A growing share of your future guests will never type your name into Google. They will ask an assistant. "Where should I eat near the old town tonight?" "Best vegan dinner in Lisbon?" "Somewhere quiet for a business lunch that does a gluten-free option?" The assistant will answer in a sentence or two and name a handful of places. This guide is about how a restaurant becomes one of those names.
The honest framing first. Nobody outside the companies that build these models knows the exact rules, and the rules change often. So this is not a set of tricks. It is a clear explanation of how AI assistants appear to choose what they cite, based on what is observable and on the parts of the web they openly read, followed by a practical checklist you can act on whatever platform you care about. Most of it is simply good housekeeping for the modern web, which is reassuring: the same work that helps a model also helps Google and helps a human.
Why this matters now
For twenty years the question was how to rank on Google. That question has not gone away, but a second one has appeared next to it: how to be the answer when someone asks an AI assistant instead of searching. People now put genuine planning questions to ChatGPT, Claude, Perplexity and Google AI Overviews, and increasingly trust the reply without clicking through to ten blue links. When the assistant names two or three restaurants and skips the rest, being named is the entire game.
This is usually called answer engine optimisation, or AEO, and sometimes generative engine optimisation, or GEO. It is the same instinct as classic search engine optimisation, pointed at a new surface. The reassuring part, which we will keep returning to, is that the work overlaps heavily with sound SEO. You are not chasing two separate, contradictory goals. You are making your restaurant legible to machines, and legible machines are what both Google and the assistants run on.
How AI assistants appear to decide what to cite
Treat everything in this section as informed observation, not a leaked rulebook. The models are proprietary and they evolve. That said, a consistent picture emerges from how they behave and from what their makers say publicly. Assistants favour sources they can read easily and trust quickly. In practice that means a handful of things.
They prefer text they can actually read
An assistant reads HTML text far more reliably than it reads a photo of a menu or a PDF. When your dish names, descriptions and prices live in real, selectable text on a page, a model can quote a specific plate, confirm a price, or tell a guest what is in it. When the menu is a flat image or a downloadable file, that information is effectively invisible to the answer. This is the single most common reason a restaurant with a perfectly good menu never gets cited: the menu exists, but it cannot be read.
They lean on structured data
Schema.org is a shared vocabulary that lets a page label what it is. A page can declare, in machine-readable terms, "this is a Restaurant, here is its name, address, telephone, opening hours, cuisine, and here is its Menu with these items and prices." That structured data does not change what a human sees, but it removes ambiguity for a machine. A model no longer has to guess whether a string of text is an address or a dish; the page has told it. Restaurants whose pages carry clean Restaurant and Menu structured data are simply easier to summarise correctly.
They reward consistency across the web
If your name, address and phone number read one way on your own page, another way on a directory, and a third way on an old listing, an assistant has to reconcile the conflict and may hedge or pick a competitor it trusts more. Consistent name, address and phone details across every place you appear are a quiet trust signal. This is old SEO advice, and it matters just as much to a model trying to decide whether your information is reliable.
They prefer fresh content
A page last updated two years ago, with a menu that no longer matches the kitchen, is a weak source. Assistants tend to favour content that looks current. A restaurant page that reflects this season's dishes, today's prices and a sold-out item removed in real time reads as a living, trustworthy source rather than an abandoned one.
They value plain-language detail, especially dietary and allergen information
When a guest asks for "a good gluten-free dinner" or "somewhere with vegan mains," the assistant can only answer well if that information exists in plain words on a readable page. A menu that states, in plain English, that a dish is vegetarian, or that it contains nuts, gives the model something concrete to match against the question. Vague or missing dietary detail means your restaurant is quietly skipped for exactly the queries where you might have won.
For ChatGPT specifically, being in Bing helps
ChatGPT search draws partly on the Bing index. That makes a practical point easy to miss: optimising only for Google leaves one of the most-used assistants partly blind to you. Being present, crawlable and current in Bing is part of being citable by ChatGPT, not an afterthought.
Assistants cite pages they can read (text, not images or PDFs), trust (consistent details, structured data), and rely on (fresh, specific, plain-language content). Almost everything below is one of those three things in a different outfit.
The restaurateur's checklist
Here is the practical part, written so you can act on it regardless of which assistant or platform you use. None of it requires you to understand the technical machinery; it requires the information to exist in a readable form.
- Put your full menu in real text. Not a PDF, not a photo of the printed card. Dish names, short descriptions and prices as text a machine can read. This is the foundation; nothing else helps much without it. Our digital menu versus paper menu comparison covers why this format wins beyond AI search too.
- Make your name, address and phone number identical everywhere. Your own page, your directory listings, your social profiles. One spelling, one format, one phone number. Fix the stale ones.
- Write clear, specific dish names and descriptions. "Slow-cooked lamb shoulder with rosemary and white beans" gives an assistant something to match against a query. "Chef's special" gives it nothing. Descriptive, sensory wording also helps humans order; field research from the Cornell University Food and Brand Lab, reported by FoodService Director, found descriptive menu labels were associated with roughly 27 percent higher sales of the labelled items, though that exact figure should be treated as indicative rather than definitive.
- State allergens and diets in plain language. Mark which dishes are vegetarian, vegan or gluten free, and list the allergens present. In the EU this is not only good for AI search, it is the law: Regulation (EU) No 1169/2011, in application since 13 December 2014, makes allergen information mandatory for non-prepacked food served in restaurants and cafes, covering the 14 regulated allergens. Our restaurant allergen compliance guide explains exactly what you must disclose.
- Keep it current. Update prices when costs move, add this season's dishes, remove what you no longer serve. A page that matches the kitchen is a page a model trusts.
- Make sure search engines can crawl the page, including Bing. If a crawler cannot reach your menu, no assistant that relies on that index can cite it. Being indexed in Bing matters specifically for ChatGPT.
- Add Schema.org structured data if you can. Restaurant and Menu markup removes ambiguity for machines. If editing code is beyond your setup, this is the one item best handled by a platform that does it for you.
If you read that list and thought "most of this is just keeping an honest, current, readable menu online," you have understood it correctly. AEO for restaurants is far less exotic than the acronym suggests. The work is mostly discipline, and the gap between restaurants is usually effort, not secret knowledge.
A note on photos, honesty and the rest of the menu
Making a page machine-readable does not mean stripping out the human parts. Photos still matter enormously to the guests who do reach your page. On a digital ordering platform, Uber Eats reports from its own 2019 global data that menus with item photos are three times more likely to be ordered from than text-only menus, and DoorDash's 2024 ordering trends report found that 38 percent of customers say they look at photos when choosing a new restaurant. Peer-reviewed work by Hou, Yang and Sun in the International Journal of Hospitality Management (2017) found that adding a picture to a clearly named dish lifted diners' willingness to pay and their intention to order, a directional effect rather than a single headline number.
The point for AI search is simply this: keep the photos for people, but never let them be the only place the information lives. A beautiful photo of a menu board is invisible to an assistant. The same menu in text, with the photos alongside, serves both audiences. And keep the photography honest, because a photo sets an expectation the plate then has to meet. The two goals, being read by machines and being chosen by humans, do not conflict; they reinforce each other.
How Opnclo automates this
Everything on the checklist can be done by hand. The reason Opnclo exists is that doing it by hand, correctly, and keeping it current forever, is more work than most kitchens have time for. The tagline is "found by humans, quoted by AI," and the product is built so that a restaurateur gets the AEO groundwork as a by-product of simply keeping their menu up to date.
Concretely, every Opnclo restaurant gets a single, fast page that carries structured Restaurant and BreadcrumbList data, with the full menu marked up so machines can read every dish, price and allergen. Alongside that HTML page, Opnclo publishes a clean Markdown export of the same content, because many assistants find plain Markdown easy to parse and cite; it is an emerging practice rather than a proven ranking factor, but it removes friction at no cost. New and updated pages are pinged to Bing through IndexNow, which matters for ChatGPT's index. Pages that serve guests in more than one language carry the right hreflang signals so the correct language is surfaced. Allergens and diets are expressed in plain English that a model can match against a question. And because updates publish instantly, the content stays fresh, which is one of the things assistants reward.
We cannot promise ChatGPT or Perplexity will name your restaurant; nobody can, and anyone who does is guessing. What a well-structured page does is remove every avoidable reason a model would skip you. It makes you readable, consistent, current and specific. The rest is the assistant's call.
SEO and AEO are the same discipline
It is tempting to treat AI search as a brand-new specialism. It is better understood as the next chapter of the work restaurants should already be doing online. A page that an assistant can read and cite is, almost by definition, a page Google indexes well and a human navigates easily. Clean structure, readable text, consistent details, fresh content and clear dietary information: that single investment pays off across Google, the assistants, and the guest standing in your doorway checking your menu on their phone. If you want the broader background, our complete guide to QR code menus covers how the digital menu underneath all of this works.
You can also see the principle in practice. Every restaurant on the Opnclo restaurants directory has a page built to this standard, and each one links to the others in its city, the kind of structure both search engines and assistants reward.
Frequently asked questions
What is answer engine optimisation (AEO) for restaurants?
It is the practice of making your restaurant easy for AI assistants such as ChatGPT, Claude, Perplexity and Google AI Overviews to read and cite, sometimes called generative engine optimisation or GEO. Classic SEO aims for a clickable link; AEO aims to be the source a model quotes when it answers a question. The underlying work, clean structured data, crawlable text, consistent details and fresh content, overlaps heavily with good SEO.
How do AI assistants decide which restaurants to recommend?
The exact ranking is proprietary and changes often, so treat any specific claim with caution. What is reasonably clear is that assistants favour pages they can read and trust: text rather than images or PDFs, structured data labelling the entity as a restaurant with its address, hours and menu, details that match across the web, and recent content. ChatGPT search draws partly on the Bing index, so being present in Bing matters for that assistant in particular.
Does my restaurant need a website to be recommended by AI?
You need a page a model can read, which is not the same as a full website. A single, well-structured page with your name, address, hours, contact details and full menu in plain text, marked up with Schema.org Restaurant data, is often more useful to an assistant than a slow, image-heavy site where the menu is a PDF. Format and structure matter more than size.
Why should my menu be in text rather than a PDF or an image?
A menu locked inside a PDF or a photo is hard for search engines and assistants to read reliably; the dish names, prices and allergens stay trapped in pixels. As plain, structured text, the same menu lets an assistant quote a dish, confirm a price, or say which items are gluten free. If you want to be the answer, the menu has to be readable, not just visible.
What is a Markdown export and why does it help with AI search?
Markdown is a plain text format with light structure and no design clutter, which many assistants find easy to parse. Publishing a clean Markdown version of your restaurant page gives a model an unambiguous source to read and cite. Opnclo publishes one alongside every restaurant page. It is an emerging practice rather than a guaranteed ranking factor, but it costs nothing and removes friction.
One honest closing thought. AI search for restaurants is still early, and the rules will keep moving. Anyone selling you a guaranteed place in ChatGPT's answers is guessing. What is durable, and what will still be true a year from now, is simpler: a clear, current, readable menu, with consistent details and structured data, is the best thing you can put in front of a machine and a guest alike. Get that right and you have done the part that actually matters. See what Opnclo costs if you would like it handled for you.