Research · July 2026

How people find restaurants in the age of AI.

Finding a restaurant is no longer one search. People hear about a place through social media, an AI answer, or a delivery app, then verify it through Maps, reviews, and the menu. About one in five U.S. diners now uses AI to choose where to eat, and AI returns a handful of names instead of a full list. A restaurant that is not in that answer is invisible.

Discovery became a chain of small decisions

People used to run one search and pick. Now discovery runs in stages. First the idea arrives through a Google or Maps search, a friend, a TikTok, an AI recommendation, or a delivery app. Then people build a shortlist, and this is where AI, social video, and delivery apps carry the most weight. Then they validate the choice against reviews, photos, prices, and the menu. Only then do they get directions, reserve, or order.

Online discovery is close to universal: 94% of U.S. diners use online resources to find new restaurants (SevenRooms). Most orders and visits still go to familiar places, so this whole chain matters most for winning the first visit, and that is the visit an unreadable menu loses.

One in five diners already choose with AI

Two independent surveys land on nearly the same number. 22% of U.S. consumers have used AI to help choose a restaurant (DoorDash 2026, n=3,001), and 20% used AI to research restaurants or bars in 2025 (Reputation / Nielsen). The average hides a generational split: 61% of 25 to 34 year olds use AI for food and drink recommendations, against 4% of those over 65.

That number is a floor, not a ceiling. Most AI restaurant discovery hides inside tools people already call "Google." AI now sits inside Google AI Overviews, Yelp's assistant, Maps summaries, and delivery-app search. About 65% of searches now end without a click, rising to roughly 83% when a Google AI answer appears. The answer becomes the destination.

AI selects a few names, so absence is invisible

A conventional engine ranks. It shows dozens of restaurants, a map, ads, and pages of links, and a weak listing still appears on page two. An AI answer returns three, five, or ten names, and everything else drops out of view. Being left out hurts more than ranking low, because the diner never learns the restaurant exists.

Two studies with different methods both found large-scale invisibility. A QSR benchmark found 83% of restaurant locations absent from AI recommendations even though most had a Google presence (Uberall). An academic audit of three model families across 304 neighborhoods found that, restricted to verified restaurants, 47.5% were never recommended (arXiv). The pattern in the audit: restaurants with more reviews and more evidence of real activity showed up more, and thinly documented restaurants got ignored, confused with another business, or invented. An image or PDF menu gives the AI nothing to read, so it cannot confidently recommend the place.

Voice is a second front door

About half of diners use voice search to find restaurant information, and food service is the largest local voice-search category. Siri, Alexa, and Google Assistant do not read ten links aloud. They speak one answer, drawn from listing data and structured menu content. Voice sharpens the same rule as AI text search: one result, no scrolling, and it rewards the same asset, a clean machine-readable menu.

What decides whether a restaurant appears

The audits and practitioner guidance point to the same short list, in rough priority:

  • Consistent identity everywhere. Name, address, phone, hours, and cuisine should match across the website, Google Business Profile, Apple Maps, Yelp, and delivery apps. AI reads agreement across sources as trust.
  • A readable HTML menu, not a PDF. Real, selectable text with category, item, description, price, and dietary tags. Image-only and PDF-only menus stay invisible to search, AI, and screen readers alike.
  • Schema markup. Restaurant and FAQ schema translate the page into machine-readable structure that AI answers pull from.
  • Recent, specific reviews that tell the AI what the place is known for.
  • Intent-matching detail: good for families or groups, quiet or lively, parking, price range, and dietary and accessibility notes.

The same properties that make a menu answerable by AI make it readable by a blind or low-vision diner. Accessibility and discoverability are the same engineering problem.

What an AI-visible menu is worth

No credible study isolates a guaranteed "+X% sales" from being recommended by AI, and attribution is hard: a diner can find a restaurant in an AI answer, search it in Maps, read the menu, and walk in, leaving a trail that looks like a direct visit. So the model below is illustrative. It does not promise a number. It shows the share of new-customer discovery now flowing through AI and voice answers, revenue that goes to whichever restaurant the answer names.

Illustrative model: 100 new-customer parties per month from online discovery, $55 average party spend. Not published averages.
AI / voice share of discoveryParties via AI answersRevenue/moRevenue/yr
10%10$550$6,600
20% (today's adoption)20$1,100$13,200
30% (near-term)30$1,650$19,800

Read each row as revenue at stake, not revenue gained. A structured, accessible menu makes the restaurant eligible to be the name the AI speaks. An image or PDF menu removes it from the answer, and a competitor with a readable menu takes the party. Separately, if better menu information lifts conversion among visitors who reach the page, even a low-single-digit lift compounds on top of this.

Honest limits of the evidence

These figures come from different surveys with different samples and years, and they do not add together. Vendor benchmarks and aggregated search statistics show direction and rough scale, not precise values. The defensible claim is not a revenue guarantee. It is that an accessible, structured menu makes a restaurant discoverable and recommendable at the moment discovery is consolidating into a few spoken or summarized answers.

Traditional search ranks every restaurant. AI names a few. An accessible, machine-readable menu is what puts a restaurant in that short answer instead of leaving it out of the conversation entirely.

Sources

Independent statistics are cited to their publishing sources. Figures come from different 2024 to 2026 surveys and do not add together. The revenue model is illustrative, not a published average or a guarantee.