A car rental company in Punta Cana with nine Google reviews was recommended by Google's AI ahead of a competitor with seventeen hundred. ChatGPT, independently, put the same nine-review company at the top of its shortlist — and it had no business listing attached at all. Both systems reached it the same way: they read its website, which publishes rates, insurance terms and delivery options in plain text.

We found that while running 208 AI queries about Dominican businesses across four platforms. Between them the assistants named 369 companies, and we verified every one. If your business has never turned up in an AI answer, the data says something fairly specific about why — and it isn't what most people assume.

It isn't your reviews

That's the first assumption to discard, because it depends entirely on what you sell.

In hotels, review volume dominates completely. Across eight runs asking four platforms for the best hotels in Punta Cana, not one independent hotel was named, and nothing with fewer than 465 reviews appeared anywhere. The recommended properties had 63,000, 61,000 and 22,000 reviews. If you run a small hotel in that market, you are not competing on merit — you are below a threshold.

But move to Cabarete and ask the same kind of question, and every single name is an independent, several with fewer than 150 reviews. Google's AI Overview led with a small owner-run kitesurf hotel that has 140 reviews, and linked to its own website twice.

And in services, review volume stops mattering almost entirely. On one web design query, Google's AI listed eight companies carrying 34, 39, 51, 49, 1, 10, 57 and 108 reviews — in that display order. A one-review business sat above a ten-review one and a 108-review firm came last. Whatever ordered that answer, it wasn't reviews.

So the honest version is: in mature, heavily-reviewed categories, review volume is a wall. Everywhere else, it's barely a factor.

It is, mostly, whether an AI can read you

Across the services portion of our study, businesses with their own website were named 2.52 times on average. Businesses without one were named 1.50 times — roughly a 40% penalty.

Here's the interesting part, and the reason we trust the finding: it has an exception, and the exception explains the mechanism.

When we first ran the numbers, the correlation reversed — businesses without websites appeared to do better. It was caused by four businesses: the veterinary clinics in Sosúa. Not one of them has a working website. One lost its domain entirely, one exists only in directories, one is Facebook-only. All three live ones were named six times each, by three different platforms.

The reason is simple. In a category where nobody has a website, the AI uses whatever exists — Facebook pages, phone numbers, community posts, directory scrapes. A website buys you nothing when your competitors don't have one either.

Strip that one category out and the pattern returns cleanly: 2.52 mentions with a site, 1.50 without. So the finding isn't "get a website and AI will find you." It's narrower and more useful:

In categories where your competitors have websites, not having one costs you roughly 40% of your mentions. In categories where nobody has one, it costs you nothing — yet. What a website changes in every category is what the AI is able to say about you, and where it sends people.

That second half matters more than the first. Being named is only half of it.

What the AI says about you comes off your own site

This was the most consistent behaviour across all 208 runs, and it cuts both ways.

ChatGPT, answering questions about Dominican web agencies, cited only businesses' own websites — four or five per run, no directories at all. It quoted their published prices back with attribution: "$399, and e-commerce from $699, according to its current site." In one Spanish answer it went further and organised the entire recommendation under a heading meaning "options with published prices." Publishing your rates wasn't just repeated — it became the structure of the answer.

Google did the same thing with descriptive copy. Where a business had a clear site, Google's AI quoted it — sometimes three different passages of the same site across three runs, reading it as a document rather than matching it as a listing.

The reverse is equally true. Businesses with thin data didn't survive. On one query, ChatGPT named two genuinely independent hotels that had no rating and no price attached to them. One run later, those were precisely the two that had dropped out of the answer. On another, an AI named seven hotels including two with no website; the next run named five, and the two that vanished were exactly those two.

That last observation is a single instance, and we won't oversell it — but it's the shape the whole dataset keeps taking. When a system needs to compress a list, the entries it can say least about are the ones that go.

Overclaim and it gets amplified

The flip side of "whatever you publish is what the AI says" is that AI does not audit you.

A dive centre in Bayahíbe describes itself on one page of its own site as the only PADI Career Development Centre in the Dominican Republic. PADI's own public register lists it a tier lower, and scopes the claim to the only such operator in Bayahíbe town. Google's AI repeated the business's version — promoted one tier, and expanded from one village to the whole country.

A law firm calls itself "the premier law firm in the Dominican Republic for immigration" on its website. Google put the word premier into its own opening sentence, as editorial judgement rather than quotation. The firm is genuinely distinguished — founded 1927, ten offices, Chambers-ranked — but it is ranked by Chambers in no immigration category, and its own site lists immigration as one of twenty-plus practice areas.

If you exaggerate on your website, an AI will restate the exaggeration in its own voice, to a customer who has no idea it originated with you.

Get your name right, or lose the click

One failure mode is worth its own warning because it hits small independents hardest.

A small kitesurf hotel in Cabarete was rendered five different ways across the study. Google's AI Overview got it right and linked its real domain. ChatGPT used a name the business doesn't use. Gemini called it "Agualinda" — twice, in two different wrong forms, with a letter inserted. No Dominican business is called Agualinda. Neither string would find the hotel in a search.

A thirteen-review guesthouse near Santo Domingo's airport was named "Hostal Doña Beba" in one run and "Doña Beba Guesthouse" in the next, minutes later — and in the second run its listing card broke entirely, showing no rating and no price, because the renamed entity no longer matched the underlying record.

Google's AI even read the domain future.do as a brand and invented a business called "Future Dose", then linked out under that name.

Across the hotel data, 40% of businesses had a name inaccuracy of some kind. The businesses this happens to are consistently the small ones whose names appear in few places, inconsistently. The defence is unglamorous: use one exact name, everywhere — your site, your Google listing, your social profiles, your directory entries — and make it the name a customer would actually type.

The practical order of work

Based on what the data actually showed, in order:

1. Claim and complete your Google Business Profile. It's free, and in our repeat testing Google's local pack was the only surface that didn't move at all — 100% stable across eleven captures, while every AI answer above it churned. It's also the layer that English-language AI answers draw on most heavily. Our thirty-minute audit covers it.

2. Put your actual information on a page as text. Services, prices if you publish them, locations, what's included. Not in a PDF, not in a designed image — as text an AI can read. This is what determines what gets said about you.

3. Publish prices if you can. In this dataset, publishing rates got them quoted accurately, repeatedly, across platforms — and on one platform it determined the shape of the answer.

4. Fix your name everywhere. One form, consistently, on every property you control.

5. Then worry about Spanish and English separately. They're not the same problem — asking Google in English returns business listings, while asking in Spanish returns website citations with no local layer at all. Genuinely bilingual pages are the only way to be in both.

What this study doesn't claim

We measured what AI systems said, and verified whether the businesses existed. We did not measure revenue, bookings or customers — nothing here says a mention is worth anything. Our sample was curated rather than random, weighted toward businesses well-known enough to appear in guides, which makes these findings conservative. It's a snapshot of two days in September 2026, and these systems change.

One correlation is not causation: businesses with websites tend to be larger, older and better documented in every other way too. What we can say is that the businesses AI could read most about were named most often, and the ones it could say least about were the first to disappear when a list got shorter.

If AI has never heard of you

Most of what's missing here is cheap to fix, and the highest-value item on the list costs nothing. At DR Web Studio we build sites that are fast, genuinely bilingual, and written as real text rather than pictures — with your Google listing wired to it properly. If you'd like us to check what the AI assistants currently say about your business, contact us for a free consultation and we'll look and tell you honestly what we find, including if the answer is that you don't need us.