AI GUIDES: GUIDE 02
How do AIs choose
which restaurants to recommend?
An AI does not answer by reading a fixed ranking table. It reads the conditions in the question, gathers public information from the web (your official site, your Google listing, dining sites, reviews) and assembles the answer into sentences on the spot. In other words, what the AI sees is not your restaurant itself, but the information published about it.
Last updated: October 30, 2026 | AI Guides
STEP 1
Step 1: It reads the conditions in the question
A guest's question contains the conditions for choosing: area, cuisine, occasion, budget, party size, as in “Italian in Asakusa, good for an anniversary, around ¥10,000.” The AI first reads these conditions and organises what a restaurant has to satisfy.
What matters here is that how the question is asked changes which restaurants appear in the answer. “Ramen nearby” and “ramen I can take the kids to” ask for different things. This is why our measurement uses several ways of asking. Our measurement principles are collected on the Methodology page.
STEP 2
Step 2: It gathers public information as material
Next, the AI gathers information on restaurants that fit the conditions. The workings differ by AI and by settings, but the public information that can serve as material is mainly:
- Official site: the restaurant's own descriptions, menus, and access information
- Google listing (Google Business Profile): opening hours, address, cuisine category, photos, reviews
- Dining and booking sites: listing descriptions, course information, reviews
- Reviews: guests' ratings and written impressions
- Articles and roundup pages: third-party content introducing restaurants by area or cuisine
Search-linked AIs consult web pages on the spot. Others answer from learned knowledge. Either way, the shared point is that the material is information that is published and readable as text.
STEP 3
Step 3: It narrows down to restaurants that fit
From the gathered information, the restaurants that remain are those that fit the question's conditions and for which the AI can write a reason to recommend them. Here, what counts is not only the “amount” of information, but its clarity and consistency:
- Can it be read from the information that the restaurant fits the conditions (area, cuisine, occasion, and so on)?
- Is the restaurant's character described in text, so a reason to recommend it can be written?
- Do multiple sources agree, so the restaurant can be introduced with confidence?
Conversely, restaurants with little information whose character cannot be read, or whose details disagree between sites, tend to drop out at this stage. The specific reasons for falling out are sorted out in “Why doesn't ChatGPT recommend my restaurant?”
STEP 4
Step 4: It introduces them in sentences, with reasons
Finally, the AI writes up the remaining restaurants together with the reasons they are recommended. Search-linked AIs may show citations to the pages they drew on.
This “writing the reasons” character is the big difference from traditional search. What is asked is not your position in a list, but whether there is enough material to recommend you in sentences. That is why whether text describing your restaurant's character is published has become more important than before.
VISIBLE / INVISIBLE
Information an AI can and cannot “see”
When thinking about what a restaurant can do, holding this line clearly avoids misjudging the situation.
- Visible: published text information (official site descriptions, listing details, menu text, review text, articles about you)
- Hard to see: information shown only inside photographs, menus turned into images, announcements that exist only in social posts
- Invisible: the quality of the food and service itself, the actual atmosphere of the room, today's seat availability (unless linked to an external system)
Unfortunately, the quality of the actual food and service is not directly visible to an AI. What the AI sees is what someone has written about them and published. This is why reviews work as material; details are in “Do reviews affect AI recommendations?”
WHY IT VARIES
Why the answer differs every time you ask
By the way they are generated, AI answers are never exactly the same twice, even for the same question. Differences in phrasing, updates to the information drawn on, and the fluctuation of generation itself change which restaurants appear, and in what order.
This is a precondition, not a defect. Judging from a single ask (“it didn't appear” or “it appeared”) is hasty; you need to look at the tendency across repeated asks. That variation is exactly why our measurement reports a share: how many times out of how many you appeared.
FAQ
Frequently asked questions
Q1How is this different from traditional search (SEO)?
Traditional search was about position: where you appear in a list of links. In an AI answer, the question is whether you are on the short list of restaurants the AI names and recommends in its sentences. At position 10 in a list, you are still seen; if your name is not spoken in the AI's text, you are not among that guest's options at all.
Q2Are the sources AIs draw on fixed?
They differ by AI and by settings. Your official site, your Google listing, dining sites, review sites, and map information can all be material, but the internal workings of how much weight each source gets are not published. Rather than betting on one particular site, the realistic response is to keep the accuracy and consistency of your public information as a whole in order.
Q3Why do some answers show citation links?
AIs that build answers in conjunction with search may show the web pages they drew on as citations. What gets cited is a page where the answer can be read as text. Whether your restaurant's character is published as text matters here, too.
Q4How much can the restaurant control?
There is no way to control the AI answer itself. What you can control is the public information the AI draws on: the accuracy of your official information, consistent notation, whether text describing your character exists, and correcting wrong information when you find it. Putting that in order is the extent of what can be done. Nobody can guarantee appearances themselves.
MORE GUIDES
More guides
Measure where your restaurant stands with an AI Visibility Check
These articles are general. The reason a particular restaurant does not appear differs from restaurant to restaurant.
With NIXUMA's AI Visibility Check, we ask several AIs repeatedly, using questions guests actually use, and report whether your restaurant was named, which restaurants were named instead, and whether anything said did not match the facts. The method is published on the Methodology page.
The service is currently in testing ahead of its formal release. For a release notice or a test-restaurant enquiry, email us, a single sentence is fine.
Requests and reports all happen in writing, by email and similar. We never call during your service hours. For the full overview, see the AI Visibility for Restaurants page.