Skip to content

How restaurants get cited by AI search

ChatGPT and AI Overviews already recommend places to eat. Here is what their crawlers read, what they skip, and the pages you need to be quoted.

by The Menugram team

2 min read

Copy link
The Menugram analytics dashboard showing revenue, orders and daily trends

People increasingly do not search for "restaurant in Shoreditch". They ask: "where should I eat in Shoreditch with a decent vegetarian menu and outdoor seating". The answer comes back as a paragraph with two or three links, not a list of ten.

That is a different game. A search engine ranks sources; an assistant writes an answer and then decides whom to credit. If it has not read you, it cannot mention you.

What an AI crawler actually reads

Briefly: HTML. That is all.

Googlebot executes JavaScript — late, but it does. The crawlers behind most AI assistants do not. They request the URL, take the response, and move on. If your site paints its content in the browser, what they saw was an empty page.

So the first question is not which keywords to use. It is whether the words are in the server's response at all.

Why short sentences win

An assistant assembles an answer out of fragments. A fragment that is true on its own is safe to quote. A fragment that depends on the paragraph above it is not.

Compare:

  • "As we saw above, this applies to the lunch menu too."
  • "The lunch menu at [the restaurant] is served from noon to 4pm on weekdays."

The second can be lifted with nothing around it. The first cannot.

In practice: phrase the heading as a question, and answer it in the first sentence underneath. Everything after that is detail for the people who want it.

The three files that cost an afternoon

  1. A robots.txt that lets the right crawlers in. GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot, Google-Extended. A blocked agent does not stop answering questions about your neighbourhood — it just answers them without you.
  2. llms.txt. A short markdown map written for an agent: what this place is, which pages matter, how to cite it. The convention is new; the file is fifteen lines.
  3. FAQPage and Restaurant markup. The questions and answers you already wrote for humans become machine-readable without writing anything new.

And the dullest thing, which works

Put an author's name and a last-updated date on the page. Visible, in the body, not only in a meta tag.

A person and a model, each facing two equally good sources, both take the more recent one with a name under it. That is not a trick — it is the same judgement you would make.

We write about this regularly. The rest of the posts are here.

If you are still assembling the foundations, start with your menu — it is the page an assistant will quote first.

Frequently asked

Should I block the AI crawlers?

If you sell content, perhaps. If you sell dinners, no. A blocked crawler does not stop existing; it just stops knowing about you, and the guest asks the assistant regardless.

What is the difference between SEO and optimising for AI?

Most of it overlaps — clean HTML, clear structure, a fast page. The difference is that a search engine ranks links while an assistant composes an answer. That is why short, self-contained statements are rewarded and long wind-ups are not.

How do I check whether they can see me?

Load your page with JavaScript disabled, or request it with curl. Whatever survives is roughly what a crawler that does not run scripts gets to read.

Sources

  1. 01llms.txt — the proposed site map for LLMs
  2. 02Google Search Central — Overview of Google crawlers
  3. 03Schema.org — FAQPage

Menugram

See what your restaurant looks like on Menugram

A QR menu, table ordering, payments and an owner dashboard — with your own brand and your own domain.