Make sure every source those tools read says the same correct thing about you, then test what they actually say. That means one name, address, phone number and set of hours across Google, Bing, Apple and review sites, a text menu on your own website, a complete Business Profile and an honest flow of reviews. Nobody can guarantee a recommendation.
Google says AI Overviews and AI Mode need no special optimisation beyond normal SEO, and OpenAI does not publish which listing databases sit behind ChatGPT's local answers. That is why checklists only get you so far. Start with a test instead: twelve questions diners ask, run three times in each tool and logged in a sheet. The log shows what the tools say about you, which errors they repeat and which competitors they name, and the fixes follow from that.
Test first: twelve diner questions, three runs each
Build the list from how guests really ask. Cover the situations that decide a booking: the category in your neighbourhood, an occasion, a dietary need, a time of day, a group size, a signature dish, your own name and a head-to-head with a competitor. Keep the wording natural, and keep it identical every month.
| Type | Example wording (swap the brackets) |
|---|---|
| Category and place | "Best [cuisine] restaurant in [neighbourhood]" and "Good-value [cuisine] in [town] for a weeknight" |
| Occasion | "Where should I take my parents for a quiet birthday lunch near [landmark]?" |
| Dietary need | "Gluten-free friendly [cuisine] restaurant with outdoor seating in [town]" |
| Time of day | "Which [cuisine] restaurants near [landmark] are open late on a Sunday?" |
| Group | "Restaurant in [neighbourhood] that takes bookings for a group of ten" |
| Dish | "Where can I eat [signature dish] in [town]?" |
| Brand | "What is [restaurant name] in [town] like? What's on the menu and when is it open?" |
| Head to head | "[restaurant name] or [competitor] for a date night?" |
Use two wordings for category, occasion, dietary need and brand, and one for the rest: twelve questions in all. Pick three or four tools and keep them fixed: ChatGPT with search switched on, Google's AI Mode and the AI Overview on a normal search, Gemini, and one of Copilot or Perplexity. Run each question three times, in a fresh chat each time, on different days if you can. Location shapes the answers, so run the test from the same place each month, ideally somewhere your diners start from rather than your own dining room.
For every run, log six things: the date, tool and question; whether you were named; your position in the list; the sources shown or cited; what it said about you (hours, prices, menu, dietary claims); and which competitors appeared. A spreadsheet with those columns is enough. Three runs matter because the same question gives different lists on different days. "Named in two of three" is a finding. "Named once" is noise.
A quick sum: twelve questions, three runs and three tools is 108 runs. At about 40 seconds to run each one and another 40 to log it, that is 144 minutes, or about two and a half hours a month. Do it monthly, and again after any change to your hours, menu or listings.
What the tools say they use, and what is guesswork
OpenAI's help page for ChatGPT search says it can rewrite your question into one or more targeted search queries, use an approximate location taken from your IP address (device location sharing is optional and off by default), and share location with trusted third-party providers to bring in local information such as business listings. It also says reservation availability can come from OpenTable, Resy or Yelp, depending on the restaurant. What it does not do is name the listing databases behind local answers. It says ChatGPT search "sometimes partners with other search providers" and links Microsoft's privacy statement among them, which fits the claim in several restaurant-software guides that Bing's index is one source. Those vendors have something to sell and OpenAI names no listing database, so I would treat Bing as likely, not proven.
Google is clearer about the shape of it and vaguer about the detail. Its guidance says AI Overviews and AI Mode need no special optimisation beyond normal SEO, and that llms.txt files and special markup are not used for them. For local questions, commentators report that AI Mode leans on the same listings and reviews as Maps and the local pack, which fits Google's long-standing description of local ranking: relevance, distance and how prominent a business is. Google hasn't published a restaurant-specific list that I could find, so read the rest as reasonable inference, not rule.
Put together, that leaves a short list of things you can influence: the facts each source holds about you, the text on your own site, the reviews and mentions on other sites, and whether the crawlers can read you. What you cannot influence is how the tools weigh them. That is why I would rather you test than trust any checklist, this one included.
A first scorecard: a 36-seat Thai restaurant
The figures below are invented, but the shape is typical of a first test. The restaurant runs the twelve questions three times each in ChatGPT, Google AI Mode and Gemini, 36 runs per tool. Nine of those runs use the restaurant's own name, so it is named by construction. The number that matters is how often it appears in the other 27.
| Tool | Named, all 36 runs | Named in the 27 open questions | Wrong facts in the 9 brand answers |
|---|---|---|---|
| ChatGPT | 14 | 5 (19%) | 4: hours (3), an old price (1) |
| Google AI Mode | 21 | 12 (44%) | 1: a price |
| Gemini | 16 | 7 (26%) | 3: hours and an old menu |
Now the diagnosis. Of the four wrong facts in ChatGPT's brand answers, three said the restaurant was closed on Mondays and one quoted an old price. Its Business Profile is right, but its Tripadvisor page still carries last year's hours, and that page was cited beside the answers. The old price traces to a PDF menu on the restaurant's website. In the open questions, the main competitor was named in 19 of 27 ChatGPT runs, with two "best of" articles, a directory and Tripadvisor cited beside those answers. The restaurant's own site appeared once.
The fix list that follows from it, in order: correct the Tripadvisor hours first, because it is the cited source; claim the Bing Places listing and copy the details across; check Apple Business; replace the PDF menu with a web page; add dietary and group-booking facts to the site and the profile where they are true; and pitch one local writer a story.
After six weeks, the same test shows the wrong facts in ChatGPT's brand answers falling from 4 to 0, and the open-question count moving from 5 to 7 of 27. Take the first result seriously, because you can verify it. Do not claim the second. A change of two runs out of 27 is inside the normal day-to-day wobble.
Fix the facts AI repeats: name, address, phone and hours
Start with one plain-text file, your source of truth, and copy from it every time. Write the name exactly as it appears on your sign, the address in one fixed format, the main phone number, opening hours including holiday hours, cuisine, price band, booking link, dietary notes, parking and step-free access.
Name: [trading name exactly as on the sign]
Address: [street address, one fixed format, no variations]
Phone: [main line]
Hours: Mon-Thu 12:00-22:00; Fri-Sat 12:00-23:00; Sun 12:00-21:00
Holiday hours: [dates and hours]
Cuisine: [two words]
Price band: [as you want it described]
Bookings: [booking link]
Dietary: vegetarian and vegan dishes marked; gluten-free options on request; no claims about allergen-free preparation
Access: [step-free entrance, parking]Then check the places the tools actually read: your Google Business Profile; Bing Places (bing.com/forbusiness); Apple Business (business.apple.com, formerly Apple Business Connect), which feeds Apple Maps; Tripadvisor and Yelp; your booking platform; Facebook; your website's footer and contact page; and any delivery app. The mismatches are usually small and old.
| Where | What to check | A typical mismatch |
|---|---|---|
| Google Business Profile | Name, hours, categories, menu link, attributes | A secondary category that describes a different cuisine |
| Bing Places | Claimed, with the same details as Google | An unclaimed listing carrying a former owner's hours |
| Apple Business | Name, phone, hours | The old phone number after a change of provider |
| Tripadvisor and Yelp | Hours, address format, menu link | "Closed Mondays" copied from last year |
| Booking platform | Sittings, party sizes, hours | Sittings that no longer exist |
| Website, phone, hours | An old domain in the website field | |
| Website footer and contact page | Exactly your source-of-truth file | The address written two ways on two pages |
Here is how stale facts linger, in an invented but common case: a restaurant moves two streets along and updates Google within a week. Six months later ChatGPT still sends a diner to the old address and cites a Tripadvisor page that nobody remembered. It was the one place nobody had updated. The edit took ten minutes, but only the test showed where the wrong address came from.
Do not rename your Business Profile to add keywords or a neighbourhood. Google's guidelines ask for your real-world business name. Put the neighbourhood in the description and in your website's page titles and headings instead. The profile itself, including hours, menu links and photos, is covered step by step in restaurant Google Business Profile: AI-drafted menu, hours and photos.
Put the menu on your website as text
A menu that exists only as a PDF or a photo is the weakest form for any tool. It is hard to read, impossible to update in one place and easy for old copies to linger. Use one web page per menu (dinner, lunch, drinks, children's), with each dish as text: the name, a one-line description in the words guests use, dietary tags and a price, plus a line saying when prices were last updated. Link the pages from your home page and your Business Profile.
The difference is easy to see. Before: an image called "Menu Nov.pdf" with the dish names set in a decorative font. After: "Pad kra pao: holy basil, chilli and garlic stir-fry with minced pork or tofu and a fried egg, served with rice. Can be made gluten-free on request. $14". The second version answers "where can I get a spicy basil stir-fry with a fried egg?" in a form a tool can quote.
If you keep your menu in a spreadsheet, AI can build the page, as long as you stop it adding anything:
Turn this menu into a plain web page. One heading per course. For each dish: the name in bold, then a one-line description, then dietary tags in brackets, then the price.
Use only the information I give you. Do not add ingredients, dietary claims, adjectives or dishes.
Menu: [paste]What I would check in the result: every dietary tag against the kitchen's own sheet, every price against the till, and every ingredient the description mentions. Tools quote dietary tags back to diners, and an invented ingredient becomes a customer's allergic reaction or a complaint. Avoid claims such as "nut-free kitchen" unless that is literally true and documented. If you also print menus or use QR codes, build them from this same text so the versions never drift; the Canva route is covered in AI restaurant menu maker: printable and QR menus in Canva.
Add a short plain-text "good to know" section as well: parking, children, dogs, step-free access, group bookings, private dining, corkage. Do not build an FAQ page hoping for rich results, because Google ended FAQ rich results on 7 May 2026. The point of the section is that a diner, or a tool answering a diner, finds the answer in words.
Reviews and mentions: what you can steer and what you can't
Reviews are probably the richest text about a restaurant that any tool can read, and mentions on other sites are the independent evidence that you exist and are good. You can steer both a little. Ask every guest the same way, because Google's rules bar review gating and incentives and, since April 2026, staff review quotas and requests for reviews that mention specific content such as a team member's name. Reply to reviews using the specifics the reviewer gave, such as the dish; that puts accurate dish names on your profile without breaking any rule. The reply routine is in restaurant reputation management: what to automate and what not.
Mentions are slower. Look at the sources shown beside your competitors' names in your test and note which local articles and "best of" lists they are. Then find out how those pages choose. Some take submissions, some follow awards, and some write about openings, new menus and local suppliers. Send writers something worth writing: a seasonal menu, an anniversary, a producer you buy from. There is no shortcut, and I would be wary of anyone who sells a guaranteed place on a list.
Leave the crawlers alone and skip the tricks
Blocking OpenAI's OAI-SearchBot keeps a site out of ChatGPT search answers. GPTBot is a separate crawler used for training, and the two are independent settings in robots.txt, so you can allow one and refuse the other. Some website builders, hosting firms and security plugins block AI crawlers by default, so open yourdomain.com/robots.txt and read it. You are looking for a blanket "Disallow: /" under all crawlers, or a block naming OAI-SearchBot. A change can take about 24 hours to be picked up. This is one way to allow search and refuse training, and the choice is yours:
User-agent: OAI-SearchBot
Allow: /
User-agent: GPTBot
Disallow: /Three tricks to skip. Restaurant markup (a type of LocalBusiness data holding hours, cuisine, price range and menu link) is a sensible extra if your website platform or an SEO plugin adds it for you, but Google says special markup is not used for AI Overviews or AI Mode and I could not confirm that ChatGPT reads it, so do not pay for it as an AI fix. An llms.txt file has no confirmed effect either: Google says it is not used for AI Overviews or AI Mode, and I found no confirmation that OpenAI's search reads one. And do not ask AI to produce forty near-identical pages such as "best [cuisine] near [suburb]": Google's spam policy, updated on 28 August 2026, lists mass AI pages without added value as scaled content abuse.
Reading the first test: symptom, likely source, fix
| What you see | Likely source | What to do |
|---|---|---|
| Closed on a day you open | Stale hours on a directory or social page the answer cites | Correct the cited page first, then every listing; check holiday hours |
| An old menu or old prices | A PDF, an old page or a menu aggregator | Publish a dated web menu; update or remove the old copy |
| A competitor named, you not | Their "best of" mentions and review volume | Earn mentions and reviews steadily; this takes months |
| Named, but described wrongly | Your categories, description or review themes | Fix the primary category, description and attributes |
| Missing from dietary or occasion questions | No page or profile says you cater for it | Add accurate facts to the site and profile, only where true |
| Wrong phone or address | A duplicate or former listing | Claim, merge or remove it at the source |
| A different restaurant with a similar name | A name clash | Add the neighbourhood to your site's titles and the profile description |
Start at the top. Wrong facts are the one category where a fix is direct and checkable, so work through those before you chase the harder rows.
Tracking without a paid tool
Two free reports help alongside your own log. Google Search Console's Generative AI performance report, available to all sites by 31 August 2026, shows impressions only and combines AI Overviews with AI Mode, so it tells you whether your pages are being surfaced, not how diners reacted. Bing Webmaster Tools has an AI Performance report, in public preview since 10 February 2026, which shows citations and the grounding queries behind them. Both need your website verified, and both describe your website, not your listings, so a restaurant whose main presence is a Business Profile will see little there.
The manual log therefore stays the main record. Keep a one-page monthly summary: questions where you were named, wrong facts found, wrong facts fixed and listings changed. For a wider method for working out whether AI search sends customers, see how to track whether AI search is sending you customers.
What stays out of your hands
Some limits are structural. The same question gives different answers on different days and for different people, so no result is permanent. A newly opened restaurant with few reviews and no mentions starts behind, and no setting changes that. Tools without live search answer from older data, so a recent move or new menu may not appear in them for a while. And a restaurant can do everything above and still be passed over for a generic "best restaurant" question in a crowded neighbourhood.
My advice is to aim at the specific questions, the dietary, occasion, dish and group ones, where an accurate page and profile can decide the answer, and to treat the generic ones as a bonus. If you want the general version of this for any local business, how to get your business recommended by ChatGPT and AI search covers it.
Questions restaurant owners ask about AI recommendations
Can I pay to be recommended by ChatGPT?
I can't confirm any route for a restaurant to buy a place in ChatGPT's organic answers. OpenAI's pricing page says its Go plan "may include ads", but that is separate from the answers themselves, and I haven't verified how a local restaurant would buy one. Be wary of anyone promising a guaranteed recommendation, because nobody controls the answers.
What if ChatGPT gets my restaurant wrong?
Find the source. Ask which pages the answer cited, look for the wrong fact there, and correct it at that source first, whether that is a directory listing, an old PDF menu or a social profile. Then correct your Business Profile and Bing listing. Reporting the answer inside the tool is worth a try, but fixing the source is the lasting remedy.
Is it worth claiming a Bing Places listing?
Probably, because it is a small job. OpenAI does not say which listing databases ChatGPT uses, but its help page lists Microsoft among its search providers, and several vendor guides say Bing's index is one source. Claim it at bing.com/forbusiness, copy your details exactly from your source-of-truth file and keep the hours current.
Can AI write my website's menu descriptions?
Yes for first drafts, but the kitchen must confirm every ingredient and dietary label before anything goes live. AI tools invent plausible ingredients, and other tools then repeat your page to diners. Never let a tool decide an allergen or dietary claim; that sign-off stays with a person.
Further reads
- What AI Can and Can't Do for Your Local SEO — What AI can and cannot do for local search in general.
- AI SEO for Plumbers: Get Named by ChatGPT and Google AI Overviews — The same question answered for a trade business.
- How to Write Menu Descriptions With AI That Still Sound Like You — Menu wording that still sounds like you.
- AI Visibility for Hotels: Get Recommended in ChatGPT Answers — Free checks and a fix list for a small hotel: what ChatGPT and Google AI say about you today, where they get it, and what you can and cannot change.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI's ChatGPT search help page and ChatGPT pricing page (checked October 2026); Google's published guidance on AI Overviews, AI Mode and local ranking; Bing Places and Apple Business pages; Search Console and Bing Webmaster Tools AI reports; restaurant-software vendor guides read as commentary only.