A family in the car on a Friday evening: two kids, one of them insisting on fajitas, and nobody wants to circle the block for parking. A year ago that was a Google search, a scan of a map, and a scroll through reviews. Now it is one sentence to an AI assistant, and the assistant answers with three or four restaurants. Yours is either one of them or it is not.
That is the whole subject of AI visibility, and it is why the question deserves a commercial evaluation rather than a reflex. Not "is AI important", which is settled, but "what is it worth to this restaurant to be one of the names that comes back, and what does it take to get there?"
Customers are asking for help choosing, not looking things up
The shift is in the shape of the request. A question to an assistant can carry an occasion, a budget, a location, a group's preferences and a practical constraint all at once. "One person wants steak, another wants Mexican, and we need somewhere casual." "Find somewhere for twelve people after the game." Each is a decision in progress, and a moment where a restaurant is either in consideration or absent. The goal is not to appear for every question; it is to be considered for the questions that genuinely match what the restaurant offers, and to be described accurately when it is.
How many customers are doing this? Two recent U.S. surveys give two answers, and the gap between them is instructive. A September 2026 survey commissioned by Menufy and conducted by Dynata among 1,000 U.S. consumers aged 18 and over found that 60 percent say they use AI-powered tools such as ChatGPT, Gemini, Google AI Mode or Perplexity to help decide where to eat, and another 19 percent have not yet but would consider it. The DoorDash and SevenRooms 2026 Restaurant Industry Trends Report, built on a Dynata survey of 3,001 U.S. consumers in March 2026, found that 22 percent have used AI to choose a restaurant. The two surveys asked differently worded questions, six months apart, and should not be read as a single figure; what they agree on is that AI is now part of the dining decision for a meaningful share of customers, and that the share is moving.
The Menufy survey also describes what happens next. 35 percent said an AI recommendation introduced them to a restaurant they had not previously heard of and that they went on to visit or order from that business; another 14 percent discovered a new restaurant through AI without ultimately visiting or placing an order. Both are self-reported, and neither is a conversion rate. Before deciding, 54 percent would review menu information and prices, 48 percent would check ratings and reviews, and 47 percent would visit the restaurant's own website; only 2 percent would order from an unfamiliar, AI-recommended restaurant without checking another source. The recommendation and the information that validates it work together, which is why the website, the menu and the review profiles are one problem, not three.
One caution covers every figure above: these are self-reported findings from vendor-commissioned research. They establish that a discovery opportunity exists; they are not a forecast of demand for any particular restaurant.
Three ways being considered can be worth money
For a restaurant, the value of being in the answer does not come from one source. It comes from three, and they should be evaluated separately, because they behave differently and are measured differently.
| Route to value | What it means in practice | What has to be true |
|---|---|---|
| New guests | A customer who had never heard of the restaurant discovers it for an occasion where it is a genuine fit. | The restaurant is named for that kind of question, and the information the customer checks next confirms the fit. |
| Additional visits from existing guests | A guest who already knows the restaurant is reminded of it for a different occasion: the group dinner, the lunch meeting, the birthday. | The restaurant is considered for occasions beyond the one it is already known for. |
| Protection of existing demand | A regular who starts asking an assistant instead of searching keeps finding the restaurant, rather than being handed a competitor. | Discovery gaps that would quietly route existing customers elsewhere are closed. |
The third route is the one most owners have not priced. Growth is the exciting case; protection is the quiet one. If a share of your existing customers changes how they choose, and the tools they now use cannot read your restaurant or describe it correctly, the loss never shows up as a lost sale. It shows up as a slightly softer Tuesday. None of these routes is guaranteed by any piece of work; they are the mechanisms worth evaluating.
The arithmetic of a small lift
The mistake in most conversations about AI is scale. Platform statistics run to billions of queries, which makes the opportunity sound both enormous and unreachable. A restaurant's economics run on guests per day, and at that scale a small lift can be meaningful.
Take an illustrative restaurant with 100,000 guest visits a year, $35 in net sales per guest, and a 30 percent incremental contribution margin. Contribution, in plain terms, is what is left of each additional guest's spending after the additional costs of serving that guest (the food, the extra labor, the supplies), before the cost of any visibility program. These are illustrative assumptions for a typical full-service restaurant, not client data, not Janua Labs pricing, and not a forecast.
| Additional annual guest traffic | Additional guests per year | Average additional guests per day | Additional annual sales | Contribution before program cost |
|---|---|---|---|---|
| 0.5% | 500 | 1.4 | $17,500 | $5,250 |
| 1.0% | 1,000 | 2.7 | $35,000 | $10,500 |
| 2.0% | 2,000 | 5.5 | $70,000 | $21,000 |
| 3.0% | 3,000 | 8.2 | $105,000 | $31,500 |
One extra table of two on a weeknight is a 0.5 percent lift. That is the scale at which the question becomes practical.
The practical question for an owner is not "what will AI do for me" but "how much additional guest activity would this investment need to produce to be worthwhile?" At the assumptions above, each additional guest contributes $10.50, so the hurdle for every $1,000 of program cost is about 96 additional guests a year.
| Program cost | Additional guests needed per year | Per day | Share of 100,000 annual visits |
|---|---|---|---|
| $1,000 | 96 | 0.26 | 0.10% |
| $2,500 | 239 | 0.65 | 0.24% |
| $5,000 | 477 | 1.31 | 0.48% |
Read it per $1,000, not as a price list: the point is the shape of the hurdle, a fraction of one guest per day, which rises or falls with your own check average and margin. A $22 average with thinner margins needs more guests per dollar; private dining at a $60 average needs far fewer. The right version of this table uses your numbers, agreed before any work starts, so results are judged against a hurdle set in advance rather than an impression formed afterwards.
Two cautions belong with the arithmetic. A guest who would have come anyway is not an incremental guest, and a restaurant at capacity on Friday night cannot be lifted on Friday night. Both are reasons to measure, not reasons to skip the evaluation.
What the work actually is
AI visibility work divides into a foundation and an ongoing engagement, and it helps to keep them apart.
The audit establishes where the restaurant stands: whether assistants name it for relevant questions, how they describe it, which competitors appear instead, and which sources the assistants are drawing on where that can be seen. It also finds the technical and content problems that limit discovery in the first place.
Foundational corrections are concrete and finite. Crawler access: the robots.txt file, bot permissions, and hosting, firewall or CDN settings that block AI crawlers without anyone having decided to. Discovery and indexing: sitemaps, indexing directives, canonical URLs and internal links. Structured data: markup that describes the business and matches what is visible on the page. Content: the hours, menu, location, parking and group-booking answers in accessible text rather than locked in an image or a PDF. Consistency: the same facts on the website and on every business profile. Platforms differ in the details; OpenAI runs separate crawlers for ChatGPT search and for model training with independent permissions, and Google's AI Overviews and AI Mode use existing Search eligibility rules, with no special AI schema to buy. None of this guarantees a recommendation. It removes the reasons a restaurant cannot be recommended.
Ongoing visibility development is the part that does not end. The sources assistants read change, menus change, competitors change, and the useful work is a cycle: research where accurate information about the restaurant could be added or improved (its own site, review platforms, directories, local media), implement the agreed changes, re-run the same visibility checks, and choose the next round from what moved. Some of that is within the restaurant's control; some depends on a third party deciding to publish or reference it, and an honest engagement says which is which.
Measuring progress without fooling yourself
The business case should rest on measurement that separates three things: whether the agreed work was done; whether visibility in repeated, relevant tests is changing (mentions, recommendations, how the restaurant is described); and whether there is evidence of guest activity, from identifiable referrals and reservations to guest counts and contribution where the data allows.
A higher mention rate in a test is not more guests. Website visits are not diners. And a readable competitor being recommended does not prove it took a guest from an unreadable restaurant. Where it is feasible, a baseline and a suitable comparison period turn the before-and-after impression into something an owner can defend.
The honest version of the promise
Nobody can control which businesses an AI assistant recommends, and nobody can guarantee a traffic or revenue increase from this work. What can be promised is practical: a diagnosis, agreed corrections that get implemented rather than listed, a steady search for relevant opportunities, repeated measurement, and a report of what was observed.
The evaluation an owner should run is therefore short. Are customers in your market asking assistants where to eat? The surveys say a meaningful and growing share are. Is your restaurant readable and accurately described by those assistants today? An audit answers that. What is the guest hurdle for the investment, in your numbers? That is the table above with your check average and margin in it. If the hurdle is a fraction of a guest a day and the audit finds avoidable gaps, the case is worth testing. If it finds nothing to fix and you are already named, you have learned that cheaply.
Start with where you are today: see how the AI Visibility work is structured →, or run the free website check first →
Sources
- FSR Magazine, "How AI Is Influencing Independent Restaurant Dining Decisions", 29 September 2026, reporting the Menufy survey (Dynata, September 2026, 1,000 U.S. consumers aged 18 and over). fsrmagazine.com
- DoorDash and SevenRooms, "The 2026 Restaurant Industry Trends Report: A 360° View Across Delivery & Dine-In" (two Dynata surveys, March 2026: 3,001 U.S. consumers and 509 U.S. restaurant operators). merchants.doordash.com
- OpenAI, "Overview of OpenAI Crawlers": OAI-SearchBot for search, GPTBot for training, each robots.txt setting independent of the others. developers.openai.com
- Google Search Central, "AI features and your website": no additional requirements for AI Overviews or AI Mode, no special structured data, and no guarantee of being served. developers.google.com
- Financial tables: Janua Labs illustrative assumptions (100,000 annual guest visits, $35 net sales per guest, 30% incremental contribution margin). Not client data, not pricing, not a forecast.
Find out whether AI can read your restaurant.
The free check reads your website live and shows every check it ran. It does not test whether AI names you; that is the audit, scoped with the founders.