Data · survey run · 50 sites measured
AI readiness of restaurants in Amsterdam
The short answer
Of the 50 websites of restaurants in Amsterdam that we measured, 32 opened for an AI reader. Of those, 0 of 32 (0%) show prices as readable text, 5 of 32 (16%) show opening hours and 9 of 32 (28%) link to an online booking system.
Sample of 50 sites from OpenStreetMap entries with a website; not a random sample. Survey run 28 October 2026. How we measured.
The numbers
| Measure | Count | Share | 95% range |
|---|---|---|---|
| Website opened for an AI readerof measured sites | 32 of 50 | 64% | 50–76% |
| Prices readable as textof sites that opened | 0 of 32 | 0% | 0–11% |
| Opening hours readableof sites that opened | 5 of 32 | 16% | 7–32% |
| Links to an online booking systemof sites that opened | 9 of 32 | 28% | 16–45% |
| robots.txt turns away AI assistants or AI searchof sites with a readable robots.txt | 0 of 32 | 0% | 0–11% |
| Open, readable prices and a booking system, not blocking assistantsof sites that opened | 0 of 32 | 0% | 0–11% |
| Published rules an agent can read (deposit, cancellation)of sites that opened | 0 of 32 | 0% | 0–11% |
| Has an llms.txt file (not a rule set)of sites that opened | 4 of 32 | 13% | 5–28% |
| robots.txt blocks AI training crawlers (a legitimate choice)of sites with a readable robots.txt | 1 of 32 | 3% | 1–16% |
Shares are of the base named in each row. The 95% range reflects sampling noise inside this sample only; it cannot correct for businesses that are not in OpenStreetMap or have no website.
What this means for an owner
Between half and three quarters of the restaurants we tried in Amsterdam opened for our reader (32 of 50, 64%). The others may be out of date in the map data or may turn away automated visitors, so check that your own address opens cleanly over https.
None of the 32 readable sites showed a price in a form a program can read. An assistant asked "how much is it?" has nothing to quote. Menus that exist only as a PDF or a photo are not counted.
9 of 32 (28%) link to an online booking system, most often Zenchef (3 sites). A large share still leaves an assistant without a way to arrange a table reservation.
Opening hours are readable at 5 of 32 (16%) of the sites that opened.
None of the 32 sites with a readable robots.txt turns away AI assistants or AI search, so in this cell the barrier is missing information, not a locked door.
Compared with restaurants in Berlin, Amsterdam is ahead on a linked booking system: 28% here against 9% there. A sizeable gap, though both samples are small.
No site in this sample did all of it: opened, did not block assistants, showed prices and linked a booking system. An owner who does all four would be, in this sample, the first.
The three most common gaps
- Prices not readable as text: 32 of 32 sites that opened (100%). The assistant cannot quote a price, so it either guesses or recommends a business that published one. Prices that only exist in a photo, a PDF or a page built by script are invisible to simple readers.
- Opening hours not readable: 27 of 32 sites that opened (84%). Without hours an assistant cannot say whether you are open on Saturday, and customers get an answer from a directory that may be out of date.
- No online reservation system linked: 23 of 32 sites that opened (72%). An assistant that wants to arrange a table reservation finds no system it can reach from the website and falls back to "call them". Phone-only and email-only routes are not counted here.
Fix it yourself: a checklist that needs no tools from us
- Publish the menu as text on a page of your own site, with prices, not only as a PDF or a photo. Update it when the prices change.
- State your opening hours as text on the site, and keep them identical to your Google Business Profile. If your website tool offers it, add them as structured data (schema.org openingHoursSpecification).
- Add one reservation link (your booking system's public page) to the home page and your Google Business Profile.
- Open your own address in a private browser window and over a mobile connection. Fix certificate warnings, redirect loops and anything that makes the page slow. Ask your web host whether a firewall turns away unknown visitors.
- Write your terms in two or three sentences as text: deposit if any, how much notice you need to cancel, and what happens if someone is late. None of the sites in the survey published a rule set an agent could read, so this is cheap to get ahead on.
- An llms.txt file is optional and not a rule set. Add one only after the items above, and keep it short.
- Ask an assistant about your business in the words a customer would use, write down what it gets wrong, and fix the first thing. Repeat monthly.
Read before quoting
- Not a random sample: OpenStreetMap entries with a website tag over-represent businesses that have a website and that someone mapped. Real availability to assistants is likely worse.
- "Readable" means readable by code that does not run a browser. A more capable agent may see more; these figures are a lower bound for that agent and a fair measure of a simple one.
- "Did not open" is not "closed". We did not retry, on purpose.
- One city, one snapshot, 50 sites. Do not generalise to a country.
Where snuepi comes in
Disclosure: snuepi is the author of this survey and of the free check. Read this section as an advertisement.
The free check runs the same reading on your own website, with no sign-up, and tells you what an assistant sees at your business. You can fix everything on the list above without us.
How to cite
snuepi (2026). AI readiness of restaurants in Amsterdam. Survey run 2026-10-28, 50 websites measured. https://snuepi.com/data/restaurants-amsterdam. Licence CC BY 4.0. Sampling frame: © OpenStreetMap contributors (ODbL).
Data: results.json · results.csv · all cities and trades.
Map data © OpenStreetMap contributors, ODbL. Aggregates © snuepi, CC BY 4.0: share and adapt with attribution.