Our own study

Which Miami businesses does AI recommend?

We asked 600 questions to ChatGPT, Gemini and Claude about local businesses in Miami — the way a customer who has never heard of you would ask. This is who shows up, how tightly shut each category is, and the full method so anyone can repeat it.

Promeseo · Miami2026-07-26

Why we ran this study

Everyone in marketing repeats the same handful of AI statistics, and almost all of them trace back to a blog quoting another blog. We did it ourselves once too: we published a figure we later had to pull because the trail ended in a vendor's sales deck.

The way out of that isn't quoting better. It's measuring. So we built the measurement, ran it, and published the data — including the two mistakes we found while doing it.

How we measured it

• 600 questions across three engines with web search on: ChatGPT, Gemini and Claude.
• Five categories: restaurants, med spas, dental clinics, law firms, home services.
• Five areas of Miami-Dade: Brickell, Kendall, Doral, Hialeah, Aventura.
• Two languages: English and Spanish, the same question in both.
• Phrased the way a customer asks — never naming a business in the question.

Finding 1 · The door isn't equally shut in every category

The practical read: in med spas, getting named is hard and staying named is durable. In law firms it's the opposite — easy to appear once, and appearing once means almost nothing. 132 of those 310 firms showed up a single time in 120 questions. That's noise, not presence.

Finding 2 · The three engines don't see the same Miami

Same questions, very different answers. Gemini named between 1.5 and 2 times more distinct businesses than ChatGPT. In med spas: Gemini named 131, Claude 71, ChatGPT 68.

ChatGPT is the most restrictive of the three and also the most used — which makes showing up there harder and worth more.

Finding 3 · Ask in Spanish and you get a different city

This is the one that matters most in Miami. Asking about med spas in Spanish surfaced 166 distinct businesses; in English, 106. Same city, same category, two different maps.

For a business serving both languages, that isn't trivia: it decides whether half your potential customers ever hear your name. And it means a measurement done only in English is telling you about half the market.

What to do with this

1. Check where you stand, in both languages. Ask each engine for your category in your area, without naming yourself, several times. Once proves nothing — models don't answer the same thing twice.
2. Work on what a model can actually read. Your Google listing, your reviews, your own pages and the third-party sources that mention you.
3. Get other sites to talk about you. A business only its own Instagram covers has one source. Models prefer the one with ten.
4. Measure again. The first report is the starting line, not the result.

The limits of this study, not in fine print

It's one snapshot of one month, in one county, in five categories. Models change their answers over time and between sessions. It doesn't measure whether being named brings customers — only who gets named. And we grouped business names by hand where the same place appeared written three different ways, which is a judgement call we've documented in the data.

The data, in full

We publish the summaries for all five categories: business ranking, concentration by area, cuts by engine and by language, the name groupings we made and what we discarded.

Download the data (JSON) — free to use with attribution, under CC BY 4.0.

The full study, with the area-by-area tables and the leader of each category, is published in Spanish: Índice de Visibilidad en IA de Miami.