GEO for restaurants
Dining queries are the most heavily constrained category in hospitality. Cuisine, price, party size, dietary provision, opening day, booking policy. Every constraint you leave unstated is a table you never hear about.
Short answer
An assistant asked “best restaurants in [city]” returns three or four names, not ten links. Which names depends on two things: whether a crawler could fetch and read a page about you, and whether the guides and review platforms it draws on describe you accurately. We work on both, and we can tell you within a call which one is your problem.
Why restaurants lose these answers
Almost never because the food is worse. The three failures we see repeatedly are mechanical.
The menu is drawn by JavaScript. Plenty of restaurant sites assemble the menu, the hours and sometimes the address in the browser. No major AI crawler runs JavaScript, so what those crawlers receive is an empty shell with a booking widget in it. The restaurant is invisible to the category query no matter how much press it has.
The opening paragraph is atmosphere. The text around your H1 is what gets stored and reused, and a paragraph about candlelight and provenance contains nothing a constrained query can match.
Nobody outside the site describes it properly. Category queries are decided largely by roundups, city guides and review platforms, and a restaurant that has never been written up accurately is competing on your own site alone, which is the half that barely participates.
The attributes that decide filtered dining queries
These are the constraints real diners put into an assistant. Each one you fail to state in crawlable text is a query family you drop out of.
- Cuisine, specifically. “Sichuan” beats “Chinese”. “Neapolitan pizza” beats “Italian”. Generic category words lose to specific ones because the query is specific.
- Price per head, with and without drinks. “Affordable” and “upmarket” are among the most common query modifiers there are.
- Neighbourhood, not just the city. “Near the station” and “walkable from the old town” resolve on this.
- Exact opening hours, including which days you are closed. “Open on a Monday” is a query that eliminates most of a city.
- Booking policy. Reservations, walk-ins, or both, and any cutoff. Whether groups need to call.
- Largest group you can seat. Party-size constraints are extremely common and almost never published.
- Dietary provision: vegetarian, vegan, gluten-free, halal, jain. State what you actually do, not what you could accommodate on request.
- Outdoor seating, and whether it is covered.
- Accessibility: step-free entry, accessible toilet. Rarely published, frequently asked.
- Whether children are genuinely welcome, and whether there is anything for them to eat.
The awkward ones matter most, because your competitors leave them out too. Publishing that you cannot seat more than six, or that the room is up a flight of stairs, wins you the queries where that is the point and loses you bookings that would have gone wrong anyway.
Where restaurants actually get described
Your own site wins the branded query, when someone already knows your name. The category query is decided elsewhere: roundup articles, city guides, review platforms and forum threads. Branded web mentions correlated with AI Overview visibility roughly three times more strongly than backlinks across a 75,000-brand study, though the authors are explicit that correlation is not causation and so are we.
One claim worth distrusting while you are here. The widely repeated story that Foursquare powers ChatGPT's local results did not survive testing: a run of 2,880 prompts across twelve metros recorded Foursquare citations at 0.00 percent, with Yelp grounding appearing in the overwhelming majority of runs after OpenAI's non-exclusive Yelp licensing deal of 23 July 2026. One test, not a settled fact, but enough to be sceptical of anyone selling Foursquare work as a ChatGPT lever.
What an engagement looks like
We sample the dining prompts for your city and category first, so the target list comes from what the assistants actually cite rather than from a generic directory checklist. Then the mechanical work, then the constraint coverage, then the outreach. We report monthly with the sample size and the margin stated.
We do not promise you a position, because there is no position to promise. Our evidence page sets out exactly which parts of this have data behind them and which are correlational.
Four disciplines, in the order the evidence supports
Common questions
A large part of the underlying work is shared, and we will tell you that rather than invent a distinction. Crawlability, consistent listings and clean site structure serve both. Where it diverges: AI crawlers apart from Googlebot do not run JavaScript at all, an unlinked mention counts where a link used to, and there is no ranking to report, so measurement has to be a sampled prompt set instead.
Sometimes. Several restaurant platforms render the menu and hours client-side, which makes them invisible to AI crawlers even though they look fine to you and to Google. We check this by fetching your pages with JavaScript disabled, which takes minutes, and we tell you before you commit to anything.
There is nothing to get into. No submission form, no index, no verification file. What exists is whether your pages can be fetched and read, and what independent sources say about you. Anyone selling you a submission service for restaurants is selling you nothing.
No, and spreading thin across all of them is the common mistake. We sample which sources the assistants actually cite for your category and city, and work on those. The list is usually shorter and more specific than people expect.
Find out which half is your problem
Book a 30 minute call. We will fetch your site the way the crawlers do, check whether the answering bots can reach it, and look at how the assistants answer your category today.