There's a behavioral pattern we've watched accelerate over the past 14 months. A traveler arrives in a city she's visiting for the first time. She has 20 minutes before dinner reservations need to be made. Five years ago she'd have opened Google Maps and scanned by star rating. Two years ago she'd have opened Yelp and read three reviews. Today she opens ChatGPT, types "best restaurants in [city] tonight," and reads the model's two-paragraph synthesis. She picks one. She books it. The decision is finished in three minutes.
This pattern shows up in our customer interviews almost universally now. The under-45 leisure traveler segment is leading; business travelers and over-45 travelers are following. The shift is most visible in dining (where the user wants a single confident pick), less visible in hotel booking (where price comparison still drives users to booking platforms), and least visible in flight booking (where AI hasn't won the comparison-shopping use case yet).
What real ChatGPT responses look like
To make this concrete, here's what we got back when we ran identical queries through ChatGPT (gpt-4o, web search enabled) on April 30, 2026:
Query: "Best restaurants in London tonight?" ChatGPT returned a five-paragraph response leading with Padella, Hawksmoor, and Dishoom — each with a one-sentence positioning ("Padella for fresh pasta in Borough Market," etc.) and a recommended use case ("for an impromptu Saturday lunch"). Total length: 280 words. No bullet list, no star ratings, no addresses unless the user asks.
Query: "Best restaurants in Paris for a special dinner?" Recommendations centered on Le Comptoir du Relais, Septime, and Frenchie — each with brief context about the chef's reputation and atmosphere. The model's defaults skew toward chef-led contemporary places over haute cuisine, mirroring how millennial-aged users tend to ask the question.
Query: "Where to eat ramen in Tokyo?" Tighter, more specific answer: Tsuta, Afuri, Ichiran (with a note that the last is a chain). Notable: ChatGPT explicitly named the neighborhood for each restaurant, and added the caveat that Tsuta moved locations in 2023.
The structural change in the funnel
From the restaurant's perspective, the visibility funnel used to look like this: customer searches Google → sees ten options → triages → clicks two or three → picks one. You needed to be in the top ten. You didn't need to be the most-recommended. Triage took the user away from default options.
The new funnel: customer asks ChatGPT → sees one to three recommendations → asks a follow-up if curious, otherwise picks the first one mentioned. The triage step is gone. The pick is fast. If you're not in ChatGPT's first three, the user often doesn't see you at all.
"Three is the new ten. AI's confidence in compressing the answer is exactly what makes the choice feel useful — and exactly what cuts your shelf space."
Why this is good news, even if your score is currently low
Two reasons. First, the AI recommendation set is much more responsive to specific business actions than the Google search ranking ever was. Six weeks of intentional signal-building (English Google profile, press, structured data, review-language coverage) can move you from outside the recommendation set to inside it. We've measured this on our own audit base. Search ranking, by comparison, takes months to shift.
Second, AI rewards specificity over scale. A small modern Anatolian restaurant in Istanbul with a clear "rediscovering forgotten regional ingredients" positioning will out-cite a much larger chain with no specific positioning, on cuisine queries. The decisive factor is whether the AI can attach a clear "this restaurant is for X" frame. If it can, you appear. If your positioning is generic, you don't.
Both of these mean a small business with intent and a budget of a few hours per week can meaningfully shift the AI's mental model of them. That wasn't true with SEO at this point in its lifecycle.
What we recommend
Three things, ranked by leverage:
- Find your existing AI mention pattern. Run "best [cuisine] in [your city]" through ChatGPT, Gemini, and Claude. See what each says. Note where the descriptions agree and disagree. The disagreements point to weak signal — those are the gaps you can move first.
- Tighten your single-sentence positioning across Google, your website, and any press. AI loves a clean frame. "Modern Anatolian rediscovering forgotten regional ingredients" beats "fine dining experience" by a wide margin.
- Get one English-language press mention this quarter. If you're outside the US/UK, this is your single highest-leverage AI signal. Pitch one outlet that covers your city's dining scene. Even a 200-word mention compounds.
Or skip all of that and run a structured audit. We measure where you stand, what AI says, and what to do — and we deliver in under 24 hours.