Diners are asking AI where to eat. Your venue is either in the answer or it isn't.

Your next regular is asking an AI where to eat tonight.

Not Google. Not Instagram. They're typing "good patio for dinner in Mount Pleasant, somewhere not too loud" into ChatGPT or Gemini, and they're trusting the answer it gives back.

That's new. A year ago this was a party trick. In 2026 it's a habit.

Across the operator cohort I track, 11% of new guests in May named an AI assistant when asked how they found the venue. In January it was 3%. Same question, same field on the same intake form, same venues — the number nearly quadrupled in five months.

That's not a rounding error. That's a discovery channel forming in real time, and most operators have no idea whether they show up in it.

How the answer gets built

When someone asks an AI where to eat, the model doesn't read your beautiful website. It reads what the rest of the internet says about you — and then it summarizes confidently, whether the source material is good or not.

Three things feed that answer → your Google Business Profile, your recent reviews, and whatever third-party listings mention you by name.

Notice what's not on that list. Your homepage. Your Instagram grid. The thing you spent the most money on.

The AI reads the room around you, not the room you built. You don't get to write your own entry. You get summarized by a machine working off other people's notes.

What I'm seeing work

I pulled the venues in my cohort that get named most often — I ran the same prompts across ChatGPT, Gemini, and Perplexity — and the pattern is boring in the best way.

They have a Google Business Profile that's actually filled out. Hours, attributes, a primary category that matches what they are. Not "restaurant." "Natural wine bar" or "Pacific Northwest restaurant."

They have reviews from the last 90 days. The models weight recency hard. A venue with 400 reviews and nothing since March loses to a venue with 80 reviews and a dozen from last week.

And they get mentioned, by name, in at least one piece a human wrote without being paid to — a neighborhood blog, a "where to eat in [neighborhood]" list, a food writer's roundup.

That's the whole game right now. It is not sophisticated. It is just unattended.

What I'm seeing fail

The venues that never get named have one of three problems → a thin or stale Google profile, reviews that dried up a year ago, or a name so generic the model can't tell them apart from three other places on the same block.

One client — terrific food, packed every weekend — does not appear in a single AI answer for their own neighborhood. Their Google category is "restaurant." Their last owner-response to a review was 2024. They have a gorgeous, expensive website the model will never open.

They're invisible to a channel sending 11% of everyone else's new guests through the door.

Why this isn't just the SEO conversation again

People hear "show up in search" and reach for the 2015 playbook. Keywords, meta tags, a blog nobody reads.

That's not this.

Ten blue links let a diner pick. One AI paragraph picks for them. The model names two or three places and moves on, and if you're the fourth-best option you're not on the list — you don't exist for that query. Second place in an AI answer is still in the answer. Fourth place is invisible.

So the bar moved. It's no longer "be findable." It's "be one of the three the model is confident enough to name."

What to do this week

You don't need a strategy. You need ninety minutes.

Open your Google Business Profile and fix the category, the attributes, and the hours. Make the category specific — it's the single line the model trusts most.

Ask your last twenty happy tables for a review. A card on the bill, a line from the server, anything that keeps a steady trickle going instead of one big spike.

Then go ask the AI yourself. Type the three prompts a real diner would type for your neighborhood and your food. See if you're named. If you're not, the model just told you exactly what it's reading instead of you.

The operators who win this channel won't be the ones with the cleverest AI strategy. They'll be the ones who noticed it existed while it was still cheap to fix.

It's June. The number was 3% in January and 11% in May.

Don't wait for it to be 25%.

— Damon