A guest who used to google "restaurant Ghent" now types: "where can I eat good Italian in Ghent tonight, outside, with a table for six?" — and gets three names back. Not ten blue links. Three names.
This is no longer a forecast. In 2026, 45% of consumers use an AI assistant at least occasionally to find a local business, up from 6% a year earlier. More than one in five guests has already asked ChatGPT or Gemini specifically where to eat; among people in their twenties and thirties that figure is 61%. And when the question is asked, the assistant names three to five places and stops. There is no page two.
The problem: a May 2026 benchmark found that 83% of restaurant locations never appear in those answers. Not because they are bad, but because a language model does not consider them confirmed enough. This guide gives you 7 concrete steps to change that, two graphics showing exactly where the thresholds sit, and a scoring tool that tells you in two minutes where you stand today.
Why AI names only three restaurants
A classic search engine hands you a list and lets you choose. An assistant chooses for you. That changes everything: where position seven still brought you visitors, there is now only in or out. Across the categories studied, the three most visible names together take 53.4% of all mentions — everyone else divides the crumbs.
The counter-intuitive part is that this is not the same as classic SEO. Of the venues that never surface in an AI answer, 86% have a neatly completed Google Business Profile. So they are findable — just not recommendable. A language model does not rank, it confirms: it looks for facts that several independent places state identically, and drops anything it is unsure about. One contradictory opening time, and you become the risk the model will not take.
100 restaurants, 17 answers
Out of every hundred restaurant locations, 17 turn up in AI recommendations. To the assistant, the other 83 simply do not exist.
Source: Uberall GEO Studio benchmark, May 2026. The painful detail: 86% of those invisible venues do have a fully filled-in Google Business Profile.
Think of it as a waiter advising a guest about a place down the street. He mentions only what he knows for certain: that they do good fish, that they close on Mondays, that it gets busy. If he is not sure, he would rather say nothing. AI assistants work exactly the same way — and your job is to write down what you know about your restaurant often enough and consistently enough that the assistant dares to repeat it.
7 steps to become visible in AI answers
The steps below are deliberately in this order. The first two are hygiene — without them the rest has little effect. Step three is by far the heaviest lever. The last four build the trust that takes you from contender to regular name.
1. Make your details identical everywhere
Name, address, phone number and opening hours: on your website, your Google Business Profile, Facebook, Instagram, TripAdvisor, the city guide and that one local platform someone once added you to. Models treat that repetition as evidence. If three places say "closed on Monday" and one says "open from noon", that is not a detail — that is a reason not to name you when someone asks what is open on a Monday evening.
So once a quarter, do a round of every place your restaurant is listed, including the forgotten ones. Write your business name identically everywhere (not once with and once without "Bistro"), use the same phone number with the same country code, and apply holiday closures everywhere at once. Your Google Business Profile is the reference point here: it is the source that gets copied most often.
2. Get your menu out of the PDF
A menu as a PDF or a photo is a closed door to a language model. Not a single dish, price or dietary label comes out of it. And that is exactly the information most questions revolve around: "where can I eat gluten-free", "which place does a vegetarian tasting menu", "where do I find fresh pasta under 25 euros".
So put your menu on your own website as plain text, with dish names written out, a short description per dish and clear labels for vegetarian, vegan, gluten-free and the 14 legally required allergens. That pays twice: it makes you findable and it sells better. Our guide on writing menu descriptions covers how to phrase that text so it works for guests and machines alike; designing your menu covers the layout around it.
3. Lift your rating over the threshold
This is the bluntest finding in the research: every assistant applies an implicit floor. ChatGPT recommends almost exclusively venues averaging 4.3 stars or higher. Perplexity sits around 4.1, Gemini around 3.9. Below that, the rest of your effort barely matters.
The star threshold per assistant
Below this average rating, an assistant rarely includes you in a recommendation at all.
A venue on 3.8 stars misses all three cut-offs. Going from 3.8 to 4.3 usually takes 40 to 60 new reviews — one season's work.
The good news is that an average is movable. A venue with 120 reviews at 3.8 stars needs roughly 45 new five-star reviews to climb above 4.3 — that is one review every two days for a quarter. Entirely doable if you ask systematically instead of hoping. Use our free Google review link generator to send guests to the right page in one tap, and read reputation management for hospitality for how to ask without being pushy.
Watch freshness too. A pile of glowing reviews from 2019 followed by silence reads, to a model, like a business that may no longer exist. Twenty reviews from the past three months outweigh two hundred old ones.
4. Let other people write about you
More than 41% of the sources assistants cite on restaurant questions are guides and review platforms — not the venue's own website. And the three big assistants each look slightly differently: ChatGPT leans heavily on independent directories, Gemini gives the most weight to the venue's own website, and Perplexity draws notably from reviews, social media and forums.
That means you need three legs to stand on. Make sure you are listed correctly on the usual guides for your city and region. Make sure your own site tells the full, current story. And make sure someone else writes about you now and then: a local food blog, the traders' association newsletter, a piece in the regional paper about your new summer menu. Our guide to press and media strategy shows how to earn a mention like that on no budget.
5. Write answers, not slogans
Nearly 79% of restaurant prompts to AI are research-style: someone describes an occasion, a party, a dietary need or a moment, and asks for a place that fits. "Where can I have lunch with ten colleagues in the centre?" "Which place is wheelchair accessible and has a quiet room?" "Where can I still get a hot meal after 10pm?"
On most restaurant websites, none of those questions is answered. There is atmosphere, there is passion, there is "welcome to our place". So add pages or paragraphs that answer the question literally, in the words guests use: one about groups and what you can handle, one about your terrace and how many seats it has, one about accessibility, one about parking. That is exactly the same exercise that strengthens classic restaurant SEO — only now it is rewarded twice.
6. Put structured data on your site
Structured data (schema.org) is the version of your information machines can read without guessing. For hospitality that is the Restaurant type, carrying your cuisine, price range, opening hours per day, geographic coordinates, a link to your menu, whether you accept reservations and your average rating.
It is one of the few steps you take once and that keeps working afterwards. If you build your own site, this is baked into any decent website solution — our free restaurant website builder ships that schema automatically. If you work with an agency, ask explicitly for Restaurant schema and test it afterwards; the 9 essential parts of a restaurant website covers what else matters.
7. Measure what AI already brings you
Almost every owner underestimates this channel, and that is down to measurement. Visitors arriving via an AI assistant usually land in your statistics as "direct", "referral" or "unassigned" — analyses show that default Google Analytics settings can understate AI traffic by up to a factor of a hundred. So the conclusion "AI brings us nothing" is often simply a measurement error.
Turn it into a separate segment: filter your referral sources for chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com, and add one question to your own booking flow: "how did you find us?". That costs nothing and gives you an honest picture within a month. How to set such a measurement up is covered in restaurant analytics.
Test your AI visibility in two minutes
The seven steps do not carry equal weight. Your rating counts most, matching details and a readable menu come right behind it, and the rest builds trust. Tick below what is already in order at your place.
Your AI visibility score
Tick what is already true today. Your score shows how likely an assistant is to name your restaurant when someone asks where to eat.
0/100
Invisible — you are in the 83%
Start with steps 1 and 2: matching details and a menu in plain text. Without those two, the rest has little effect.
Then run the test no tool can replace: open ChatGPT, Gemini and Perplexity and ask each three questions as if you were a guest — "where do I eat well in [your town]", "best [your cuisine] in [your city]", "restaurant with a terrace in [your neighbourhood]". Write down literally who gets named and what is said about them. That is your baseline, and at the same time the most honest competitor analysis you can get for free.
Your action plan for the next 90 days
You do not have to tackle everything at once. This is the order that pays off fastest:
Week 1 — measure and clean up:
- Ask your three guest questions to ChatGPT, Gemini and Perplexity and note who gets named
- List every place your restaurant is mentioned online, including the forgotten ones
- Correct name, address, phone and opening hours until they match everywhere
Weeks 2 to 4 — make what you offer visible:
- Put your full menu on your site as text, with dietary labels and allergens
- Add Restaurant schema (cuisine, price range, hours, menu, booking link)
- Write three answer pages: groups, terrace or accessibility, and parking
Months 2 and 3 — build trust:
- Ask for reviews systematically, with a review link on the bill and in your confirmation email
- Aim for one external mention a month: local blog, guide or newsletter
- Track your AI traffic separately and repeat your guest questions monthly to see progress
Conclusion: make the assistant dare to name you
Being visible in AI answers is not a trick and not an ad budget. It is the sum of seven sober things: matching details, a readable menu, a rating above the threshold, other people writing about you, pages that answer real questions, structured data and honest measurement. Get those right and you get named; leave them and you disappear into the 83% — however good the food is.
The nice part is that every step pays off outside AI too: the same work earns you a higher Google position, more direct bookings and fewer phone calls asking things your website could have answered. At HappyChef that is built into the system — your own booking page with Restaurant schema, guest profiles, automatic review requests after the visit and 0% commission on your reservations. See how it works and make sure the next guest who asks an assistant hears your name.