How AI Recommends Restaurants and How to Influence It

A guest walks into a suburban American restaurant and asks about a Wednesday special that Google's AI overview recommended. The problem: that special doesn't exist. The AI made it up. The owner eventually posted on social media asking customers to stop using Google's AI to check specials, because the information kept coming back fabricated.

That story captures the tension at the heart of AI restaurant recommendations: the systems are powerful, growing fast, and sometimes confidently wrong. Ignoring them is a risk. Trusting them blindly is also a risk. The useful response is somewhere in between.

The Scale of the Shift

Three AI assistants now reach a combined audience that dwarfs any single review platform. ChatGPT has over 900 million weekly active users and powers integrations across Microsoft Copilot, Apple Intelligence, and Samsung Galaxy AI. Gemini has surpassed 750 million monthly active users through Google Search, Google Maps, and Android. Meta AI has crossed one billion monthly active users across WhatsApp, Instagram, Messenger, and Facebook.

These numbers matter because consumer behaviour is following. As of 2026, 45% of consumers use AI tools to find local businesses -- up from just 6% a year earlier. Tripling in twelve months is not the kind of trajectory you plan around with "maybe next year."

The fundamentals still dominate discovery, for now. Toast's latest restaurant trends report found word of mouth drives 38% of how guests find new restaurants, in-person walk-by accounts for 30%, and Facebook 27%. Traditional search sits at 16%.

Read the two numbers together. AI is still a minority discovery channel. It's also the one growing fastest by a wide margin, and the gap is closing every quarter. The question isn't whether AI matters yet. It's whether your restaurant will be visible when it does -- and whether what the AI says about you is accurate.

How AI Picks Its Favourites

AI assistants don't rank restaurants the way Google does. There's no paid placement, no bidding on keywords, no local pack algorithm. Instead, the model synthesises everything publicly available about a restaurant -- reviews, website content, social media, press coverage, menu descriptions, food blogs -- and forms an opinion.

When someone asks "where should I take my parents for their anniversary?", the AI doesn't just retrieve a list. It matches the request against what it believes about each restaurant: the atmosphere, price range, cuisine, service style, and what kinds of occasions it suits.

This means two restaurants with identical star ratings can get very different treatment. The one with more detailed, specific, and recent public information gives the AI more to work with. The one with sparse or outdated information gets passed over.

Volume of Information Matters More Than You'd Expect

A study by MyPlace analysing AI recommendation patterns found that restaurants recommended by ChatGPT average 3,424 Google reviews, compared to 955 for those not recommended -- a 3.6x gap. Star ratings above 4.4 barely influenced whether a restaurant appeared. What mattered was the sheer volume of information available about the business.

That finding aligns with what an analysis of 350,000+ business locations revealed: ChatGPT recommends only 1.2% of all local business locations. Gemini does better at 11%, and Perplexity sits at 7.4% -- all three still a fraction of what appears in Google's local 3-pack.

For independent restaurants, the gap widens further. Chain restaurants have massive web footprints -- thousands of mentions, standardised listings across every platform, media coverage. An independent restaurant with 80 Google reviews and a basic website has a dramatically smaller digital presence for the AI to draw from.

The 83% Invisibility Problem

Local Falcon analysed nearly 190,000 AI search results and found that 83% of restaurants don't appear in ChatGPT's recommendations at all. The same restaurants, searched on traditional Google, had only a 14% invisibility rate.

That gap is the story. AI search is one discovery channel among several, and being invisible on it doesn't mean 83% of restaurants are failing. It means the channel rewards a specific type of online presence -- one built on volume, consistency, and specificity of public information. The quality of food has almost no bearing on AI visibility. The quality and quantity of public digital information does.

The Accuracy Problem

Here's where it gets uncomfortable: AI assistants are not always right about the restaurants they do recommend. Testing across ChatGPT and Perplexity found that AI-generated business information is only about 68% accurate.

MIT research found that AI models are 34% more likely to use confident language when generating incorrect information. The model doesn't signal uncertainty -- it states fabricated details with the same tone it uses for verified facts.

For restaurants, this means the AI might confidently state wrong opening hours, invent menu items, or describe an atmosphere that doesn't match reality. And only 12% of ChatGPT users visit external websites to verify what the AI told them, compared to 73% of Google users.

That combination -- confident inaccuracy plus low verification rates -- creates real operational consequences. Guests show up expecting things that don't exist.

The Financial Stakes

Let's work through concrete numbers. Harvard research by Michael Luca found that a one-star increase on Yelp leads to a 5-9% revenue jump for independent restaurants (this effect applies to independents, not chains).

Take a small independent restaurant doing $400,000 in annual revenue. A one-star improvement could mean an additional $20,000 to $36,000 per year. Now consider the inverse: an AI assistant that confidently tells potential guests your food is "average" or that you close at 9 PM when you're actually open until 11 PM. Each inaccurate AI interaction is a potential guest lost -- not because of your quality, but because of bad data.

74% of consumers specifically want reviews from the last three months when evaluating a business.

If your most recent visible information is stale, AI models treat your restaurant as less relevant. That recency gap compounds over time: fewer recommendations lead to fewer guests, fewer guests lead to fewer reviews, and fewer reviews lead to even fewer AI recommendations.

What Restaurant Owners Can Actually Do

The fundamentals come first. Nothing in this section is specific to AI -- it's about building a strong, accurate online presence that serves every discovery channel. If you neglect your Google Business Profile, fix that before worrying about ChatGPT.

Make Your Information Complete and Consistent

AI models pull from every public source they can find. When your website says "modern Nordic cuisine," your Google profile says "Scandinavian," and TripAdvisor says "European," the AI gets a muddled picture. Consistency across platforms gives the AI (and human visitors) a clear signal.

Fill in every field on your Google Business Profile. Upload your current menu. Keep hours accurate, including holiday hours. Mark every relevant attribute: outdoor seating, wheelchair access, private dining, dietary options.

77% of diners check a restaurant's website before visiting, and nearly 70% have been discouraged from visiting by a poor website experience.

Your website is the source of truth. Make it specific: your cuisine, your concept, your atmosphere, what occasions you're suited for, your actual menu with real prices. Generic descriptions produce generic AI understanding. Specific descriptions produce accurate AI understanding.

Earn Detailed, Recent Reviews

97% of consumers read online reviews before choosing a business.

Reviews that mention specific dishes, occasions, and experiences teach AI models more than five-star ratings with no text. "Perfect anniversary dinner, the tasting menu was creative and the wine pairing was excellent" gives an AI model material to work with. "Great food" does not.

You can't write your guests' reviews. But you can shape the conversation. Train staff to ask specific follow-up questions: "How was the risotto?" rather than "How was everything?" That specificity often carries over into the review.

A steady stream of recent reviews matters more than a burst of old ones. Encourage reviews consistently rather than in campaigns.

Monitor What AI Actually Says About You

Ask ChatGPT, Gemini, and Meta AI about your restaurant. What they say might surprise you. If an AI model has your cuisine wrong, your hours wrong, or your vibe wrong, that's the starting point for correction. We cover the specific fix process in a separate article.

Respond to reviews that contain incorrect information. When a guest writes "great Italian food" about your Nordic restaurant, your response is a chance to correct the record: "Thank you -- we're glad you enjoyed our modern Nordic menu." AI models read both reviews and responses.

Don't Optimise for AI at the Expense of Everything Else

Here's a counter-argument worth taking seriously: a restaurant that obsesses over AI visibility while neglecting food quality, service, and the fundamentals of hospitality is solving the wrong problem. The discovery channels that have always mattered -- word of mouth, walk-by traffic, a loyal guest base -- still drive the majority of new business.

From building Nine Tables, we've seen this pattern repeatedly. The restaurants with the strongest AI presence are the same ones that already had their fundamentals in order: a complete Google profile, an accurate website, a steady flow of genuine reviews, and consistent quality that generated organic word of mouth. AI visibility was a byproduct of getting the basics right, not a separate project.

The right framing: treat AI as one more channel that rewards accuracy, completeness, and recency of your public information. The same work that improves your SEO and your presence on review platforms also improves how AI models perceive you.

How Nine Tables Fits In

Nine Tables monitors how AI models perceive your restaurant. Each month, the system queries ChatGPT, Gemini, and Meta AI about your restaurant and tracks what each model believes -- your recommendation scores, which guest types you're matched with, perceived strengths and weaknesses, and which dishes they associate with you.

This matters because you can't fix what you can't see. If ChatGPT thinks you close at 9 PM, or Gemini doesn't associate you with outdoor dining despite your terrace, those gaps are costing you guests you never know about. The monitoring surfaces these discrepancies so you can address the root cause: the inconsistent or incomplete public information feeding the models.

Where This Is Heading

52% of diners say they're open to AI recommendations based on their past orders and preferences.

That number suggests AI-driven restaurant discovery will keep growing, particularly as these assistants get better at personalisation. The models will improve in accuracy over time. But for now, their understanding of any individual restaurant is only as good as the public information available.

The restaurants that will benefit most from this shift are not the ones gaming an algorithm. They're the ones making sure the algorithm has something accurate to work with. A complete profile, a clear website, fresh reviews, and consistent information across every platform.

That's not an AI strategy. It's just good practice for being found -- by humans and machines alike.