The Evidence Layer
Published Observations
I study how AI interprets restaurants. These are the readings I have published — real businesses, read as the machines read them, with the gap between what they are and what they are understood to be named plainly.
- Case Read · August 2026Ghost Gap
A Caribbean bar the machines never learned to describe.
A Caribbean bar — New York
Published anonymously. Recognizable to people who already knew it, nearly invisible to systems meeting it for the first time — with a same-name distortion pulling signals in from another city.
Read the observation - Infrastructure Shift · August 2026Industry Shift
ChatGPT can now book your table. That changes what “getting recommended” is worth.
Reservation infrastructure — United States
Reservations now happen inside the conversation — no website, no menu, no second search. Booking was always the easy part. The question that just got more expensive: what makes AI choose your restaurant before it ever checks whether a table is open.
Read the observation - Field Note · August 2026Ghost Gap
AI doesn't recommend the best restaurant. It recommends the most validated one.
Validation volume — Category-wide
An award-recognized independent with roughly 800 reviews loses to a chain location with more than 5,000 — not on quality, but on volume of machine-readable validation. What actually closes the gap.
Read the observation - Case Read · July 2026Distortion Gap
A Harlem Italian restaurant, read on a Tuesday night.
Anonymous — Harlem, New York
A candlelit Sicilian dining room reads to machines as a casual neighborhood red-sauce spot. Published anonymously, with the corrections that close the gap.
Read the observation - Field Note · July 2026Commodity Gap
Menu descriptions are now the discovery layer.
Menu language — Category-wide
“Chicken sandwich” tells a machine nothing. Menu language has quietly become the primary surface AI reads to decide what kind of restaurant you are.
Read the observation - Case Read · May 2026Distortion Gap
A Dominican bakery a loyal customer had to call to order from.
Anonymous — Upper West Side, New York
A returning customer who'd eaten in the dining room a couple of days before opened a delivery app to order the same dish, and found only an English breakfast-and-sandwich menu — no trace of what she'd just had in person. She had to call the restaurant directly; staff recommended different food items at a similar price, then corrected the order by hand. Four competitor restaurants surfaced before them in her search, but she still ordered because she wanted the food from there.
Read the observation - Collection · June 11, 2026Editorial Collection
The Interpretation Gap, Volume I.
Collected observations — Location not specified
An editorial collection of early restaurant interpretation observations: restaurant identity, misunderstood signals, and what AI can support from the sources it reads. Not a verified international research study.
Read the observation
Commission a reading
Every observation started as a reading.
The Guest Read is the first act of the work. It shows you how AI currently interprets your restaurant — before anyone edits a word.