Can an AI Actually Read Your Restaurant's Website?

Somebody three blocks from your restaurant just asked an assistant where to get dinner. Whether your name comes up has very little to do with how good your website looks.

It depends on something almost no owner has checked: whether an automated reader can fetch your pages at all.

This is not the same problem as Google ranking, and the difference matters. A search engine that finds your site mediocre still lists you on page three. An assistant that cannot read your site has nothing to say about you — there is no page three. You are either in the answer or you are not in it.

Here is how to check yours. It takes about two minutes and costs nothing.

The two-minute test

1. Open yoursite.com/robots.txt in a browser.

That file tells automated readers what they may look at. Three outcomes:

• It loads and says Allow: / — good, keep going.
• It names crawlers and blocks them. Look for GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot or Google-Extended followed by Disallow: /. That is an explicit instruction to stay out.
• It does not load at all. If a plain text file meant for robots cannot be fetched by one, that is worth asking your platform about.

2. Try to select your menu text with the mouse.

If you can highlight the item names, a reader can parse them. If you cannot — because the menu is an image or a PDF — then your prices and dishes are effectively invisible, not only to assistants but to screen readers and to ordinary search.

3. Ask an assistant about your restaurant.

"What are the hours for [your name] in [your city]? What's on the menu?" Wrong answers are informative. Confident wrong answers are the ones that cost you covers, because the guest has no reason to doubt them.

What we found

We ran the fetch step against two restaurant website platforms on August 6, 2026. Verbatim results:

The Popmenu result is the interesting one: the file whose entire purpose is to tell robots what they may do could not be retrieved by a robot.

Now the part most competitive posts would skip. That 403 does not prove Popmenu blocks AI crawlers. Generic bot protection — the kind that stops scrapers and card-testing — routinely returns 403 to unfamiliar clients while still allowing named crawlers like GPTBot through an allowlist. We could not confirm either way, precisely because we could not read the file that would have told us.

What it does establish is narrower and still worth knowing: one ordinary automated reader, on that date, got nothing. If your site is on a platform, that is a question worth putting to your account manager in writing.

We would rather publish the narrow true version than the wide one that sounds better.

What good looks like

A site an assistant can actually use has four properties:

1. A readable robots.txt that names the AI crawlers and allows them, rather than leaving it to a wildcard nobody has reviewed.
2. The menu as text, in the page. Not a PDF. Not a photograph of a chalkboard.
3. Hours, address and phone in the markup, not baked into a hero image.
4. A machine-readable endpoint — structured data, or an API that returns the menu as data. This is the difference between an assistant guessing from prose and quoting your actual prices.

The first three are hygiene. The fourth is where most restaurant websites stop, and it is the one that decides whether an agent can order, not merely describe.

Why we care about this particular problem

OpenPOS was built around the assumption that software would stop being the only thing reading a restaurant's data.

Our robots.txt names ten AI crawlers and allows them explicitly. Store data — menu, hours, location — is public by default, and an agent can read a menu without a key. There are 41 tools exposed over the Model Context Protocol at /api/mcp, covering the menu, inventory, reservations and orders, so an assistant can do more than paraphrase a page.

We are not the only ones moving this way, and we will not claim otherwise — several vendors have opened ordering to agents, and that is genuine progress. The distinction we would draw is between exposing the checkout and exposing the operation.

But you do not need to take our word for any of it, which is rather the point of a test you can run yourself. Open your own robots.txt first. Whatever platform you are on, you should know what it says.