Restaurant Website Design in the Cayman Islands: Why ChatGPT Cannot Read Your Menu

A tourist standing on Seven Mile Beach at 6pm does not open Google, scroll a results page and compare ten restaurants. They open ChatGPT, or the AI answer that now sits above Google’s results, and they ask a question: where should we eat tonight, we have kids, we want somewhere close.

Whatever comes back is the shortlist. If your restaurant is not in it, you were never in the running — and nothing on your website told you that happened.

This is not a prediction about the future of search. It is how a meaningful share of dinner decisions on this island already get made, and the thing that decides whether you are in the answer is almost never the thing restaurant owners are told to worry about.

The two-second test you can run right now

Open your restaurant’s website on your phone. Go to the menu. Try to select the text of a dish with your finger, the way you would to copy it.

If the words highlight, they are text. If nothing happens — if your finger just drags the whole page around — your menu is a picture, or a PDF, and to a machine it is an empty box.

That is the whole test. It takes two seconds and it is the single most predictive thing about whether an assistant can recommend you, because everything else depends on it.

Why a PDF menu is invisible

A PDF feels like a document because it has words in it. To a crawler it is an attachment sitting next to your website, not part of it. Assistants read the text of a page. They do not open your attachments, and they do not reliably read words baked into a JPEG.

So when someone asks which restaurants on Seven Mile Beach do gluten-free pasta, the assistant is matching that question against text it can actually see. Your gluten-free pasta may be item fourteen on page two of a beautifully designed PDF. It does not exist.

The same applies to the designed graphic your agency exported for Instagram and then dropped onto the website as the menu. It looks correct to every human who visits. It is blank to every machine that does.

This is the part that catches owners off guard: the site looks finished. Nothing is broken, nothing errors, the photos are gorgeous. The failure is silent, and the only symptom is a phone that rings less than it should.

What structured data actually means for a restaurant

Making the menu readable is the floor. The next step is telling a machine what it is looking at, in a format built for exactly that: structured data, sometimes called Schema markup.

Plain text says Kung Pao Chicken — $24. Structured data says: this is a MenuItem, it belongs to a MenuSection called Mains, the price is 24 in KYD, it is served by a Restaurant at this exact address, at these coordinates, open these hours.

The difference matters because the assistant no longer has to guess. Without it, a model reading your page has to infer that the number next to a dish is a price, that the address in your footer is where the food is, and that “Sun–Thu 5–10” means you are open tonight. Sometimes it guesses right. When a competitor down the road has stated all of it explicitly, guessing is what loses.

The pieces that carry the most weight for a restaurant, in order:

• Restaurant with a full postal address and geo coordinates. “Grand Cayman” is not a location to a machine. A latitude and longitude is.
• OpeningHoursSpecification. Machine-readable hours, including the seasonal ones. This is what answers “is it open now”, which is most of the question when someone is hungry.
• Menu, MenuSection and MenuItem. Your actual food, itemised, with prices. This is what gets quoted back to somebody asking about a dish.
• servesCuisine. Trivially small, and it is what puts you in the running when the question is “Thai food near Camana Bay” rather than your restaurant’s name.

What that actually looks like in the page

Structured data is invisible to visitors — it sits in the source of the page, not on the screen. Here is a single dish, marked up properly. If you have a developer, this is the shape of the thing to ask them for:

{
  "@context": "https://schema.org",
  "@type": "MenuItem",
  "name": "Kung Pao Chicken",
  "description": "Wok-fired chicken, peanuts, dried chilli, Sichuan pepper.",
  "offers": {
    "@type": "Offer",
    "price": "24.00",
    "priceCurrency": "KYD"
  },
  "suitableForDiet": "https://schema.org/GlutenFreeDiet"
}

Nine lines. That is the difference between a dish a machine can quote and a dish it has to guess at. Note the last line in particular: suitableForDiet is how you get surfaced for “gluten free near me” without writing a single blog post about gluten.

The contradiction that costs you the table

Here is the failure we see most often on this island, and it has nothing to do with your website’s design.

Your website says you open at 5pm. Your Google Business Profile says 5:30. A delivery platform has last winter’s hours. TripAdvisor has something else again.

Faced with four versions of the truth, an assistant does not pick the most likely one. It does what any careful system does with an unreliable source: it lowers its confidence and recommends somebody it is more certain about. You are not penalised for being wrong. You are skipped for being unclear.

Fixing this costs nothing and takes an afternoon. It is also the single most common thing left undone, because it is nobody’s job — the website was built by one person, the Google profile was claimed by another, and the delivery listing was set up by whoever was on shift that day.

What we found across six Cayman restaurant sites

We build and maintain restaurant and bar websites here, so rather than describe this in the abstract, here is what is actually deployed on the six live ones: Uncle Liu, Coconut Room, San Si Wu, Carnivore Smash Burger, Yallah and 19-81 Brewing Co.

All six declare their venue type in structured data — Restaurant, FastFoodRestaurant or Brewery — each with a full postal address, geo coordinates and machine-readable opening hours. Four of the six publish their menu as itemised structured data with prices, as page content rather than as a download. The other two are bars where the drinks list changes weekly, and we deliberately did not pretend a fixed menu exists.

That last detail matters more than it looks. Structured data has to describe what is true. Markup claiming a menu that is not really on the page is worse than no markup, because it is a machine-readable statement that turns out to be false, and that is exactly the unreliability you were trying to fix.

In a session we recorded on 21 August 2026, we asked ChatGPT where to eat on Seven Mile Beach and it named Uncle Liu and Coconut Room — two sister venues that share an owner and, more to the point, share the same setup: menus as page content, hours identical across the website and the Google profile, and both digital profiles kept in sync so an assistant never has to work out which is which.

We want to be precise about what that is and is not. It is one recorded session on one date. AI assistants are not deterministic — the same question, asked differently, from a different place, on a different day, returns different answers, and no agency on this island can promise you a fixed position in them. What we can say is that the structure underneath those two restaurants is the structure an assistant can read, and that on the day we asked, it read it.

What to do this week

None of the following requires a new website. In order of how much they matter:

1. Run the two-second test. If your menu does not highlight, that is your whole problem and everything else is secondary.
2. Republish the menu as page content. Real headings, real dish names, real prices as text. Keep the PDF for printing if you like — just stop making it the only copy.
3. Reconcile your hours. Website, Google Business Profile, Apple Maps, delivery platforms, Instagram bio. One version, everywhere. Set the special hours before the holiday, not after.
4. Check the primary category on your Google profile. It decides which searches you are eligible for at all, and it is the field most owners set once in 2019 and never revisit.
5. Then add structured data. Last, because marking up a menu that is not on the page is marking up nothing.

Steps one through four are free and you can do them yourself. If you would rather see where your site stands before deciding anything, our free visibility checker reads any website the way an assistant does and tells you in plain English what it could and could not find — no signup, about a minute.

And if you want the whole thing built properly, that is what we do, from George Town, for restaurants on this island.

Questions we get asked

Do I need a new website for this?

Usually not. If your site is on WordPress, Squarespace or Wix and loads reasonably fast, the menu can be republished as page content and the structured data added to what you already have. A rebuild is worth it when the site is slow on mobile, when the menu lives inside a booking widget you cannot edit, or when nobody has the login any more — which is more common than you would think.

My menu changes every week. Do I have to update the markup every time?

No, and if the answer were yes this would not be worth doing. The markup should be generated from the same content you edit, so changing a price in the page changes it everywhere at once. If your developer proposes maintaining the menu in two places, that is a design you will abandon within a month.

Is this the same as SEO?

It overlaps, and the foundations are shared. Traditional SEO is about ranking in a list of blue links. What this article describes is about being readable and quotable when there is no list — when an assistant answers the question directly and names two or three places. The same work feeds both, which is why we do not sell them as separate products.

How long before it makes a difference?

The menu and the hours are read as soon as the page is crawled again, which is usually days rather than months. Whether that turns into more covers depends on things no markup controls: your reviews, your photos, your location, your food. Structured data does not make you the best restaurant on the beach. It makes you legible to the thing being asked which restaurant to pick, which is a prerequisite, not a guarantee.

Does any of this help with Google, or only with ChatGPT?

Both, and the same markup serves both. Google has used structured data for rich results — the star ratings, the price ranges, the opening hours in the panel — for years, and its AI answers draw on the same signals. There is no version of this work that helps assistants and hurts your Google ranking.