I Asked ChatGPT to Recommend a Business in My Town — Here's What Actually Decided the Winner

The Question That Started It

I do this for a living, and it still caught me off guard.

On a slow Sunday, I opened ChatGPT with search enabled and typed the kind of thing a real person types — no keyword gymnastics, no SEO brain: "I'm looking for a good physical therapy clinic near [my town]. Which one should I go with?"

I live in a mid-sized suburb — call it Riverton, a stand-in for the thousands of towns exactly like it, because the point isn't the place, it's the pattern. There are eleven PT clinics within a fifteen-minute drive. I know most of them. A few are excellent. One or two are, frankly, coasting.

The model didn't hedge. It named one clinic first, described it in a confident sentence, listed two runners-up, and moved on. No ten blue links. No "here are some options for you to evaluate." A recommendation. A verdict.

And here's the part that made me put my coffee down: the clinic it named first is not the one with the best reputation in town. Ask any local. There's a clinic everyone raves about — packed schedule, word-of-mouth machine, the one your doctor mentions by name. ChatGPT didn't pick that one.

It picked the one that was legible to a machine.

That gap — between the business humans love and the business the AI recommends — is the single most important shift in local marketing right now, and almost nobody is paying attention to it. So I spent the next week doing what I do for clients: I reverse-engineered the win. I ran the same question across ChatGPT, Perplexity, Gemini, and Google's AI Overviews. I pulled up every clinic's digital footprint side by side. I mapped what the winner had that the beloved-but-invisible clinic didn't.

This is what I found. It's not what most people assume.

Why This Isn't a Niche Problem

Before the "nobody searches like that" crowd sharpens their knives — they do now.

According to AIrecommend.ai's State of AI Search 2026 research, 37% of consumers now begin a purchase or local-service decision with an AI tool rather than a search engine or a map app. That's not a fringe behavior anymore; that's more than a third of your potential customers, and it skews toward exactly the people you want — higher-intent, higher-income, decisive.

And the kicker: between 58% and 68% of those AI queries end in zero clicks. The person reads the answer, absorbs the recommendation, and acts on it — books, calls, drives over — without ever visiting a website or comparing a list. The AI didn't send them to a shortlist. It was the shortlist. In the old world you competed to rank. In this world you compete to be the sentence.

When the machine names one business, second place is invisible. There is no scroll. There is no page two. There's the answer, and there's everyone the answer forgot.

So the question I was really asking on that Sunday wasn't "which clinic is best." It was: what makes a language model confident enough to say your name out loud?

Reverse-Engineering the Winner

Here's what I expected to find: the winner had the most reviews, or the slickest website, or was paying for something. Wrong on all three. The beloved clinic actually had more Google reviews and a higher star average.

Here's what the winner actually had. Four things, in order of how much they mattered.

1. The machine knew exactly what it was — entity clarity

The winning clinic was unambiguous about its own identity, and it was unambiguous everywhere. Its name, its category ("outpatient physical therapy clinic"), its specialties (post-surgical rehab, sports injury, vestibular therapy), and its service area were stated the same way on its website, its Google Business Profile, its Yelp page, its Healthgrades listing, and its own "About" copy.

The beloved clinic? On its website it called itself a "wellness and movement studio." On Google it was a "Physical therapist." On another directory it was tagged "Massage." Warm, human, on-brand — and to a language model, a blur. The AI couldn't confidently answer "is this a PT clinic?" because the business itself never gave one answer.

This is the concept most people miss: AI doesn't rank pages, it resolves entities. It's trying to build a confident internal model of what you are. Every inconsistency is a reason to doubt — and doubt is what gets you left out of the sentence. The winner made itself the easiest entity in town to be certain about.

2. It said the same thing everywhere — citation consistency

I pulled the NAP — Name, Address, Phone — for all eleven clinics across a dozen sources. The winner was pixel-identical everywhere: same suite number, same phone format, same legal name. The beloved clinic had three different phone numbers floating around the web (an old one from a prior location still lived on two directories) and two spellings of its own street.

To a human, that's a shrug. To a retrieval system deciding whether two mentions refer to the same business, it's the difference between one strong, corroborated entity and three weak, fragmented ghosts. AI systems reward corroboration. When five independent sources agree on who you are, the model treats that as truth. When they conflict, it hedges — and hedging means it reaches for the competitor it can be sure about.

Consistency isn't hygiene. It's a confidence signal, and confidence is the whole game.

3. Its reviews said words, not just stars — review substance

This is the finding that surprised me most, so sit with it: the star rating barely mattered. The language of the reviews mattered enormously.

The beloved clinic's reviews were a wall of five-star warmth: "Love this place!" "Best team ever!" "Highly recommend!" Genuine, glowing — and semantically empty. The winner's reviews were full of nouns and specifics: "fixed my rotator cuff after surgery," "great for ACL rehab," "helped my vertigo," "got me running again after a stress fracture."

When someone asks an AI for "a clinic that's good with post-surgical knees," the model isn't counting stars. It's matching the meaning of the question against the meaning of everything it knows about each option. The winner's reviews were dense with exactly the vocabulary that maps to real patient questions. Its customers had, without knowing it, written its AEO copy for it — and the clinic had quietly prompted them to, by asking departing patients what specifically they'd been treated for.

Stars get you in the room. Substance gets you named.

4. It handed the machine the facts — structured data

Finally, the boring one that punches above its weight. The winner's site had proper LocalBusiness schema markup: name, geo-coordinates, hours, service list, all machine-readable. It had a genuine FAQ section written in plain "question → direct answer" format — the exact shape AI systems love to lift. Its service pages were headed with real questions ("What does vestibular therapy treat?") and answered them in the first two sentences, before any marketing throat-clearing.

The beloved clinic buried its answers in beautiful, meandering brand prose. Lovely to read. Miserable to extract. When a model has to guess what you offer, it discounts you. When you hand it clean, structured, unambiguous facts, you're not asking to be trusted — you're making trust the path of least resistance.

What This Actually Means

Sit with the through-line, because it reframes everything.

The clinic that won didn't win by being better. It won by being legible, consistent, corroborated, and specific — by making it easy and safe for an algorithm to stake its reputation on saying its name. The beloved clinic was better in every way that a human notices and invisible in every way a machine reads.

That is the entire discipline of Answer Engine Optimization in one sentence: you are no longer optimizing to be found. You are optimizing to be chosen, and to be repeatable. The AI has to be able to name you confidently and defend that choice with evidence it can point to.

The encouraging news is that this is buildable. It is not luck, budget, or being the oldest name in town. When we take a client through the full AEO framework at AIrecommend.ai — entity resolution, citation cleanup, review-language strategy, and structured data — the pattern in our State of AI Search 2026 data is consistent: 3.8× more AI citations and roughly 142% more AI-referred leads within 90 days. Not because the business changed. Because its legibility did.

The beloved clinic could overtake the winner in a quarter. It has the reputation. It just never translated that reputation into a form a machine can read.

The Playbook

Here's exactly what I'd do, in order, if I ran a local business tomorrow and wanted to be the name the AI says.

1. Run the experiment on yourself, tonight. Open ChatGPT, Perplexity, and Gemini and ask, as a customer would, for a recommendation in your category and town. Note who gets named and who's invisible. This is your real scoreboard now — treat it like one.

2. Nail your entity. Pick one clear category and one consistent description of what you do and who you serve. Deploy that identical language across your website, Google Business Profile, and every major directory. Kill the clever, blurry self-labels. Be boringly clear.

3. Clean up your citations. Audit your Name, Address, and Phone across every listing you can find. Make them pixel-identical everywhere, and hunt down zombie listings from old locations or numbers. Corroboration is a ranking signal for machines.

4. Engineer your reviews for meaning, not just count. Prompt happy customers to name the specific problem you solved — the procedure, the condition, the outcome. A hundred substance-rich reviews out-recommend five hundred "love it!" reviews.

5. Feed the machine structured facts. Add LocalBusiness and FAQ schema. Write real question-and-answer content that resolves the query in the first two sentences. Put the extractable answer before the brand poetry, not after it.

6. Re-run the experiment every month. AI answers drift as the model's confidence shifts. Track whether you're getting named, in which tools, and for which questions — and close the gaps you find. This is a system, not a project.

I went looking for a curiosity on a Sunday morning and found the future of how customers choose. The businesses that win the next decade of local search won't be the ones with the best story. They'll be the ones a machine can tell that story for — confidently, consistently, out loud, by name.

Go ask the machine about your town. Then go make sure it can say yours.

About the author

Scott Tischler is the Founder & Chairman of AIrecommend.ai and a practitioner-authority on AI search and Answer Engine Optimization. With 20+ years in marketing technology — including American Express, MetLife, and UBS — and executive study at Wharton, Harvard, Yale, and Oxford, he helps businesses become the ones AI recommends.