How AI Assistants Decide Which Restaurant to Recommend
Someone opens an AI assistant and types: “where should we eat tonight near Wicker Park, nothing too expensive, my friend is vegetarian.”
Four restaurants come back. Yours isn’t one of them.
The instinct is to ask how to rank higher. That’s the wrong question, because there is no ranking. An AI assistant doesn’t sort a list of every restaurant in town. It retrieves a handful of sources, reads them, and writes an answer from what it found. If your restaurant wasn’t in what it read, you were never a candidate. Not ranked low. Absent.
How Big Is This Channel, Really?
Bigger than most operators assume, and growing fast.
Pew Research Center found that 34% of U.S. adults had used ChatGPT by early 2025, roughly double the 2023 share (Pew Research Center). That’s one assistant among several.
Then there’s the AI built into ordinary search. In an August 2025 survey, 65% of Americans said they at least sometimes see AI summaries in search results (Pew Research Center). So even guests who never open a chatbot are increasingly reading an AI-written answer before they see a list of links.
The Part That Should Worry You: Fewer Clicks
Pew also tracked what people actually do when an AI summary appears. Analyzing the real browsing of 900 U.S. adults in March 2025, it found that users clicked a traditional search result on 8% of visits when an AI summary appeared, versus 15% when none did, and clicked a link inside the summary itself only 1% of the time (Pew Research Center).
For a restaurant, that has an uncomfortable implication. More of the decision now happens inside the answer. If the summary names three places and yours isn’t one of them, the guest may never scroll down to find you, and nothing in your analytics will record the miss.
What These Systems Actually Read
The details vary by product and change often, but the logic is consistent: assistants answer from text they can retrieve and trust. For a restaurant, that usually means three places:
- Your listings, Google Business Profile above all: category, hours, attributes, and the menu link.
- Your own website and menu, if they’re readable text.
- Reviews and other mentions that describe your food and your room.
Google is explicit about its own AI features. Its guidance for site owners says there are no special requirements to appear in AI Overviews or AI Mode beyond normal search eligibility, and it tells site owners to keep important content available as text, make sure structured data matches what’s visible on the page, and keep Business Profile details current (Google Search Central). That’s not a trick. It’s housekeeping.
What the Academic Research Adds
The term “generative engine optimization” comes from a peer-reviewed paper by researchers at Princeton and IIT Delhi, presented at KDD 2024 (Aggarwal et al.). They tested what makes a source more visible inside AI-generated answers and found that adding concrete statistics, quotations, and citations raised visibility by up to about 40%, while old-school keyword stuffing performed poorly. They also found the effect varies by subject, so there’s no universal formula.
Translate that to a dining room and it points in one direction: specific, verifiable detail beats vague marketing language. “Wood-fired, 72-hour dough, vegan cheese available on any pie” gives a model something to match. “The best pizza in town!” gives it nothing.
Where Restaurants Fall Out of the Answer
Assistants answer specific questions. Not “recommend a restaurant” but “somewhere with good vegetarian options,” “gluten-free pasta near the stadium,” “open past 11 and not too loud.” To be retrieved for those, the words have to exist somewhere readable and connected to your business.
For most restaurants they exist on the menu. And for most restaurants the menu is a PDF or a photo, which means the richest, most specific information about your restaurant is the one part of your web presence a machine reads worst.
It’s also why short, concrete dish descriptions matter more than they used to. A dish listed as “The Gardener” with a price tells an assistant nothing. “Roasted cauliflower, tahini, pomegranate, pistachio — vegan” answers three different questions people actually ask.
That’s the case for MenuOnGo in this context. Your menu lives as real text on its own page, linked from your Google listing, with dietary labels on each dish on every plan. On Pro, AI suggests short, appetizing descriptions for the dishes that never had one, in the style you set, and you approve every word before it goes live, so the detail an assistant needs is finally written down. You can scan your current menu in for free and see what it looks like as text.
What Actually Moves the Needle
In rough priority order:
- Keep your Google listing accurate and complete, including the menu link.
- Make your menu readable text, with dish-level and dietary detail.
- Write concrete descriptions: ingredients, technique, dietary notes.
- Keep facts consistent across your listing, your site, and your menu. Contradictory hours undermine trust for people and machines alike.
- Earn and answer reviews. They’re one of the few places others describe your food in their own words.
Notice what isn’t on the list: stuffing pages with keywords or buying a mysterious “AI ranking” service.
A Reasonable Caveat
This field is young and moving monthly. Products change how they retrieve sources, and most published “AI visibility” studies come from companies selling visibility services. The independent evidence is thinner but points the same way as Google’s own guidance: publish accurate, specific, readable information about your restaurant in the places people and machines look. That work pays off in ordinary search and with human guests too, whatever the models do next.
What AI Assistants Need From a Restaurant
| What They Look For | Where It Lives | What You Control |
|---|---|---|
| Category, hours, location | Google Business Profile | Accuracy, completeness |
| Dish names, ingredients, dietary info | Your menu, if it’s text | Format and detail of every item |
| Descriptions of food and atmosphere | Reviews, articles, your site | Service quality; responses to reviews |
| Consistency across sources | Everywhere you’re listed | One set of facts, kept current |
| Specific, verifiable claims | Your own content | Concrete descriptions over slogans |
Frequently Asked Questions
How do AI assistants choose which restaurants to recommend? They don’t rank every restaurant. They retrieve a small set of sources relevant to the question, usually listings, websites, menus, and reviews, and write an answer from what they find. A restaurant that isn’t in those sources isn’t considered.
How many people use AI assistants or AI search? Pew Research Center found 34% of U.S. adults had used ChatGPT by early 2025, and 65% of Americans said they at least sometimes see AI summaries in search results.
Do AI summaries reduce clicks to restaurant websites? Pew’s analysis of real browsing found people clicked a traditional search result on 8% of visits when an AI summary appeared, versus 15% without one, which means more decisions are made inside the answer itself.
Is there anything special I need to do to appear in Google’s AI results? Google says no special requirements or markup are needed beyond normal search eligibility. Its guidance is to keep important content available as text, keep structured data consistent with the visible page, and keep Business Profile details current.
Why doesn’t my menu show up in AI recommendations? Usually because it isn’t readable. Menus stored as PDFs or images can’t be parsed reliably, so the dish names and dietary details that conversational questions depend on aren’t available to the model.
Sources: Pew Research Center (ChatGPT use, 2025; AI summaries and clicks, 2025; attitudes toward AI summaries, 2025), Google Search Central, Aggarwal et al., “GEO: Generative Engine Optimization” (KDD 2024). Images: Pexels (free to use). Data current as of October 2026.
Read it on your break. Your menu keeps working.
Thirty days of every Pro feature, menu import included, and nothing to pay.