We asked four AI assistants the same question twice. Once in English: "web design company Dominican Republic." Once in Spanish: "diseñador web República Dominicana."

Same industry. Same country. Same week. Same assistants, each run twice in a fresh session.

Across the sixteen runs, 43 distinct Dominican businesses were named. Six of them appeared in both languages.

Six. Out of forty-three.

What that actually means

If you are a Dominican business, you do not have one presence in AI search. You have two, and they barely know about each other.

The overlap between the English and Spanish answer sets was 14%. Put the other way: 86% of the businesses an assistant named in one language did not appear when the same question was asked in the other.

This is not a ranking difference. It is not that you placed third in English and seventh in Spanish. For most of the 43 businesses, it is the difference between existing and not existing, depending on which language the customer typed.

The two platforms that flipped completely

The aggregate number undersells it. Two of the four assistants did not shift their answers between languages — they inverted their behaviour entirely.

Google's AI Overview produced nothing in English. Both English runs returned no AI Overview at all for this query. Our field notes record it as the third consecutive capture of that query with none generated. Asked the same thing in Spanish, it produced an AI Overview both times, naming four businesses in the first run and five in the second.

Gemini did the opposite. In English it named five businesses in the first run and seven in the second. In Spanish it named zero, both times — it returned a generic answer about how to choose a web designer without naming a single company.

So for this query, in this week: if your customer was searching in English, Gemini could recommend you and Google's AI Overview could not. If they were searching in Spanish, the reverse. A business optimising for "AI visibility" as a single thing was optimising for a system that does not exist.

The two assistants that did answer in both languages still barely agreed with themselves. ChatGPT named 16 distinct businesses in English and 12 in Spanish, sharing 5. Google AI Mode named 8 in English and 7 in Spanish, sharing exactly 1.

Why this is specifically a Dominican problem

Most markets do not have this shape. The Dominican Republic does, for three reasons.

Your customers genuinely split by language. A hotel in Bávaro, a dive shop in Bayahibe, a dentist doing dental tourism, a lawyer handling residency applications — all of them serve Dominicans searching in Spanish and foreigners searching in English, for the same service, often in the same week.

AI Mode is now live here in both languages. Google's own supported-countries documentation lists the Dominican Republic and lists Spanish. This is not a future problem.

And Spanish is where the name errors concentrate. Across our whole 117-business sample, 25% of businesses had their name rendered inaccurately by at least one assistant. On the Spanish query specifically it was 5 of 13 — 38%. Accents, tildes and the ñ get dropped or guessed. The language with more of your customers is the language where machines are worse at spelling you.

What the six businesses in both sets had in common

This is the useful part of the finding, and it is a small sample, so treat it as a lead rather than a law.

The six that appeared in both languages were not the biggest agencies in the country. What they shared was being easy to resolve: a single consistent business name, a working website on their own domain, and a Google listing that matched both.

Of the six, four had a working website of their own. Two could not be verified to exist as trading businesses at all — a reminder that appearing in both languages is not the same as being real, and that these systems reward legibility rather than quality.

What to do about it

Four things, in order of how much they matter.

Publish in both languages properly, on separate URLs. Not a translate widget. Our hotel study found zero translate widgets across 36 sites, but 92% had distinct English and Spanish versions — the large operators already know this. If you have one Spanish page with a flag icon that machine-translates on click, there is nothing in English for an assistant to read. Here is how hreflang works.

Write your business name identically everywhere, accents included. Your Google listing, your website, your footer, your invoices. One spelling. This is the single cheapest fix on the list and it is where the 38% error rate comes from.

State your city and service in plain text in both languages. "Diseño web en Punta Cana" on the Spanish page and "Web design in Punta Cana" on the English one, as readable text rather than inside a designed graphic. Google's AI guidance is explicit that important content must be available in text form.

Check both languages when you check at all. And check more than once. Run-to-run overlap across our study averaged 0.63, and one platform swapped half its list between two runs minutes apart. A single reading tells you nothing — and a single reading in one language tells you less than half of nothing.

If you genuinely only serve one language

Some businesses do. A colmado in Santiago has no English-speaking customers, and a villa rental agency in Cap Cana may have almost no Spanish-speaking ones.

If that is you, the finding is simpler but not softer. It means the assistants that behave well in your customers' language are the only ones that matter, and the ones that go generic or produce nothing are irrelevant to you. In our pair, a Spanish-only business could ignore Gemini entirely for this query and should care a great deal about Google's AI Overview. An English-only business would need the reverse.

What you should not do is read a general article about "optimising for AI search," built on English-language testing in the United States, and assume it describes the system your customers are using. For this query, in Spanish, two of the four assistants behaved nothing like their English selves.

What this does not mean

It does not mean you should build two separate websites. One site with proper language versions is the answer, not two businesses.

It does not mean AI search is now your main channel. In our restaurant study, 98% of Dominican businesses had a Google Business Profile and 16% had a usable menu on their own site. The basics are still where the returns are, and we have put a number on that gap.

And it does not mean these results will hold. This is one query pair, four platforms, two runs each, in September 2026. Models get retrained and rollouts change. What we expect to persist is the shape of the problem — two languages, two answer sets, one business — not the specific numbers.

Method and disclosure

Eight queries were run across ChatGPT, Google AI Mode, Google AI Overview and Gemini, twice each, every run in a fresh session with the query typed verbatim and no follow-ups. The English and Spanish web design queries described here are one matched pair from that set. Business names were normalised for punctuation and legal suffixes before comparison, so "Example SRL" and "Example" count as one business. Every named business was then verified by hand.

DR Web Studio is one of the six businesses that appeared in both language sets. One of the eight queries was about Dominican web design companies, and our own business was named in it. We disclose this in every post built on this study and publish the counts so anyone can check them. Here is why we work this way.

If you want someone to look at whether your site actually exists in both languages — properly, on separate URLs, with a consistent name — DR Web Studio builds bilingual sites for businesses across the Dominican Republic. Contact us for a free consultation.