How to check if ChatGPT recommends your restaurant

Most restaurant owners have tried it once. They open ChatGPT, type "best restaurants near me" or their own name, and either feel relieved or alarmed by the answer. Neither reaction is reliable. A single answer from a single assistant, asked once, from one phone, is an anecdote — not a measure of how often guests hear your name.

The questions guests type are changing shape. In the 28 days to 30 October 2026, 43 of the 119 queries Google Search Console listed for this site were written as full questions ending in a question mark — "is ai search replacing google for restaurant discovery" among them. People are asking search engines and assistants the way they would ask a friend. A check worth doing asks the same way.

AI assistants do not return a fixed list the way a results page used to. Ask the same question twice and the shortlist can change. Answers shift with wording, with the location the assistant believes you are in, with whether you are signed in, and with what the assistant remembers from earlier conversations. The underlying models and search systems are also updated regularly.

That does not make checking pointless. It means treating each answer as a sample and reading the pattern across many samples. There is no "position one in ChatGPT" to report, and anyone quoting one is describing a screenshot, not a ranking.

Start with ten to fifteen questions, written in the words guests use rather than the words on your menu. Cover five kinds:

1. 01Name questions — "Is [your restaurant] open on Sunday evenings?", "Does [your restaurant] have vegan options?"
2. 02Category and area — "Best seafood restaurant in Howth", "Good Italian near Grand Canal Dock"
3. 03Occasion — "Somewhere for a 12-person birthday dinner in Dublin 2", "Romantic restaurant for an anniversary in Galway"
4. 04Constraint — "Gluten-free friendly restaurant near Stephen's Green", "Restaurant near the Aviva for a pre-match dinner"
5. 05Comparison — "[Your restaurant] or [a nearby competitor] for a business lunch?"

The wording does not need to be invented. Three places hold it already: the queries report in Search Console, the search terms in your Business Profile Performance report, and the questions your team hears on the phone and in reservation notes. If guests keep asking whether you take groups of twelve, that is a question for the list.

Consistency is what turns answers into evidence. Change the method between months and you cannot tell whether the result moved or the test did.

Fifteen questions across four assistants, asked twice, is 120 answers. That sounds like a lot; in practice it is a focused hour, and it produces a record you can compare against next month.

Most owners record only whether they were mentioned. The other two columns are where the useful work comes from.

Named

Whether your restaurant appeared in the answer at all, and which competitors appeared alongside or instead of you.

Answer accuracy

Whether what the assistant said about you was true today: hours, address, cuisine, price level, dietary options, booking method, whether you are open at all. A wrong answer about a restaurant that is named is often more damaging than not being named.

Sources

Which pages the answer relied on. ChatGPT search, Perplexity and Google's AI features usually show the links they drew from. Click them — they tell you exactly which part of your public record the assistant is reading.

Turn the log into two numbers

A month of answers is easier to compare when it is reduced to two rates, kept separately for each assistant. Do not average them into a single "AI score": an assistant that names you often but gets your hours wrong needs different work from one that never names you.

There is no published benchmark for either rate, and anyone quoting an industry average is guessing. The useful comparison is your own last month, and the competitors named in your place.

Assistants rarely invent restaurant details from nothing. When an answer is wrong, there is usually a page somewhere on the web saying the wrong thing. Fix the page and the answer has a chance to follow. Arguing with the assistant does not change what it tells the next guest.

The last row is the most common and the most fixable. An assistant cannot recommend you for private dining, Sunday lunch or a pre-theatre menu if nothing it can read says you offer one. Our guides to private dining pages and menu pages that rank cover what those pages need.

Each AI search product uses its own crawler, and each one obeys your robots.txt file separately. A rule written years ago to block "bots", or a security setting on your hosting, can quietly stop an assistant from reading your menu and hours — and it will then rely on whatever other sites say about you.

Two details are worth knowing. First, these search crawlers are separate from the crawlers used to train AI models. OpenAI's GPTBot and Anthropic's ClaudeBot, for example, can be blocked without affecting whether you appear in their search answers. Second, Google states that there are no extra requirements for appearing in AI Overviews or AI Mode: a page needs to be indexed and eligible to show a snippet in ordinary Search. There is no special file, tag or markup that buys inclusion.

Source: OpenAI — Overview of OpenAI crawlers — OAI-SearchBot, GPTBot and ChatGPT-User, and what each one controls.

Source: Perplexity — Perplexity crawlers — PerplexityBot and Perplexity-User, with published IP ranges.

Source: Anthropic — How site owners can manage Claude's crawlers — ClaudeBot, Claude-User and Claude-SearchBot, and what disabling each one does.

Source: Google Search Central — AI features and your website — How AI Overviews and AI Mode use website content, and the controls available.

Whether to decline AI training is a business choice, not a visibility requirement. The search crawlers above are the ones that decide whether your pages can appear in answers.

What a Dublin sample shows about readability

Crawler access is only half of it; the page also has to say something a machine can read. When we checked the public homepage HTML of 62 Dublin restaurant websites for our Dublin Restaurant Visibility Report on 3 October 2026, 52 of the 62 (84%) had no restaurant structured data, 4 served a homepage with almost no readable text before scripts ran, and 5 linked to their menu as a PDF from the homepage. That check did not test robots.txt or AI answers, so it does not show how often those restaurants were named. It does show how often the basic facts an assistant needs are missing from the restaurant's own page.

Run the same questions on roughly the same day each month. Progress looks like three slow movements: named in more of the questions, fewer inaccurate answers, and your own website and Business Profile showing up more often among the sources. Expect noise from month to month. A restaurant that moves from being named in three of fifteen questions to seven of fifteen over a season has learned something real; a restaurant that celebrates one good screenshot has not.

Read the competitor column too. If the same two venues are named whenever you are not, open their websites and profiles and compare them with yours. The difference is usually visible: a page about the occasion, a text menu, a fuller description, more recent reviews that mention the thing the guest asked about.

• Do not pay for "AI submission" or "AI indexing" services. There is no submission pipeline into these answers.
• Do not stuff pages with lists of questions written for machines. Thin pages weaken the record an assistant reads; they do not strengthen it.
• Do not write or buy reviews. Apart from the platform rules it breaks, assistants quote review language back to guests, and invented praise reads as invented.
• Do not block a search crawler by accident while trying to block AI training. They are separate user agents with separate rules.
• Do not treat ads as a shortcut. OpenAI says ads in ChatGPT are separate, clearly labelled and do not influence its answers.

Source: OpenAI Help Center — Ads in ChatGPT — OpenAI's statement that ads run on separate systems and do not shape, rank or alter answers.

Everything an AI check uncovers is ordinary local search work: an accurate Business Profile, consistent listings, a website with readable menus and real detail, and reviews from real guests. The check simply shows you, in the words guests use, which parts of that record are missing or wrong. Our free visibility audit now includes an AI check of this kind, and the Growth and Growth+ plans on our pricing page include monthly AI search tracking against named competitors.

Written by

Alex Opsenica

Founder, Moolah Media

Founder of Moolah Media, a founder-led restaurant marketing specialist based in Ireland, working with restaurants worldwide on websites, SEO and Google Maps, and Google Ads. Around 20 years in hospitality, including hands-on restaurant management.

More about how Moolah works.