Earlier this year we measured 36 Dominican hotel websites and found that the small independent hotels outperformed the international resort chains by more than two to one. That wasn't our hypothesis. We had assumed the opposite — that corporate marketing budgets would produce better sites — and we published the finding that proved us wrong, in the first section, under a heading that says so.

Then we checked 64 restaurants and found that 84% have no working online menu. Then we ran 208 AI queries across four assistants and found that the correlation we were testing for reversed in one category, for reasons that undermined the simple version of our argument.

We're a small web studio in Punta Cana. Running original research is not an obvious use of our time, and several people have asked why we bother. Here's the honest answer.

Because nobody else has these numbers

If you want to know how fast Dominican hotel websites load, or how many restaurants here publish a readable menu, or which local businesses AI assistants actually name, there is no source. The international studies don't cover this market. The local conversation is anecdote.

That's not a gap we found and cleverly exploited. It's just a gap. Anyone could run these studies — the tools are free and public, and we publish the method every time precisely so they can. We were simply the ones who sat down and did it.

The result is that when we tell a hotel its website is slow, we're not offering an opinion. We can say that the median property in our sample took 17 seconds to show its main content, that 86% scored under 50 out of 100, and that the fastest site we measured belonged to a small independent in Bayahíbe. That's a different conversation from "your site could be faster."

Because the research disciplines us

This is the part we didn't anticipate, and it's turned out to matter more than the marketing.

Before the hotel study, we would have told you that Dominican hotels have a bilingual problem. We'd have said it confidently, because it's true of most of the market. Then we measured it: 97% had a real English version, 92% had Spanish, and not one used a translate widget. The hotel sector has solved multilingual publishing. We were wrong, and we'd have gone on being wrong — and giving that advice to hotel clients — if we hadn't checked.

The restaurant study corrected us again. We expected the story to be "restaurants don't have websites." The actual finding was sharper: nearly all of them have a Google listing, and roughly one in six of those listings points at something broken — a deleted Drive file, an expired domain parked for sale, a 2020 flipbook. The highest-value fix turned out to be free and to have nothing to do with hiring us.

The AI study did it a third time. We tested whether businesses with websites get recommended more. They do — except in one category where nobody has a website at all, where the correlation flips completely. We published that too, because the exception is what explains the mechanism.

Three studies, three times our own assumptions were corrected. We'd rather find that out from data than from a client six months into a project.

Because publishing the inconvenient parts is the whole point

A study that only produces findings flattering to the person who commissioned it isn't a study. So we've made some rules, and they cost us things.

We publish the limitations. Every study carries a section saying what it doesn't show. The restaurant study can't claim anything about revenue, because we didn't measure revenue. The AI study's sample is curated rather than random, which makes the findings conservative, and we say so.

We publish when the instrument wobbles. In the hotel study, one re-run of a resort's homepage returned a perfect score of 100, a load time of 0.8 seconds and a total page weight of 0.0 MB — because nothing had loaded at all. The tool had measured an empty response and scored it as the fastest website in the world. If we'd run that test once, on that attempt, we'd have published it as a great result. We published the failure instead.

We publish against ourselves. In the AI study, our own business appeared in the dataset, ranked first on one Google surface. So we re-ran it with personalisation switched off, watched ourselves drop to fifth, and printed that. We also printed that ChatGPT has never named us once across 36 company mentions in two languages.

None of those disclosures help us sell anything. They're the reason the other numbers are worth reading.

Because we name the businesses doing it well, and not the ones doing it badly

This is a deliberate editorial rule and we've applied it in every study.

When we found the best-performing hotel website in the country, we named it: Hotel Villa Iguana in Bayahíbe, 84 out of 100, re-measured twice to be sure. When we found ten restaurants publishing genuinely excellent menus, we named all ten.

When we found a resort scoring in the teens, or a beloved restaurant whose website is a single image reading "under construction," we described it without naming it.

The argument is the same either way — "the lowest-scoring large resort managed 17 out of 100" carries exactly the force of naming them — and the difference is that one version makes an enemy of a business in a small market, and the other doesn't. These are mostly small independent companies who did nothing wrong except hire someone who didn't come back. Using them as negative examples in an article that sells web development would be predatory, and we'd deserve the reputation it earned us.

There's a second reason, less noble but true: a study that names and shames invites a dispute, and a dispute is a story about us rather than about the finding.

Because the method is more useful than the findings

Every study we publish comes with the full method, the sample, the tools and the dates. The hotel study can be reproduced by anyone with a free Google API key — the measurements come from PageSpeed Insights, which is public and which anyone can point at any website, including ours. The restaurant study needs a browser and an afternoon. The AI study needs nothing but a laptop and the patience to ask the same question twice.

We publish the method because findings expire. Our hotel numbers describe one day in September 2026; the AI numbers will be wrong within months, because those systems change constantly. What doesn't expire is how to check — and a Dominican business that learns to measure its own site, or to search its own name across four assistants, is better off than one that read our number and believed it.

We'd genuinely rather someone re-ran our hotel study in six months and corrected us than cited our figure in 2028 as though it were still true.

What it costs, honestly

Time, mostly. The restaurant study was three hours of manual browser checking before any writing started. The AI study was 208 individual queries in fresh sessions, each recorded by hand. Nobody pays for that.

It also costs us some sales arguments. It would be easier to tell every restaurant they need a new website; the data says the first thing many of them need is a free thirty-minute fix to a Google listing, and we publish that instead. It would be easier to imply structured data lifts rankings; our own measurements show the fastest hotel site in the country carries none at all.

We think that trade is worth making. An informed client is a better client — they choose well, they know what they're paying for, and they don't spend the project wondering whether they were oversold.

What's next

More of it, and ideally repeated. The hotel study is worth re-running every six months, because the second measurement is more valuable than the first and by the third year it's a time series nobody else has. The AI study needs re-running as those systems change. And the study we most want to do is the one that gets past correlation: take a business with no website, record what the assistants say about it, build the site, wait ninety days, and ask again.

If you'd like to see the studies, they're all on our blog, with the data and the method attached. If you'd like to argue with the numbers, we'd welcome that — and if you find an error, tell us and we'll correct it in public, which is the only way any of this works.

At DR Web Studio we build fast, bilingual websites for Dominican businesses. If you'd like us to measure yours before anyone talks about building anything, contact us for a free consultation.