Will ChatGPT or other AI assistants recommend my restaurant?

How the answer gets built

When someone asks an assistant where to eat, roughly two things happen. The model draws on what it absorbed in training, and — increasingly — it retrieves live pages to check current facts.

Both routes depend on the same thing: your information existing as text somewhere a machine can read.

There is no submission form, no paid placement, and no service that can put you in the index. Anyone selling that is selling something else.

Why most restaurants are invisible to it

Not through neglect. Through three specific and very common choices:

The menu is a PDF or a photograph. No readable text means nothing to cite. This one disqualifies more restaurants than everything else combined.

The details live only on social. Hours in a bio, specials in stories that expire, the address in a caption. Largely closed to the open web, and expiring content cannot be a durable fact.

Nothing on the site answers a specific question. A homepage with photographs and a phone number answers nothing. Assistants return specifics — "yes, they have a set lunch on weekdays" — and they can only do that if someone wrote it down.

What actually improves your odds

Answer real questions on real pages. The questions you are asked twice a week are the ones being typed into assistants. Each deserves a page, or at least a clearly headed section, in plain language.

Put the facts in text, in one obvious place. Hours, address, phone, cuisine, price range, whether you take bookings, whether there is parking, what the dietary options are. Boring, and it is precisely the set an assistant needs.

Keep your Google Business Profile complete and consistent. It is a major source for local answers. Details that disagree between your site and your listing reduce confidence in both.

Be quotable. Write the answer as a self-contained sentence rather than a build-up. "We serve a set lunch Monday to Friday, 12 to 3" can be lifted directly. "Come and discover our exciting midday offering" cannot.

What does not work

Keyword stuffing. Assistants are generating language, not counting words. Repetition reads as noise.

Claiming things you cannot support. "The best pizza in the city" is not a fact anything can verify or repeat.

Buying your way in. There is nothing to buy. Be suspicious of anyone who says otherwise.

The honest expectation

This is not a switch you flip. It is the accumulation of being legible: readable pages, consistent details, real answers to real questions. The same work that makes you findable in search makes you citable by an assistant, which is convenient — there is one job here, not two.

And the first step is almost always the same one: if your menu is a picture, nothing else on this list can help you yet.

Making the first step a non-event

This article ends on the right note: if your menu is a picture, nothing else helps yet. Worth adding how that actually gets fixed, because "make the menu text" is easy to agree with and easy to never do.

Menus stay as images not out of ignorance but because the photograph is the only version the restaurant can update themselves. The designer sends a PDF, someone screenshots it, and up it goes. Text on the page would be better — but text on the page has historically meant a developer, and a developer means a wait, so the picture wins every time there is a price change.

That is a tooling problem wearing an SEO problem's clothes. On Helm a menu is a list you edit: each item with a name, a description, a price and a photograph, in both languages if you run in both. Adding a dish takes about thirty seconds.

The consequence is not that the menu is text once. It is that the menu is still text the week after a price change, which is the only version of this that actually works.