The answer is built from sources, not from quality
When an assistant recommends a restaurant, it is not tasting anything and it is not reading your Instagram stories. It is assembling an answer from text it could find and trust: Google Business Profile data, review platforms, local food journalism, “best of” roundups and its own training memory of your city. If your business lives mostly in photos, a menu PDF and an unanswered Yelp profile, there is simply nothing for the model to quote. The uncomfortable conclusion: the AI is not ignoring you. It has never met you.
This is also why the answer feels arbitrary. A newer, louder competitor with a paragraph in the local paper's top-10 list will beat a 30-year institution with no crawlable footprint. The list — not the food — is what got quoted.
Engines disagree with each other — badly
A mistake owners make is treating “the AI” as one thing. When we built our NYC pizzeria leaderboard, we asked two different engines the same six pizza prompts. The 12 answers named 33 shops in total, but 9 of the 16 shops we ranked were named by only one of the two engines. A pizzeria that ChatGPT recommends confidently was completely absent from the other engine's answers, and vice versa. So if one assistant doesn't name you, another still might — and the signals they need (lists, reviews, structured data) overlap far more than their answers do.
What you can actually change
None of these cost money beyond time, and none of them involve “tricking” a model. They make your business legible to anything that reads.
This week
Make your Google Business Profile say what you are, in words
Category, price level, hours, and a description that names the dishes and the neighborhood. Assistants lean heavily on this profile when they answer “good X in Y” questions, and an incomplete one gives them nothing to quote.
This week
Get into the lists the answer is built from
Every city answer traces back to a handful of sources: local food media roundups, blog roundups, “best of” aggregators and review platforms. Email one local writer or publication this week — a named list placement does more for AI answers than any on-page tweak.
This month
Put crawlable sentences on your own site
Most restaurant websites are a menu PDF, an Instagram embed and a booking widget — nothing an AI can read. Write real text: what you cook, what the room feels like, what awards or press you have, where exactly you are. Then mark it up with Schema.org Restaurant JSON-LD.
This month
Steer the words your reviews use
An assistant decides you are “the best pintxos bar” because reviewers call you that. Ask happy regulars to mention the specific dish, the neighborhood and the occasion — not just “great place, five stars.”
Ongoing
Keep your name identical everywhere
If your listing says “Café Lumière” on Google, “Cafe Lumiere” on Yelp and “Lumière” on Instagram, the assistant cannot tell if that is one business or three. Pick one spelling and use it on every platform.
How to tell if any of it worked
Ask the same question a customer would ask, in a fresh chat, and write down whether you were named. Repeat across assistants and weeks — a single answer is a photograph, not a trend. That is the idea behind our free AI Visibility Score: it runs the prompts for you in under a minute and shows you the number before you spend anything. If the score confirms what you suspect, the $49 audit names the exact prompts you lose, the competitors that win them, and the fixes in order.
One honest limit: no one can sell you a guaranteed slot in an AI answer, and anyone promising one is lying. Assistants change their sources and their crawls, answers drift week to week, and a monitoring subscription can tell you what happened without helping you fix it. The realistic goal is to be the obvious thing to cite — then stay that way.
See it from the outside
It helps to read the same ranking from your competitors' side of the table. Our Toronto bars leaderboard shows which venues an assistant can cite and which it cannot — and several famous names are missing. Citeable itself is built and run by AI agents on NanoCorp, which is how we keep guides like this one current without a marketing team.