"Where can we eat tonight in Lyon with a terrace?" More and more guests put that kind of question to an assistant rather than a search engine. Being cited by ChatGPT, Perplexity or Google's AI Overviews is what people call GEO (Generative Engine Optimization). This guide covers GEO for a restaurant: what you can verify and control, and what is a gamble.
Read this first: no AI engine publishes a recipe for being cited, and nobody can guarantee it. What we describe falls into two groups: documented facts from the providers (crawler access, how structured data works), and good practices in local visibility that make your venue easier to find and understand, with no promise of results.
GEO for a restaurant: what are we talking about?
When an AI assistant recommends a restaurant, it generally draws on web sources: websites, profiles, reviews, articles. Depending on the product, it may cite its sources with links. Your realistic goal is not "to be number 1 on ChatGPT": it is to be findable, readable and consistent everywhere these systems can read.
What Google itself says about its AI features (AI Overviews and AI Mode) is clear and sober: there are no additional requirements to appear in them, and no special optimisation is needed. A page must be indexed and eligible to appear in Google Search with a snippet. The usual fundamentals still apply: allow crawling, make content easy to find through internal links, provide a good page experience, keep important content in text, and keep your Business Profile information current.
In other words, for Google, GEO is not a separate discipline: it is good local SEO. And Google states that meeting all these rules does not guarantee a page will be crawled, indexed or served.
What is verifiable: crawler access in robots.txt
This is the most factual point, and the one most often ignored. Each service uses crawlers identified by a name (user-agent), which you can allow or block in your site's robots.txt file.
OpenAI distinguishes three crawlers, according to its documentation:
- OAI-SearchBot: used to surface sites in ChatGPT's search features. To appear, allow it in your robots.txt (and allow its published IP addresses if a firewall filters traffic). A site that blocks it will not appear in ChatGPT search answers, though it may still show as a navigational link.
- GPTBot: crawls content that may be used to train OpenAI's models. Blocking it signals that your content should not be used for training.
- ChatGPT-User: visits a page when a user asks for it inside ChatGPT. robots.txt rules may not apply, because the visit is triggered by the user.
These crawlers are controlled independently: you can allow OAI-SearchBot and block GPTBot. OpenAI says to allow roughly 24 hours for a robots.txt change to be picked up by its search systems.
Perplexity documents two crawlers:
- PerplexityBot: surfaces and links sites in Perplexity search results. It is not used to train foundation models. To appear, allow it in robots.txt and accept its published IP ranges.
- Perplexity-User: visits a page at a user's request, and generally ignores robots.txt, since a person triggers the request.
Google: the Google-Extended token lets you control, in robots.txt, whether your content is used to train its Gemini models and to ground answers. Google states that it does not affect Google Search: it has no bearing on inclusion in results and is not a ranking signal. For AI Overviews and AI Mode, Google points to the usual Search indexing conditions.
How to read this: a robots.txt containing Disallow: / for all crawlers, or a firewall that rejects all unknown traffic, can block these crawlers without your knowing. So check:
- that your robots.txt (at the root of the domain) does not forbid the search crawlers of the services where you want to be visible;
- that your host, CDN or firewall does not filter them: Perplexity notes, for instance, that Cloudflare's Bot Fight Mode can still challenge them despite custom rules;
- that you made a conscious choice about training crawlers (GPTBot, Google-Extended): it is an editorial decision, not an obligation.
Worth knowing: robots.txt manages crawling; it is not a way to keep a page out of Google. Reputable crawlers follow its rules voluntarily.
If your site runs on Webflow, Wix or another CMS, you sometimes edit this file through a setting rather than by hand. Check the result by opening yoursite.com/robots.txt in a browser.
Structured data: helping machines read your information
Structured data (schema.org, in JSON-LD format) describes your venue in a format machines understand without guessing. For a restaurant, Google documents the Restaurant type, a subtype of LocalBusiness.
According to Google's documentation:
- the required properties are the name (name) and the address (address, with as many fields as possible);
- the recommended properties include opening hours (openingHoursSpecification), phone, the venue page URL, geographic position, price range (priceRange), the menu link (menu) and cuisine type (servesCuisine);
- the markup goes in a script tag of type application/ld+json and is validated with the Rich Results Test;
- late-night hours, seasonal closures (validFrom, validThrough) and closed days have a precise syntax.
Let us be exact about what this brings. Google writes that there is no special schema.org markup to add to appear in its AI features, and that structured data must match the visible text on the page. Structured data is therefore a good practice for clarity and for eligibility to some Google displays: not a guaranteed AI lever. As for ChatGPT and Perplexity, we found no official documentation saying they read this markup systematically: do not treat it as a switch.
Rule of method: never mark up what is not displayed on the page. If your hours change, update the page and the markup together.
Google profile, reviews and consistent information
For local searches, the Google Business Profile is a central source for Google itself. Among its best practices for AI features, Google explicitly lists keeping Business Profile information current. Our guide covers the full setup: Google Business Profile for a restaurant.
Reviews play a double role: they feed local prominence in Google, and they contain guests' real vocabulary ("quiet terrace", "good for groups", "gluten-free"). We cannot tell you to what extent ChatGPT or Perplexity use them: to our knowledge, no official documentation says. What we do know is that a venue with recent reviews, owner replies and consistent information sends a clearer signal, to humans and machines alike. See replying to Google reviews.
Rules to respect: do not buy reviews, do not offer anything in exchange for a review and do not filter your requests to approach only happy guests: Google prohibits these practices.
Consistent information: the main lever within your reach
An assistant cross-checking several sources runs into contradictions: different hours on the site and the profile, an old name, an old address, an outdated menu. When in doubt, it may get it wrong, or say nothing. Consistency is therefore the most profitable work:
- name, address, phone identical on the site, Google profile, social accounts, directories and reservation platform;
- hours up to date everywhere, including annual closures and public holidays;
- menu and prices available as text on the site, not only in an image or scanned PDF;
- stable facts described simply: cuisine type, atmosphere, capacity, accessibility, private hire, dietary options.
For the menu, favour a readable web page: see online menu in HTML or PDF.
Writing clear content that can be cited
An AI engine extracts passages. Your content is more likely to be understood, and cited, if it is written to make sense out of context. This is an editorial good practice, not a guarantee.
- A short answer at the start of each section. "We welcome groups of up to 20 in the back room; beyond that, contact us for private hire." Then the details.
- Headings that reuse real questions: "Can I bring a dog?", "Are there vegetarian options?", "Where can I park?".
- One page per useful topic: menu, private hire, access, events, FAQ.
- Self-contained sentences: avoid "as seen above" or "our famous dish". Name things.
- No content hidden in images: real text for the menu, hours and offers.
- Dated facts: "autumn menu in effect since September 15".
A short FAQ on your site (4 to 8 real questions from your guests) is useful to humans first. If you mark it up with structured data, follow Google's rules: the marked-up text must be visible on the page.
Checking your visibility without fooling yourself
Testing questions in ChatGPT or Perplexity is a good reflex, but answers vary by person, day and place. A single test proves nothing.
A reasonable method:
- list 10 real guest questions ("Italian restaurant with terrace in Bordeaux", "Sunday brunch near the station");
- ask them every month, note whether your venue is mentioned, with what information, and whether it is accurate;
- fix any wrong information at the source (profile, site, directory): it is your information you need to correct, not the assistant;
- compare with your site traffic and bookings in your usual tracking tools. The absence of measured traffic does not prove the absence of citations: a user can read the answer without clicking.
GEO for a restaurant: frequently asked questions
Do I need an llms.txt file to be cited by AI?
Google writes that no machine-readable files or special markup are needed to appear in its AI features. We have not verified an official position from OpenAI or Perplexity on this file. Focus first on crawler access, consistent information and clear content.
Does blocking GPTBot stop me appearing in ChatGPT?
According to OpenAI, GPTBot relates to model training, while OAI-SearchBot handles appearing in ChatGPT search. They are set separately. Blocking GPTBot therefore does not remove you from search, but check that OAI-SearchBot remains allowed.
Can you pay to appear in AI answers?
For Google's local ranking, Google says you can neither request nor pay for it. For other assistants, we cannot claim that no offer exists: be wary of any promise of a guaranteed position in AI answers.
How long before I see an effect?
There is no reliable delay. OpenAI says roughly 24 hours for a robots.txt change to be taken into account, and Perplexity up to 24 hours, but that says nothing about when an assistant will cite you, or whether it ever will.
Need help being better understood by search engines and AI assistants?
Making a venue readable to Google and AI assistants takes a clean website, consistent data and regular follow-up. māhie supports more than 90 venues on their digital visibility, with more than 70 websites delivered. Concretely, we can:
- audit crawler access (robots.txt, host, firewall) to your site;
- add and validate structured data markup on your Webflow site;
- restructure your pages (menu, private hire, FAQ) so they are clear and citable;
- harmonise your information across site, Google profile and platforms;
- set up monthly tracking of what is visible and what it brings in.
See our websites and Growth & outsourced CMO, then Contact māhie.
