesale to a plumbing business — but the mechanism travels. A customer who arrives after an AI has already vouched for you is warmer than one comparing ten open tabs.
How the assistant decides who to name
This is the part worth understanding precisely, because it is where the actionable work lives. When you ask a modern assistant for a local recommendation, the answer is assembled from two things rather than recalled from a directory:
- Training knowledge. What the model absorbed from the public web during training, frozen at a cutoff date. It shapes the model's general sense of your category and whether your name registers at all.
- Live retrieval. Most assistants now fetch current web results at the moment you ask, read them, and synthesize an answer — retrieval-augmented generation. This is why an assistant can cite a page it never saw in training, and why recent, crawlable, specific content matters.
Weighing across both, models reward businesses described verifiably, specifically, and consistently across many sources. If your site, your Google Business Profile, two review platforms, and a directory listing all state the same concrete facts, the model can assert them confidently. If sources contradict each other — different hours, an old address, services you dropped two years ago — the model hedges, and hedging usually means naming someone else. Ambiguity is the enemy here more than obscurity.
Note the asymmetry this creates: being described by others carries weight your own site cannot manufacture. You can write anything about yourself; you cannot write a third-party review. That is exactly why the model leans on outside sources when the stakes of the recommendation are real.
What this looks like in practice
A concrete pattern we see repeatedly when scanning local businesses: the business ranks perfectly well on Google, has decent reviews, and is simply absent from the AI answer — while a smaller competitor with a thinner website gets named. The usual cause is not that the competitor did clever AI marketing. It is that the competitor's facts are consistent and specific everywhere, and the larger business has three different phone numbers across the web, a service list that doesn't match its own homepage, and a Google profile last updated in 2023.
The second pattern is worse and more common than owners expect: the business is named, with wrong details. Old address, hours from before a move, a service they discontinued. The assistant states it confidently, because confidence is a property of the model's phrasing, not of the underlying evidence quality. Customers act on it. Nobody tells you.
How to check where you stand, today
You can do this manually in about five minutes, and you should do it before buying anything from anyone — including us:
- Ask the need, not the name. Open ChatGPT, Gemini, and Perplexity and ask "who is the best [your service] in [your city]?" the way a customer would phrase it, with a real constraint attached (insurance, budget, urgency, a specific problem).
- Ask the name directly. "Tell me about [your business] in [your city]." Check every fact in the answer against reality. Wrong facts are more urgent than absence.
- Look at the citations. Where assistants show sources, note them. Those are the pages actually shaping your reputation with AI — often not the ones you would guess, and frequently not your own site.
- Repeat across phrasings. Answers vary by wording, by user, and by week. One query is an anecdote; five give you a pattern.
Then fix the boring things first, in this order: make your name, address, phone, hours, and service list identical everywhere they appear; write plain, specific sentences that answer the questions customers actually ask, in language a model can lift verbatim; and build genuine third-party evidence through reviews and mentions. The AEO checklist walks this in order, and how to get recommended by ChatGPT goes deeper on the engine-specific mechanics.
The honest caveats
The version of this story that sells courses drops the limits. Keeping them:
- Google has not disappeared. It remains the largest single surface by a wide margin. The claim here is that AI assistants are now a meaningful and fast-growing first step, not that search is over.
- "Research with AI" is broader than "chose a business via AI." The 65% figure (Clutch, 2026) covers using AI tools to research products before buying, which spans everything from comparing features to drafting a shortlist. Don't inflate it into a claim about final purchase decisions.
- Nobody can guarantee a citation. Outputs vary by phrasing, user, model version, and time. You improve the evidence models read, which shifts the odds in your favor. Anyone selling guaranteed placement in ChatGPT is describing a lever that does not exist — we cover why in Is AEO real?
- This is not instant. Consistency work takes weeks to propagate through the sources models read, and training knowledge updates on its own schedule entirely.
The bottom line
The first question about your category is increasingly asked of an assistant, not a search box, and the shortlist that comes back is built from evidence you can influence but not control. The businesses adapting well are not doing anything exotic — they are making true things about themselves specific, consistent, and easy to verify everywhere a model might look. The uncomfortable part is that you cannot see any of this in your analytics. The customer who asked and never heard your name left no trace. So the only way to know is to go ask.
Last updated July 21, 2026.