How to Get Your Restaurant Recommended by ChatGPT (and Found on Google)

You fill the dining room every Saturday, your Google listing is well kept, and still, when someone asks ChatGPT "where should I have dinner tonight?", your restaurant doesn't exist. It isn't bad luck: showing up in ChatGPT and getting AI to recommend your restaurant runs on different rules than the restaurant local SEO you already know, and almost nobody is applying them yet.

Short answer: AI doesn't rank listings. It resolves entities and cites external evidence. Six technical moves take any restaurant without a developer from invisible to citable: a consistent NAP, a Bing Places listing, schema.org/Restaurant markup, an HTML menu, review volume and recency, and citable content. None of the six requires paying for tools; all of them require method.

In 60 seconds

How to get AI to find, understand, and recommend your restaurant:

1. 83% of restaurants are invisible in ChatGPT even when they do fine on Google, and only 1.2% of local businesses get recommended by AI versus the 35.9% that show up in the local pack (Local Falcon and SOCi, 2026), so the window is open for the first ones who do the technical work.
2. Unify your name, address, and phone number across Google, Bing, TripAdvisor, OpenTable, and your website: without a consistent entity there is no recommendation.
3. Claim Bing Places and Apple Business by importing your Google listing: 15 minutes, free, and you cover Copilot, part of ChatGPT, and Siri.
4. Add a schema.org/Restaurant JSON-LD block to your website and validate that it says the same thing as the page.
5. Get the menu out of the PDF and the widget: an HTML page with dish, price, and allergens in real text.
6. Work on review volume and recency over the decimal point of your rating, and not only on Google.
7. Audit every month with 5 fixed prompts in ChatGPT and Gemini, and log the trend.

Half of these moves live on your website. If yours was built years ago and nobody knows how to touch it, Plattio's website builder generates the schema, the HTML menu, and the consistent data automatically: the part of the checklist you shouldn't be doing by hand.

How this guide was created

This guide rests on the primary studies of local business visibility in AI published between 2024 and 2026: SOCi's Local Visibility Index, the two Local Falcon analyses of restaurants in ChatGPT, BrightLocal's Local Consumer Review Survey, the MyPlace review study, the DoorDash trends report with Dynata, the Princeton academic paper on Generative Engine Optimization (KDD 2024), the Ahrefs experiments on schema and AI citations, Vercel's measurement of AI crawlers, and the official documentation from Google, OpenAI, and schema.org. For the European picture, SE Ranking's analysis of AI traffic in Spain. This is what the industry calls GEO or AEO: optimizing so that generative engines find you and cite you.

Three honest caveats. Almost all the hard numbers come from the US. The mechanisms transfer, the exact figures don't. The AI search ecosystem changes by the quarter, so we date every perishable claim; last verified: July 2026. And most of these studies measure correlations, not causes; when a figure is a correlation, we say so.

What you should take away from this article

• Certainty: what has actually been measured about how ChatGPT, Gemini, and AI Overviews recommend restaurants, with source and date, separated from the hype.
• Judgment: why showing up on Google doesn't get you into AI, which signals carry weight (review volume, data consistency, readable HTML), and which ones are myth.
• Process: the six technical moves in order, each with its step-by-step, no coding required.
• Checklist: a downloadable CSV with 30 checks and the monthly 5-prompt audit to know whether it's working.

Why is your restaurant invisible to AI even though it shows up on Google?

Because they are two different systems that barely talk to each other. The figure that sums it up best: 83% of restaurants are completely invisible in ChatGPT, versus only 14% invisible on Google, according to Local Falcon (2026, 189,905 ChatGPT results analyzed). SOCi's Local Visibility Index (January 2026, some 350,000 locations, all sectors) confirms it from another angle. ChatGPT recommends 1.2% of locations versus the 35.9% that appear in Google's local pack; Gemini reaches 11% and Perplexity 7.4%.

That isn't a sentence: it's a window. If almost none of your competitors are citable, the first ones who do the technical work split the recommendations between them.

Two names for that work, defined once and for all. Restaurant local SEO is the set of practices for appearing in the search results of your area: a well-kept Google listing, reviews, an optimized website. GEO (Generative Engine Optimization) is its extension to generative engines, optimizing your restaurant's data and content so that ChatGPT, Gemini, or Perplexity can find it, understand it, and cite it in their answers. This guide covers both, because from here on they travel together.

The demand is already here

In the US, 45% of consumers already use AI to find local businesses, up from 6% a year earlier (BrightLocal, March 2026), and 22% of diners have already picked a restaurant by asking an AI (DoorDash with Dynata, May 2026).

And in Europe? The data is thinner, but it points the same way: in Spain, AI-referred web traffic grew 30% year over year, with ChatGPT concentrating 70% of it (SE Ranking, June 2026). It's still a small share of traffic; but that channel doesn't send visits. It sends recommendations that decide dinners.

Why ranking your restaurant on Google doesn't get you into ChatGPT

The starting mistake is assuming AI "reads" Google's ranking. It doesn't. According to SOCi (as covered by Search Engine Land), only 45% of the businesses that win on Google also appear in AI answers. The myth of "if I'm fine on Google, AI already finds me" dies here. Ranking your restaurant on Google is still necessary, but it's no longer sufficient. They are different pipelines, and the rest of this guide explains where the AI ones run.

How do ChatGPT, Gemini, and AI Overviews decide which restaurant to recommend?

Each engine has its own pipeline, but the 2026 studies observe a common three-step pattern. First, entity resolution, establishing that your restaurant exists as a single, consistent entity (same name, address, and phone everywhere). Second, weighting external evidence, what sources you don't control (reviews, directories, editorial lists) say about you. Third, context fit, what matches the specific question (romantic, kid-friendly, gluten-free, in a given neighborhood). Fail the first check and the other two never happen.

Engine by engine: where each AI gets its restaurants

Treating "AI" as a single block is the mistake behind almost everything written on this topic. As of July 2026, the map looks like this:

One nuance about ChatGPT: according to what Search Engine Land observed by asking it about its own process, it doesn't read your Google listing or Bing Places directly. It scans 20-30 search results and filters them down to 3-5 verifiable sources (its shifting relationship with Bing comes up in the Bing Places section).

AI doesn't take your word for it: whom it cites when recommending where to eat

Cloro (July 2026) analyzed 200 dining prompts across 6 engines and measured whom they cite: editorial "best of" lists 40%, user content 32% (Reddit at the top, with 22% of all citations), and booking platforms 25%. The restaurant's own website almost never shows up in that discovery phase. So your job isn't to shout louder on your own site. It's to make sure the sources AI does believe are telling your correct story, starting with all of them saying the same thing about you.

Do your name, address, and phone number tell the same story everywhere?

The NAP (name, address, phone) is your restaurant's proof of identity before the machines. An inconsistent NAP doesn't penalize you: it dissolves you. AI splits your identity across several half-formed entities and none of them reaches the minimum confidence to be cited.

What entity resolution is (and why it affects you)

The knowledge graphs that feed search engines and assistants group scattered mentions into a single entity: if "The Copper Pot" on Google, "The Copper Pot Restaurant - Modern European Kitchen" on TripAdvisor, and "Copperpot" in an old directory also carry two different phone numbers, the system can't assert they're the same place. That's how these graphs work in general (each AI's exact algorithm isn't documented), but the data points in the same direction. According to Local Falcon's Restaurant AI Visibility Index (May 2026, 10,000 US restaurants), 74.9% of restaurants never appear in Google's AI recommendations, and having a claimed profile, your own website, or an active social page is associated in every case with more AI visibility. They're correlations, but they all say the same thing: a complete, consistent entity means more visibility.

Define your master NAP (10 minutes)

Write down, in one document, the official and only version: exact name with no slogan ("The Copper Pot", not "The Copper Pot - Manchester's Best Sunday Roast"), a single way of writing the address, one phone number, one URL. From today, everything you publish uses that version, letter for letter. A sanity note. The real entity killers are different phone numbers, slogan-stuffed names, duplicates, and contradictory addresses; "St" in one place and "Street" in another is not cause for panic. Don't turn this into typographic neurosis.

The free 15-minute audit

Search Google and also Bing (remember: it feeds other AIs) for your old phone number in quotes, your name with its variants, and your address. Every result with stale data is an identity leak: correct it or request its removal. While you're at it, hunt down duplicate listings on Google and TripAdvisor and request their merge or closure: a duplicate with 40 old reviews is a ghost competitor of yourself.

The 4 core listings (and the 50-directory myth)

You don't need to sign up to 50 directories; that advice belongs to another era. Four core listings plus one booking platform cover 80% of the result: Google Business Profile, Bing Places, Apple Business, and TripAdvisor, plus OpenTable, the reservation platform wired directly into ChatGPT (in much of continental Europe, TheFork plays the same role). Across most of Europe and the UK, Yelp is secondary; to go further, Whitespark maintains free lists of the top citation sources by country. Your website is the canonical reference, with footer, contact page, and schema carrying the master NAP, identical. And when something changes (phone, opening hours), protocol: the 4 core listings within 48 hours, a review of the rest every quarter.

Why is Bing Places the back door into ChatGPT?

Nuance first, because stale advice abounds here. In January 2025, 87% of ChatGPT Search citations matched Bing's top results (Seer Interactive); since then that alignment has fallen to single digits and OpenAI has diversified toward other sources, so the dependency is no longer exclusive (Profound). But Bing still counts: an April 2026 study covered by Search Engine Land repeated the same prompt 68 times, and the hotel that ranked well on Bing appeared in ChatGPT about 13 times more often than a comparable competitor. And there is one engine that does read Bing Places directly: Microsoft Copilot, which also feeds Yahoo, Windows Search, and Edge; an unclaimed listing there is guaranteed invisibility. Put it plainly. Bing Places isn't "the only door" into ChatGPT, it's the cheap insurance with the best effort-to-impact ratio in this entire guide.

The 15-minute sign-up

1. Go to bingplaces.com with a Microsoft account (don't worry if it redirects you to bing.com/forbusiness: it's the same tool at a new address).
2. Use "Import from Google Business Profile": it pulls in your complete Google listing in about 5 minutes, and turn on periodic sync so future changes propagate on their own.
3. Verify the listing. If your Google profile is already verified, it's usually automatic; if not, phone or SMS verification takes minutes. Full publication can take 3 to 14 days.
4. Review the 6 fields the AIs read: exact name with no added keywords, address identical to your master NAP, a local phone number, category (Restaurant + cuisine type), a natural 200-300 word description, and the URL of a website with an HTML menu and schema.

The other sign-ups that feed the AIs

An agent being able to close the booking without leaving the chat connects directly with how you look after the guest experience from the reservation onwards. And to size up the whole: the Yext AI citations study (October 2025; 2.2 million citations in food service) attributes 41.6% to listings on third-party platforms and 39.8% to the business's own website: 81% comes from sources you manage directly. The 86% Yext puts in its headline is the average across all sectors and also includes the reviews and social profiles a brand can manage.

What is schema.org/Restaurant and why does your website need to talk to machines?

Schema.org is a standard vocabulary for labeling your website's content so a machine understands it without ambiguity: this is the name, this is the opening hours, this is a dish and this is its price. It's implemented with a JSON-LD block invisible to the visitor.

What it is and what it isn't

There's plenty of snake oil around this one, so let me be blunt. Schema does not guarantee that AI will cite you. Ahrefs (May 2026, 6 million URLs) found that AI-cited pages carry JSON-LD almost 3 times more often, but its causal experiment (1,885 treated pages versus 4,000 controls) detected no improvement in citations 30 days after adding it. Take it for what it is: a readability condition, not a citation guarantee. In its favor, one direct signal. Microsoft stated in March 2025 that schema helps its LLMs understand content.

And one surprising fact: Google only requires two fields to mark up a restaurant: name and address. Everything else (opening hours, cuisine, prices, menu) is "recommended". And it's exactly what an AI needs in order to recommend you.

The complete JSON-LD block, ready to copy

A complete example for a fictional restaurant, "The Copper Pot" in Manchester. Replace each value with your own (using your master NAP) and insert it in the head of your homepage. Three details: menu and hasMenu coexist on purpose, because Google documents menu but schema.org marks it as superseded by hasMenu; suitableForDiet is what answers "gluten-free restaurant in Manchester"; and acceptsReservations with a URL tells an AI agent where to book. One safety caveat before you label anything. Marking a dish as GlutenFreeDiet amounts to a "gluten-free" claim regulated by EU Regulation 828/2014, which also applies to food served in restaurants and demands a ceiling of 20 mg/kg of gluten, so use it only on dishes your kitchen can guarantee with cross-contamination control; otherwise, declare the allergens and the risk of traces instead.

{
  "@context": "https://schema.org",
  "@type": "Restaurant",
  "@id": "https://www.thecopperpot.co.uk/#restaurant",
  "name": "The Copper Pot",
  "description": "Modern European bistro in Manchester's Northern Quarter: wood-fired cooking, seasonal plates, and low-intervention wines.",
  "url": "https://www.thecopperpot.co.uk/",
  "image": "https://www.thecopperpot.co.uk/img/copper-pot-frontage.jpg",
  "telephone": "+441612345678",
  "priceRange": "££",
  "servesCuisine": ["Modern European", "British", "Wood-fired"],
  "acceptsReservations": "https://www.thecopperpot.co.uk/reservations",
  "address": {
    "@type": "PostalAddress",
    "streetAddress": "42 Edge Street",
    "addressLocality": "Manchester",
    "addressRegion": "Greater Manchester",
    "postalCode": "M4 1HW",
    "addressCountry": "GB"
  },
  "geo": {
    "@type": "GeoCoordinates",
    "latitude": 53.48430,
    "longitude": -2.23690
  },
  "openingHoursSpecification": [
    {
      "@type": "OpeningHoursSpecification",
      "dayOfWeek": ["Tuesday", "Wednesday", "Thursday", "Sunday"],
      "opens": "12:00",
      "closes": "16:30"
    },
    {
      "@type": "OpeningHoursSpecification",
      "dayOfWeek": ["Friday", "Saturday"],
      "opens": "12:00",
      "closes": "23:30"
    }
  ],
  "menu": "https://www.thecopperpot.co.uk/menu",
  "hasMenu": {
    "@type": "Menu",
    "name": "The Copper Pot menu",
    "inLanguage": "en",
    "hasMenuSection": [
      {
        "@type": "MenuSection",
        "name": "Mains",
        "hasMenuItem": [
          {
            "@type": "MenuItem",
            "name": "Wood-roast chicken with charred leeks",
            "description": "Free-range half chicken, smoked butter, roast potatoes. For two.",
            "offers": { "@type": "Offer", "price": "17.50", "priceCurrency": "GBP" },
            "suitableForDiet": "https://schema.org/GlutenFreeDiet"
          },
          {
            "@type": "MenuItem",
            "name": "Braised beef shin with mash",
            "offers": { "@type": "Offer", "price": "18.00", "priceCurrency": "GBP" }
          }
        ]
      },
      {
        "@type": "MenuSection",
        "name": "Starters",
        "hasMenuItem": [
          {
            "@type": "MenuItem",
            "name": "Heritage tomato and whipped feta salad",
            "offers": { "@type": "Offer", "price": "9.50", "priceCurrency": "GBP" },
            "suitableForDiet": "https://schema.org/GlutenFreeDiet"
          },
          {
            "@type": "MenuItem",
            "name": "Charred hispi cabbage",
            "offers": { "@type": "Offer", "price": "8.00", "priceCurrency": "GBP" },
            "suitableForDiet": ["https://schema.org/VeganDiet", "https://schema.org/GlutenFreeDiet"]
          }
        ]
      }
    ]
  },
  "sameAs": [
    "https://www.instagram.com/thecopperpot",
    "https://www.facebook.com/thecopperpot",
    "https://www.tripadvisor.co.uk/thecopperpot"
  ]
}

And if "insert it in the head" sounds like gibberish? You don't need to code: in WordPress you paste it with a snippets plugin or from your SEO plugin (Rank Math and Yoast accept custom code per page), and Wix and Squarespace have a native head code field in their advanced settings. If none of that rings a bell, this block is, word for word, the exact request to send to whoever maintains your website.

Data parity: everything in the schema, visible on the page

Non-negotiable rule: every piece of data in the JSON-LD must also appear in the visible HTML. When ChatGPT visits your website live (with its ChatGPT-User agent), it reads the HTML a human sees and may ignore the JSON-LD, according to the searchVIU analysis covered by Ahrefs. Opening hours or prices in the schema but not on the page is the worst combination: readable for some machines, invisible to others, and suspicious to Google.

Validate for free in 5 minutes and let the bots in

Two validators, not interchangeable: Google's Rich Results Test checks what produces rich results on Google (and doesn't evaluate Menu/MenuItem), and validator.schema.org validates the full vocabulary, menu included. Pass both. And open the door. Check that your robots.txt doesn't block OAI-SearchBot, because blocking it removes you from ChatGPT's searches; it's a different bot from GPTBot, which only affects training.

What's still alive and what's dead (July 2026)

Menu in PDF or HTML? The decision AI has already made for you

Your menu is, in Yext's words, "the densest set of answers a restaurant owns": it answers what you cook, what it costs, what gluten-free options you have. The question is whether machines can read it.

The honest diagnosis: indexable, structured, and citable are not the same thing

Let's dismantle the usual myth first: Google does index PDFs, and AI crawlers can extract the text from a non-scanned PDF. The problem isn't reading. It's structure. A PDF doesn't tell the machine which price belongs to which dish or which allergen label goes with which preparation. Indexable doesn't mean structured, and structured doesn't mean citable.

And there is a second, bigger problem. No major AI crawler executes JavaScript. Vercel measured it on its own network, and found that GPTBot and ClaudeBot download JavaScript files, but never execute them (the exceptions are Gemini, which uses Google's infrastructure, and AppleBot). The same holds for OAI-SearchBot, the crawler that actually decides whether you show up in ChatGPT Search, which doesn't execute JavaScript either, while GPTBot only handles training. So the "invisible menu" conclusion rests on the search crawler, not just on GPTBot. If your menu lives in an embedded PDF viewer or a widget that loads with JavaScript, as far as ChatGPT is concerned your restaurant has no menu; the "but it's already on my website" myth dies here. And the volume justifies taking it seriously. GPTBot alone already generates around an eighth (about 12.6%) of Googlebot's monthly requests, according to the same measurement.

The 10-second test

Open your menu page, press Ctrl+U (view source), and search for the name of a dish. If it appears as text, machines can read it. If it doesn't, your menu is invisible to AI, no matter how good it looks on screen.

Migrating without rebuilding the website: 7 steps

1. Create a /menu page in HTML: one heading per section and, for each dish, a name, a one-line description, a price, and labels.
2. Add the allergens as text on every dish. EU Regulation 1169/2011, retained in UK food law after Brexit, already obliges you to declare the 14 allergens: in HTML, the legal obligation and the signal AI needs are the same piece of work.
3. Demote the PDF to a "printable version" linked at the end; if the PDF had an indexed URL, 301-redirect it to /menu.
4. Add the Menu/MenuItem JSON-LD (from the block above) and validate it at validator.schema.org.
5. Pass the Ctrl+U test.
6. Propagate the URL: the "menu link" field on your Google listing, Bing Places, TripAdvisor, and OpenTable.
7. Measure at 90 days in Search Console the impressions for searches with dish names and "menu + your neighborhood".

Does it work? The best-documented case is Les 3 Canonniers, a 50-seat restaurant in Nantes, which posted +178% organic sessions in 10 months and +32% direct reservations after an overhaul whose core was structuring the menu in HTML, reported by the agency that signed the project: an agency case, not an independent study, but the direction matches everything above. And if maintaining a double menu (HTML for machines, PDF for printing) sounds like extra work, that's exactly what Plattio's digital menu automates. The menu is served as HTML from the server, with the JSON-LD and allergen labels generated automatically, and the PDF remains as a printable export.

What do AIs read in your reviews?

This section is only about the signal your reviews send to AI. The work of earning them (when to ask, the QR code, the negative ones) is covered in full in our guide on how to get more Google reviews; we won't repeat it here.

Volume over rating: the finding that dismantles the obsession with decimal points

The most-cited 2026 study on reviews and AI is MyPlace's (February 2026, 230 restaurants in 5 US cities). The restaurants AI recommends average 3,424 Google reviews; the ones it ignores, 955, with only a 0.03-star difference in rating. Below roughly 1,000 reviews, almost no venue appeared. It's a correlation with a small sample, not a law, but its two extreme cases are eloquent: Provare, with 4.8 stars and 456 reviews, zero recommendations; Canlis, with 4.4 and 5,213, recommended by all four AIs tested. If your plan was to suffer your way from 4.6 to 4.8, change it, because above ~4.4 stars volume rules, and the gap that matters is yours versus the recommended restaurants in your city.

Recency, text, and consistency

AI doesn't just count reviews: it reads them. Birdeye (May 2026) describes how the engines build a semantic profile from the text. If twenty independent reviews mention "the best carbonara in the neighborhood" and "quiet terrace", that's what AI will learn to say about you. Recency weighs just as much. Whitespark's 2026 local search ranking factors place sentiment and reviews from the last 90 days among the signals associated with AI visibility, and 74% of consumers only take into account reviews less than 3 months old (BrightLocal, 2026). It's legitimate to ask for the review at the moment of the signature dish so the guest mentions it in their own words; writing it for them or buying it, never: that's the red line of this whole system.

The owner's reply, a reinforcement signal

Replying to 100% of reviews, mentioning the praised attribute and without templates, reinforces twice: before the consumer (80% favor businesses that respond to everything, according to BrightLocal) and before the machine, which sees confirmed in your voice what the independent voices are saying.

Don't concentrate everything on Google

Each engine drinks from different sources: Gemini from Google Maps, Perplexity from Yelp and Tripadvisor. And the least-known nuance: ChatGPT can't read the stars on your Google listing, because they load via JavaScript; it sees your reflection on Bing, in directories, and on your website. Among the review sources ChatGPT cites via Bing's index, Whitespark measured that Facebook beats Yelp and that TripAdvisor climbs the ranks in dining. The takeaway for a European independent: Google plus TripAdvisor plus your booking platform, and testimonials in real text on your own website.

How do you write content AI can cite?

This is where the reference academic paper, Princeton's on Generative Engine Optimization (KDD 2024), provides the numbers: adding quotations, statistics, and sources to a page improved its visibility in generative answers by up to 40% in its experiments (measured on general domains, not dining: we're not promising you that number). Its most hopeful finding is that the page in fifth position gained +115% visibility; that is, GEO favors the small player. And the negative one: keyword stuffing scored -8%. Repeating "best restaurant in Manchester" twenty times costs you.

The "answer first" rule and how real people actually prompt

Real prompts aren't the ones you imagine. The average prompt in ChatGPT's search mode is 8.7 words long (Semrush), and around 60% are questions, with "best" as the star word (Stella Rising, via Search Engine Land). The structure that answers that: headings phrased as short questions, a self-contained 40-60 word answer with one hard fact in the first paragraph and the detail afterwards, all in plain HTML; the pattern of this guide. And the context that makes it urgent: with an AI summary on Google, only 8% of users click a result, versus 15% without one (Pew Research, July 2025). Either the answer cites you, or you don't exist.

The 7 citable pieces of a restaurant website

Discovery vs verification: the distinction almost nobody explains

Go back to Cloro's citation split. On discovery questions ("best restaurant in X"), AI cites lists, reviews, and third-party platforms; your website almost never. But when the diner asks about you ("does restaurant X have outside tables?", "average price at X?"), the tables turn and your website becomes the main verification source. That match does depend 100% on you: the 7 pieces above are your starting eleven. One efficiency note. ChatGPT cites on average about 15 sources per answer and Gemini only 3 (Semrush); in Gemini, you're either a top-3 source or you're out.

The anti-patterns, to close: keyword stuffing (the -8% from the Princeton paper), marketing prose without a single fact ("a magical corner where flavors come alive" answers no question), and important information that only lives on Instagram or in a PDF. And one signal that's hard to fabricate but real: unlinked brand mentions correlate three times more with AI presence than backlinks (Ahrefs, 75,000 brands); being talked about in the local press and on Reddit is worth more than a link exchange.

How to check in 10 minutes whether AI recommends your restaurant

An AI visibility audit requires no paid tool: 5 prompts, 2 engines, a temporary chat, and 10 minutes a month.

Preparation

Open ChatGPT in a temporary chat, which uses no memories or history; for the most neutral result, also disable memory or test logged out. Repeat everything in Gemini in incognito, which is grounded in Google Maps, so its answers and ChatGPT's come from different pipelines. And keep in mind that each answer usually names only 3-5 restaurants and closes the list (Bloom Intelligence, July 2026), so not appearing in the generic one is the statistical norm, not a verdict.

The 5 prompts (copy them verbatim, replace the brackets)

1. Generic discovery: "What are the best restaurants in [neighborhood or city]?" This one measures raw brand strength; not appearing here is normal.
2. Category + area: "Where can I get good [your specialty: Sunday roast, tapas, ramen…] in [neighborhood]?" This is your core business query: here you should appear.
3. Occasion: "Looking for a restaurant for [a romantic dinner / a work lunch / going with kids] in [area]". It tests whether your reviews and website describe attributes and atmosphere.
4. Specific need: "Which restaurants with gluten-free options are near [square or landmark]?" The long tail, and your most achievable win if the menu is in HTML with labels.
5. Direct verification: "Tell me about [name] restaurant in [city]: opening hours, type of cuisine, is it worth it?" It doesn't measure whether you appear, but whether what AI believes about you is true.

Two tricks: if the answer used web search and mentions you, ask "Which sources did you use for this recommendation?" and note the URLs (only when there are real citations; without web search, the model will make sources up). And if you're in a tourist area, repeat prompt 1 in your visitors' main languages.

The 4 scenarios and what to do in each one

The monthly routine

The same day each month, the same 5 prompts word for word, in the same two engines, and the results logged in the audit rows of the CSV checklist. LLMs are not deterministic, so the same question can give different answers, and you evaluate the three-month trend, never a one-off absence; "I tried it once and I wasn't there" is not a diagnosis. If you want tools, two genuinely free ones exist: HubSpot's AEO Grader (no account; it evaluates the model's knowledge, not live search, and it may return "unknown brand" for a small restaurant, which is itself the diagnosis) and the free plan of LLMrefs (1 keyword across 8+ engines, no card). Paid ones make sense with several venues; for an independent, the manual audit is more reliable and free. Fold the monthly result into the rest of your growth metrics.

The complete checklist: the technical requirements and how to cover them without writing code

Everything above, condensed into the checklist's 7 blocks:

Download the checklist. AI visibility checklist for restaurants in CSV. Open it in Excel or Google Sheets: 30 checks with their verification method, priority, and estimated time, including the 5 prompts as the monthly audit rows. Two honest notes about its limits: it's a work log, not a guarantee of appearing; and nobody has measured how long a correction takes to show up in AI, so note the date of every action and compare it against the next two audits instead of waiting for deadlines no study backs.

And the motivation to start today is in the Yext figure you already saw. In food service, 81% of what AI cites comes from listings and websites you manage directly. A good share of that is your website: correct schema, an HTML menu, data identical to your listings. That's exactly what Plattio's website builder generates without you touching code. The site is published with the Restaurant and menu JSON-LD already in place, and changing an opening hour or a dish updates at once what your guests see and what the machines read. Your work stays where no software can do it for you: the listings, the reviews, and a kitchen people want to write about.

Put AI to work filling your dining room

83% of your competitors are invisible to ChatGPT today. With the six moves in this guide and a 10-minute audit each month, you can be on the other side of that statistic before it stops being an advantage.

Try Plattio in your restaurant · See a guided demo