What AI actually reads on a restaurant website

A restaurant hires a designer. The designer does good work: a wide hero photo, the menu set beautifully, hours in the footer, a booking button that feels right. The owner loves it. It photographs well. It cost real money.

Then someone asks an AI where to eat tonight, and the restaurant isn’t in the answer.

The menu is a picture of a menu

Here is the part that surprises people: nothing went wrong. The site is not broken. It is just that most of what a restaurant site says, it says in pixels.

The menu is an image. The hours are baked into a graphic in the footer. The “about” paragraph is set in a background image because it looked better that way. A person reads all of it without noticing. A crawler reads none of it.

Strip a typical restaurant site down to the text a machine can actually extract and you often get something like this:

Olive & Ash. Menu. Visit. Reserve a table. Seasonal cooking, five nights a week. Tuesday to Sunday from five. 14 Fairview Lane. Follow us on Instagram.

Twenty-five words. That is the whole restaurant, as far as an AI is concerned. No dishes. No dietary options. No private-dining room. No parking note. None of the things people actually ask about.

What people actually ask

The questions that send a table your way are specific, and they are almost never the words on your hero image:

• “best restaurant for a birthday dinner near me”
• “where to eat with vegetarian options in Fairview”
• “does Olive & Ash take reservations”
• “restaurants with private dining for 20 people”
• “somewhere open late on a Tuesday”

An AI answers those by finding a source that states the answer plainly. If your site never states it in text, you are not a candidate, no matter how good the photography is.

The fix is boring, and it works

There is no trick here. You make the answers legible:

1. Put the menu in HTML. Real headings, real dish names, real prices. Keep the beautiful PDF for print; the web version needs to be text.
2. Answer the recurring questions on their own pages. Reservations, private events, dietary options, parking, opening hours. One page each, titled the way people ask.
3. Add structured data. Restaurant schema with cuisine, address, opening hours, and whether you take bookings. It costs nothing and removes ambiguity.
4. Write the first sentence as the answer. Not “Welcome to our restaurant” but “Olive & Ash is a 34-seat neighbourhood restaurant on Fairview Lane, open Tuesday to Sunday from 5pm.”

Same craft, now legible

None of this makes the site uglier. The design stays. The photography stays. You are not trading beauty for visibility. You are adding the layer that was always missing, the one written for the reader that has no eyes.

That is the whole job: build it twice. Once for the person looking at it, once for the machine reading it.