Restaurant discovery is conversation-shaped now — “somewhere special for an anniversary, not too formal, near the coast” is exactly the kind of question people put to ChatGPT, because it's exactly the kind Google was always bad at. Ask it for your town and watch the answer: review platforms and listicles dominate, with a couple of restaurants named — almost never because they're the best, but because their websites say things machines can repeat.
The menu-PDF problem
The single most common failure we see on restaurant sites: everything that would win you an AI recommendation is locked in a PDF or an image. Your menu, your prices, your Sunday roast, your vegan options, your dog-friendly terrace — if it's a downloadable menu scan, a machine reading your site learns nothing. The restaurant next door with a plain HTML menu page gets the “good vegan options in X” recommendation, whatever its food is like.
The occasion angle matters too. Diners ask machines by occasion — birthday, client lunch, walk-friendly, kids welcome — and models can only match you to occasions your site actually mentions. A sentence like “private dining room for up to 14, popular for birthdays and small weddings” is a machine-recommendable fact. Most sites never write it down.
What a machine needs from a restaurant site
- Menu as text, on a page — with prices. Keep the PDF for printing if you like, but the machine-readable version is the one that earns recommendations.
- Restaurant structured data: cuisine, price range, opening hours, address, booking link, accepted occasions — the schema.org Restaurant type exists for exactly this.
- The occasions and specifics named in text: dog-friendly, sea view, step-free, Sunday roast until 4pm, pre-theatre menu. Every specific is a query you can win.
- Your reviews linked from your own site — Google and the review platforms you're strong on, connected via structured data so the model can attribute them to you rather than to the platform.
- AI crawlers not blocked — booking-platform site builders sometimes block them by default, which ends the story before it starts.
Sixty seconds to see where you stand
Run the free check — it reads your site the way the machines do and shows what they can and can't see. If you want the actual answers, the £19 citation audit asks ChatGPT, Claude, Gemini, Perplexity and Grok twelve real diner questions for your area — “best Sunday roast near X”, “somewhere for a big family birthday” — and shows exactly who gets the table you should have had. Fixes are quick web work; the citations follow over two to four months as engines re-crawl — longer than anyone likes to admit — but the restaurant that starts first in your town keeps the slot.