More and more people don't search for "where to eat in Pollença": they ask an assistant. And an assistant doesn't pick from the top ten results. It picks from the sources it can read, understand and cite.
If your restaurant isn't in that group, you're not coming last. You don't exist in the answer at all.
This guide is about the second problem: what your site needs so an AI can recommend you, why almost no restaurant meets the bar, and what you can fix this week.
What's the difference between ranking on Google and being recommended by an AI?
It isn't the same game, even if it looks like it. Google puts you in an ordered list and the user decides. The assistant decides itself, and it can only decide with what it managed to read.
| Classic search engine | AI assistant |
|---|---|
| Puts you in a list of ten | Names two or three, or none |
| The user compares and chooses | The assistant has already chosen for them |
| A good-looking site is enough | It needs data it can extract |
| Reads the text on the page | Reads structured data above all |
| A PDF is just another result | A PDF is usually a dead end |
| Rewards links and age | Rewards clarity, specifics and consistency |
The practical consequence: you can have a beautiful, fast, well-ranked website and still be invisible to the assistant. They are different requirements.
The five pillars for an AI to be able to recommend you
1. Your menu on the site, not in a PDF
This is hospitality's most repeated mistake. A PDF opens slowly on a phone, doesn't fit the screen and, above all, is a closed document: nobody can answer "do you have gluten-free options?" by reading a file that doesn't even know what a dish is.
The menu has to be a page, with its categories, its dishes, its prices and its allergens as text. Then "how much is the ribeye?" can be answered by citing you.
2. Restaurant structured data
It's an invisible tag inside your site that says, in a format machines understand without interpreting anything: this is a restaurant, it's called this, it's here, it opens at these hours, it serves this cuisine, this is its price range, it takes bookings and its menu is at this address.
Without it, the assistant has to guess by reading your text. With it, it knows. That's the difference between being deduced and being read.
3. The same name, the same address and the same phone number everywhere
Your site, your Google listing, the directories, the booking platforms, your social profiles. If in one place you're "Casa Manolo" and in another "Restaurante Casa Manolo S.L.", or you're still dragging an old phone number around, you're giving the assistant reasons to doubt which one you are.
Consistency isn't perfectionism: it's what lets all those mentions be tied to a single business.
4. Real opening hours, including the day you close
Very few get this right, and it's one of the most asked questions. "Is it open now?" is one of the most frequent things an assistant is asked about a restaurant.
And watch the detail: the closing day has to be declared. A site that implicitly says it opens seven days a week sends a customer to a locked door — and that experience gets paid for in reviews.
5. FAQs written the way people actually ask
Not "Useful information". The literal questions: whether dogs are allowed, whether there are vegan options, whether you can park nearby, whether there's a set lunch menu, whether there's a terrace, whether you take large groups, whether there are high chairs.
Each of those questions, answered in two or three clear sentences, is a fragment an assistant can take and hand back as it is. It's the most citable form of content there is.
What the AI reads about you that you never see
Your site is only one of the sources. The rest count too, and are usually neglected:
- Your Google listing, with category, hours, photos and complete attributes.
- The reviews, and above all your replies to them: that's your own text, public and recent.
- Local directories and aggregators, where your data is often years out of date.
- Local press and area blogs, which weigh more than you'd think in towns and neighbourhoods.
- Your social profiles, especially if the bio says what you are and where you are.
An uncomfortable and very useful exercise: ask an assistant today about your kind of restaurant in your town. If it names you, look at what it says about you. If it doesn't, look at who it names and why they're easier to read than you.
Checklist: twelve checks for this week
- The menu is on the site as text, not as a PDF.
- Every dish has a visible name, description and price.
- Allergens are declared and validated by the kitchen.
- The site declares restaurant structured data.
- Opening hours are complete, closing day included.
- The address is identical on the site and on the Google listing.
- The phone number is the same everywhere, with no old numbers left.
- There is a page or section with real frequently asked questions.
- People can book without leaving your site.
- The site loads fast on a phone on mobile data, not on the restaurant's fibre.
- Your Google listing has the right category and recent photos.
- You reply to reviews, bad ones included, with your own words.
If you tick fewer than eight, you have work ahead. If you tick fewer than four, it isn't that they don't recommend you: it's that they can't.
Does any of this matter if nobody clicks?
It's the honest objection and it deserves an equally honest answer: a citation is not a visit.
We know first-hand. Our own guide to answer engine optimisation has been cited 648 times by Bing's assistant between April and August 2026. That same page, in classic search, has 39 impressions and one click.
So no, it doesn't fill your site with traffic. It does something else, and for a restaurant that something is worth more: the recommendation arrives anyway. The customer who asks "where should I eat around here" and gets your name doesn't need to visit your site. They need the name, the address and for you to be open. They show up at the table, not in the analytics.
That forces you to measure differently — by direct bookings and by people who arrive saying they'd heard of you, not by sessions — but it's exactly the same deal you've been accepting from word of mouth for decades.
Where to start if you can only do one thing
Get the menu out of the PDF and put it on the site as text.
It's the hardest one and the one that changes the most, because it unlocks everything else: the dishes, the prices, the allergens, the questions about what people can eat, and even the structured data, which stays half-done without a readable menu.
The full method behind all of this is in our AEO Guide 2026.
If you'd rather see it done: Tirano Street Food and Tierra de Fuegos are on the same street in Puerto Pollença, and both have their menu, their hours and their bookings on their own site. If you want that for yours, get in touch.
Your next project
starts here.
Fast websites for real businesses. Let’s talk about yours.
See what we build for restaurantsRelated articles
Por pedido
−30%
de cada ticket
Por comensal
−€
por cada persona sentada