Restaurant AI Search: Where the Hype Ends and the Money Starts

Nine out of ten conversations about AI search are fear sold by the slice. I work inside restaurant tech, and I keep seeing the same play. A vendor lands in a café owner’s inbox, shows a scary “your traffic is dying because of AI” chart, and offers a dashboard that tracks your “visibility in ChatGPT.” The chart is sometimes real. The dashboard almost never is.

So let’s split the topic in two. There is a part you cannot sell, because it cannot be measured. And there is a part that hits revenue for real, but looks boring, which is exactly why nobody packages it.

The part you can’t measure

“Find out your ChatGPT ranking” is a contradiction in terms. ChatGPT answers the same question differently for two different people. Often it doesn’t search the web at all. It assembles a list from training data, meaning the internet as it looked a year or two ago, then delivers it with the confidence of fresh research. You can’t measure a “position” inside a system that doesn’t reproduce its own output. You can only invent a number and sell it.

There is a deeper problem: a single “AI ranking” doesn’t exist. The engines don’t share sources. ChatGPT, when it does search, leans on Bing’s index, with roughly 87% of its citations tracing back to Bing’s top results. Perplexity runs a fresh search on every query and rewards recency hard, citing content under 30 days old at an 82% rate. Google’s AI Mode runs on its own index and, oddly, cites mostly different URLs than classic search (semantic overlap near 86%, URL overlap around 14%). Being “on top” in Perplexity tells you nothing about Google AI Mode. A dashboard that flattens all of this into one “AI visibility score” is adding up numbers that don’t belong in the same column.

That is the hype. Not because AI is irrelevant, but because an industry of tools grew up around it to measure fog.

The part you can measure, and it stings

Now the part I can’t wave away as a skeptic, because it lives in Search Console, not in a vendor deck.

Google’s AI Overview sits above the results and often answers the question before anyone scrolls. When it appears, the click-through rate on the first organic result drops hard. Ahrefs measured the position-1 loss at 34.5% in spring 2025, then 58% by December, across 300,000 keywords in Google Search Console. Seer Interactive reported a 61% year-over-year drop on AI Overview queries. Different samples, same direction, and the direction is getting worse.

Then there is how people behave. Pew Research tracked the real browsing of 900 U.S. adults in March 2025, published that July. With an AI summary present, people clicked a traditional link 8% of the time, against 15% without one. Clicks on the links inside the AI answer: 1%. And in 26% of cases the AI answer ended the session, against 16% on normal result pages. Google called the methodology flawed. Grant that, and every other dataset still points the same way.

Where restaurants actually get hit, and where they don’t

Here I’ll correct a claim I see repeated, including in my own earlier notes.

“Restaurants are the worst-hit vertical” is half true. Restaurants did see one of the fastest jumps in AI Overview coverage after the March core update, near the top of every category. But the damage concentrates on informational, long-tail queries. “Best ramen in the neighborhood,” “where to eat vegan brunch with outdoor seating.” That is the intent AI swallows whole.

Short local and navigational queries are far better protected. Pew found that one or two word searches trigger an AI summary about 8% of the time, while searches of ten words or more trigger one 53% of the time. “Pizza near me” and “[your restaurant] reservation” are short, local, action-driven. AI can summarize a cuisine. It can’t seat a table. Those clicks still land.

Map that onto your funnel. Top-of-funnel discovery (“where should we eat”) is being absorbed. Bottom-of-funnel intent, your name and your location and a booking, still clicks through. That one distinction should reorder your priorities.

The paradox that changes the strategy

Branded queries flip the whole picture, and almost nobody talks about it.

AI Overviews appear on only 4.79% of branded queries, and when they do, CTR rises by 18.68% (Amsive, 700,000 keywords). The system punishes “category” and rewards “name.” Searched as “Italian near me,” you lose the click to the machine. Searched by your own name, the machine politely points at you.

The takeaway is worth more than any dashboard. Your best defense against AI search is the strength of your own brand, not a fight for category rankings you are going to lose to a summary anyway.

One more thing before you do anything

“Optimize for AI” is a meaningless instruction, because AI is not one thing. ChatGPT is, at its core, a Bing problem: if Bing doesn’t index you, ChatGPT won’t cite you. Perplexity is a freshness problem: stale pages lose to newer ones on the same topic. Google AI Mode is its own index with its own taste in URLs. There is no single lever.

There is a common denominator, though. It is old, unglamorous, and nobody sells it as magic. That is the next piece.

The winners won’t be the ones who “did AEO.” They’ll be the ones who were never sloppy to begin with.

Disclosure: I’m a product owner at a restaurant platform. Structured data and listings are part of what we build. That is exactly why it irritates me when basic hygiene gets sold as “AI magic.” I’m in this up to my neck, and I’m still telling you not to buy the ChatGPT visibility dashboard.