The MichelinGPT Moment

What just happened

OpenTable shipped a public concept called MichelinGPT — a guided experience that produces Michelin-shaped recommendations in ChatGPT's voice. The interface is clean. The language is fluent. The branding is borrowed from the most authoritative guide in the industry.

The reaction inside hospitality was immediate, and it split in two. Some operators saw it as the future of discovery. Others felt something was missing — the picks were technically correct and emotionally empty. Both groups are right. That is the point.

The thesis, validated in public

I have spent the last eighteen months saying one sentence in every room I walk into: I study how AI interprets restaurants.

MichelinGPT is what that sentence looks like when it goes mainstream. A trillion-dollar reservation platform now agrees, in public, with a logo on it, that AI is the layer where restaurant identity is being decided. The argument is over. The only question left is who teaches operators how to be legible to that layer — and who teaches the layer how to read them back accurately.

Why "the witness, not the expert" matters

The easy mistake — and the one MichelinGPT invites — is to treat ChatGPT as a critic. To ask it whether a restaurant is good. To rewrite a menu until the machine likes it. To let the model become the authority on what a restaurant is.

That is the wrong stance. ChatGPT has never eaten anywhere. It cannot judge a kitchen. What it can do is repeat back, in plain language, what the public surfaces of a restaurant currently say. That is not expertise. That is witness testimony.

The Signal Integrity Diagnostic uses the machine the way a courtroom uses a witness. We do not ask the witness for an opinion. We ask the witness what they saw. Then we compare that account to the truth of the room — and we correct the surfaces until the two match.

The four shapes the gap can take

Every Reading we run resolves to one of four diagnostic archetypes. They describe the shape of the gap between the restaurant that exists and the restaurant the machine currently describes.

• Ghost Gap. The witness does not see you at all.
• Distortion Gap. The witness sees a restaurant that is not yours.
• Commodity Gap. The witness sees you correctly, but generically — one of ten.
• Alignment. The witness describes you the way you would describe yourself, for the right guest, for the right occasion.

MichelinGPT, at its best, can deliver Alignment-shaped language to a restaurant that has already done the work. At its worst, it papers over a Distortion Gap with confident prose — and operators mistake the fluency for accuracy.

What this changes for operators

Nothing, and everything. The work has not moved. The Operator Vocabulary, the Reading, the discipline of writing one specific sentence instead of five generic ones — all of it is still the work. What has changed is the audience for the work. It is no longer marketing. It is the witness layer that now sits between every guest and every reservation.

The restaurants that win the next decade will not be the ones that beg ChatGPT for a better review. They will be the ones whose surfaces are so clear that any witness — human, machine, critic, guide — repeats the same sentence back.