A diner describes the night — four people, Friday, quiet enough to talk, one of them can’t eat gluten — and asks AI where to go. Here’s what AI reads before it names a restaurant, and why a menu locked in a PDF takes you out of the running.
To get a restaurant recommended by ChatGPT and AI search, your menu has to exist as readable text rather than a PDF or a photo, with dish names, descriptions, prices, and dietary labels an assistant can quote. Add restaurant and menu structured data, answer the occasion questions diners actually ask, keep reviews and holiday hours current, and hold a live reservation listing. Restaurants that do this get named. Restaurants hiding the menu in a download don’t.
No other local business hands a customer its entire inventory before they walk in. Your menu is the product list, the price list, and the reason someone picks you over the place two doors down — and on most restaurant websites it is a PDF, a photograph of a chalkboard, or a slideshow that fades between images. A person can squint at that on a phone. An assistant reading the page finds a link to a file and nothing else. So when somebody asks which place nearby does a proper bowl of pho, or has a gluten free crust, or serves brunch on Saturday, your restaurant is not in the running. Not because the food is worse. Because the words are missing.
The second problem hides behind the first. Restaurants are the most reviewed, most listed, most photographed businesses in any town, and all that abundance feels like visibility. You are on Google, on Yelp, on TripAdvisor, in a handful of aggregators, and in a few hundred people’s camera rolls. Nearly all of it describes the experience — nice patio, friendly server, a bit loud on weekends — and almost none of it answers what a diner is really asking. The question is rarely “Italian restaurant near me.” It is four people, Friday at seven, somewhere we can hear each other, one vegetarian, home before the sitter turns into a pumpkin. Those constraints are the search. A restaurant that has written them down somewhere readable wins answers that no amount of review volume will buy.
“We need a table for six tonight and everywhere is booked. Who nearby still has availability?”
The highest-intent restaurant prompt there is, and it resolves through live reservation data. If you hold no OpenTable or Resy inventory, an assistant has no way to see your open seven-thirty top and skips you for a place it can book.
“Best gluten free pizza near me that isn’t a chain?”
Two filters at once: a dish and a dietary constraint. AI answers from menus where the crust is actually labeled, so a page reading “we can accommodate dietary needs” matches nothing while a marked gluten free crust matches exactly.
“Where can I take my parents for an anniversary dinner, somewhere quiet with real table service?”
An occasion prompt with a noise constraint. Nothing on a typical restaurant site addresses either, so AI falls back to review snippets and best-of lists — and names whichever restaurant those sources described as quiet.
“Is Bellwether Kitchen & Bar still open, and has it gone downhill?”
Restaurants close and change hands more than any other local business, so assistants hedge on ones they can’t confirm. A steady flow of recent reviews and current holiday hours is what turns a hedge into a recommendation.
“What does dinner for two run at a decent place downtown, with a drink each?”
AI builds the estimate from whatever prices it can read. Restaurants with a text menu get quoted with real numbers; restaurants with a PDF get lumped into a generic dollar-sign band nobody can act on.
“Restaurant near the theater where we can eat and still make a 7:30 curtain?”
A landmark plus a clock. The winning answer comes from a restaurant that says how long dinner takes and mentions the theater by name — a sentence almost no menu-first website contains.
“For a birthday dinner for ten, is Bellwether Kitchen or Twelve Oaks Tavern the better call?”
A head-to-head on party size. Whichever restaurant states its largest table, its private room capacity, and its group policy gets described in detail, and the one that says nothing gets described as an unknown.
“Who around here makes birria tacos, and are they open on Sunday?”
A dish-level search, which is how people actually crave food. Every named plate on a readable menu is a query you can win; every plate trapped in an image is a query you are guaranteed to lose.
“Which places near me have a private room for a work dinner of about twelve?”
High-value, low-competition, and almost never answered on a restaurant website. A short private events page naming capacities and a per-person minimum makes you the only citable option in most towns.
“Any decent happy hour or early dinner deal near me on a weeknight?”
Deals live on chalkboards and Instagram stories that vanish, so AI mostly cannot see them. Putting the days, times, and actual items on a permanent page turns a promotion into something an assistant can repeat.
When an AI assistant names a restaurant, it’s synthesizing a handful of sources it considers trustworthy. These carry the most weight for this industry — and consistency across them matters as much as presence on any one.
The only place your dishes, descriptions, prices, and dietary labels exist in your own words. Every other source describes the experience of eating at your restaurant; only your menu says what is on the plate, and only if it is written as text.
Carries the things assistants check first for a restaurant: cuisine category, the menu link, dine-in and takeout attributes, holiday hours, the reservation or order button, and the review stream that signals whether you are currently thriving or quietly fading.
Restaurants are the one category Yelp never lost, and its structured attributes — noise level, good for groups, good for kids, parking, ambience — map almost perfectly onto the constraints diners put in their prompts. Yelp also licenses its data into assistant answers.
The booking layer assistants now reach into directly: ChatGPT’s restaurant suggestions are powered by OpenTable, with a reservation link attached to the recommendation. A restaurant with no live inventory can be mentioned but not booked, and often is not mentioned at all.
“Best tacos in town” questions get answered from city magazine roundups, neighborhood newspaper features, food blogs, and long Reddit threads. This is the citation layer that decides superlative prompts, and no other trade has anything like it.
Where visitors, hotel guests, and out-of-town business travelers look, which makes it the source behind “where should we eat near the convention center” prompts. Its cuisine and price filters give AI a clean way to narrow an unfamiliar town.
Dish photos, captions, and tagged locations are how a room’s atmosphere and a plate’s appearance get described in words at all. It rarely wins the answer alone, but it feeds the vocabulary the lists and the reviews then repeat.
Each of these maps to one of the six dimensions we score. They’re the failure patterns we see over and over in this industry — and every one is a content fix, not a budget fix.
A PDF, a photo of the printed menu, or a fading slideshow is still the most common way restaurants publish the single most important page they have. Assistants read a link to a download and learn nothing: not one dish name, not one description, not one price. Meanwhile diners search at the level of the plate — the birria taco, the cacio e pepe, the gluten free crust, the kids grilled cheese. Every item on a text menu is a question you can be the answer to, and every item locked in an image is a question you have decided in advance to lose.
Restaurants can describe themselves to machines more precisely than almost any other kind of local business. There is dedicated vocabulary for the cuisine served, the price range, whether reservations are accepted, the hours for each service, and for the menu itself broken into sections, dishes, and prices. Most restaurant sites carry a generic business block from their template and stop there. That leaves an assistant guessing at facts you could simply have declared, which is the difference between being quoted confidently and being described vaguely.
Almost every real restaurant search carries constraints attached: how many people, how loud, how long, whether there is a patio, whether a stroller fits, whether a vegetarian will have more than a side salad, whether you can be out the door in an hour. Those are the deciding facts, and they are missing from nearly every restaurant website in the country, which instead offers a paragraph about passion for fresh ingredients. Answering the constraints plainly is unglamorous work that wins prompts nobody else is even competing for.
No category churns like this one, and assistants have learned to be careful about sending someone to a place that may no longer exist. Recency is doing the heavy lifting: a stack of reviews from three years ago carries a fraction of the weight of a handful from last month, and a listing with stale hours, no recent photos, and nothing posted since the last renovation looks indistinguishable from one that quietly went dark. A restaurant with fewer reviews but a steady weekly trickle often gets named ahead of the town institution with ten times the total.
Ask an assistant for the best ramen within twenty minutes and it will not compute an answer — it will repeat one. That answer is assembled from city magazine roundups, neighborhood paper features, food blogs, and the Reddit thread where somebody asked the same thing two years ago. Restaurants are also the rare trade where the contractor directories are simply irrelevant, so skipping them costs nothing, but skipping the editorial layer costs everything. Being genuinely easy to describe — one cuisine, one neighborhood, one dish worth traveling for — is what gets a restaurant onto lists in the first place.
None of these rewrites required a developer or a redesign — just replacing copy that says nothing with copy that states facts an AI can quote.
A sample report for Bellwether Kitchen & Bar, a fictional restaurant in La Grange, IL. Click through the tabs — your report will look exactly like this, scored against your real website.
Bellwether is the most heavily cited sample on this site and still loses two thirds of the prompts that matter, which is the whole lesson of this page. Being on every platform got the restaurant recognized; it did not get the restaurant described. An assistant can confirm Bellwether exists, sits in La Grange, and is well liked, and can say almost nothing else. It does not know about the gluten free pasta, the twelve-seat back room, the pre-theater seating that turns tables by seven, or the fact that Tuesday is half price on the raw bar. All of that is true, all of it is on a chalkboard or in a PDF, and every bit of it is a prompt going to somebody else.
These scores reflect how likely AI platforms such as ChatGPT, Google AI, Perplexity, and Claude are to include this business when recommending local services. They are based on the same signals AI systems use to decide which businesses to surface.
How the site performs across all six dimensions
Prioritized fixes specific to this report — in order of impact
A phased action plan based on these scores
Showing 6 of the 12 prompts tested, on 2 of the 4 platforms. Your report includes all of it: Perplexity, ChatGPT, Gemini, and Claude.
Bellwether Kitchen & Bar is a fictional business created for this sample — the scores illustrate patterns we see in the industry, not a real report. Your report reflects your real website.
The rule that beats every hack: your business name, address, phone, and hours must be identical on every one of these. AI cross-checks them, and disagreement reads as risk.
| Directory | Priority | What to do |
|---|---|---|
| Google Business Profile | Essential | Pick the most specific cuisine category rather than the generic Restaurant one, attach a link to your text menu, fill in the dine-in, takeout, delivery, outdoor seating and reservation attributes, and set holiday hours before every holiday rather than after. |
| Yelp | Essential | Complete every attribute field, because noise level, good for groups, good for kids, parking, and ambience are exactly the constraints diners put in their prompts. Upload the menu as text where the platform allows it rather than as a photo. |
| OpenTable | Essential | Keep live inventory here if you take reservations at all, since ChatGPT’s restaurant recommendations are powered by OpenTable and attach a booking link. Fill in cuisine, dining style, price band, parking, and dress code, because those fields are what the recommendation gets filtered on. |
| Resy | High value | Worth holding alongside or instead of OpenTable if your room is chef-driven or your regulars already book there. Keep the description, the cuisine tags, and the special events current, and make sure your table inventory really reflects what the host stand has open. |
| TripAdvisor | High value | Claim the listing and complete the cuisine, meal, price range, and dietary filters, because this is where visitors and hotel guests get matched to a restaurant they have never heard of. Respond to reviews so the profile reads as actively managed. |
| Apple Business Connect | Worth having | Claim your place card so Apple Maps and Siri carry correct hours, photos, and the same menu link. It is a separate ecosystem from Google, and a surprising share of iPhone-first diners never leave it. |
| Worth having | Put the neighborhood, the cuisine, and a link to the text menu in the bio rather than only a link tree, and caption dish photos with the actual dish name. The captions are what turn a photograph into words something can read. | |
| Local food media and best-of lists | High value | Identify the city magazine, neighborhood paper, and food blogs that publish roundups in your area and pitch them one specific thing you do better than anyone nearby. These lists are what superlative prompts get answered from, and nobody assigns them to you automatically. |
It is the single biggest one. A PDF is a picture as far as an assistant is concerned, so every dish name, description, and price on it is invisible, and dish names are exactly what people search for. A text menu page is also easier to change than a PDF, because you edit one line instead of redesigning and re-exporting a document. Keep the printed version for the table and let the website carry the words.
Because reviews tell an assistant that people enjoyed themselves, not what you serve or who you suit. They also age. Recency now carries far more weight than raw volume in this category, so a restaurant collecting a few fresh reviews every week can be recommended ahead of one with ten times the total sitting untouched since 2022. Volume proves you were good. Recency proves you still are.
If you take reservations at all, it changes what an assistant can do with you. ChatGPT’s restaurant recommendations are powered by OpenTable and come with a booking link attached, so a restaurant with live inventory can be booked inside the conversation while a restaurant without it can only be mentioned. For a walk-in-only counter it matters far less, but say so plainly on your site so nobody assumes you are simply full.
As specific as your kitchen honestly allows, including the limits. Mark which dishes are gluten free as written and which can be modified, and say plainly whether you have a dedicated fryer, whether pasta shares water, and what you can guarantee for a celiac guest. Vague reassurance matches no search and helps nobody. An honest, labeled menu wins the prompt and keeps the wrong guest from having a bad night.
Yes, and these are some of the highest-value sentences you can publish. Group dinners, work parties, and pre-theater seatings are worth hundreds of dollars a booking, and the people planning them are asking AI questions almost no restaurant website answers. Naming your largest table, your private room capacity, your food and beverage minimum, and how long a party of eight normally takes makes you the only citable option in most towns.
Make the things diners filter on readable. Put the full menu on your site as text with prices and dietary labels, add restaurant and menu structured data, write down the practical details about groups, noise, parking, patio, and timing, keep holiday hours right and reviews flowing, and hold a live reservation listing. ChatGPT names restaurants it can describe in detail, and a downloadable menu describes nothing.
Almost always because AI knows you exist without knowing what you serve. You are listed everywhere, so your name and your town are safe, but the menu is a file, the dietary labels are in the server’s head, the private room is on a chalkboard, and the specials live in stories that disappeared. Every dish-level, diet-level, and occasion-level question is decided by text you have not published yet.
Your score is free — no credit card. The full report with every fix, prioritized, is a one-time $30.
Emergency calls now start in a chat box. What AI reads before it names a plumber — and the fixes that get you named.
Read the guideNo-heat calls and $8,000 replacements both start with an AI question now. What AI checks before it names an HVAC company.
Read the guidePatients ask AI for a dentist who takes their insurance and sees them fast. What AI needs to read before it says your name.
Read the guideAfter a hailstorm, homeowners ask AI who is local, licensed, and certified. What AI reads before it names a roofer.
Read the guideClients describe the problem to AI before they call a firm. What AI reads before it names a lawyer — and how to be specific without breaking bar rules.
Read the guideThe trade with the most fake local listings anywhere. What AI checks before it trusts a garage door company — and how to prove you’re the real local shop.
Read the guideDo plumbing, HVAC, and electrical? AI may recommend you for none of them. Why a multi-trade site confuses AI — and how to give each trade a real home.
Read the guideThe agent is the product, but the portals own the reputation. What AI reads before it recommends an agent — and how to surface your record where you control it.
Read the guideBusiness owners describe their entity, their software, and their mess — then ask AI who handles it. What AI reads before it names an accounting firm.
Read the guideEvery industry guide in one place — the prompts, sources, and fixes for each.
Read the guideFrom the blog: AI Visibility Gets You Found. Reviews Get You Picked. · Why ChatGPT Ignores Most Local Businesses
The fundamentals: What generative engine optimization (GEO) means for a local business · How to show up in AI answers (AEO)
The free score takes about a minute: overall AI Visibility Score, letter grade, and how many fixable gaps we found on your site.
Check Your ScoreYour score is free — no credit card. The full report with every fix, prioritized, is a one-time $30.