Twenty times the people, not one more seat

Short answer: ask for a dentist in Miami (442,000 residents), then in New York (8.8 million): the shelf has the same number of seats. Across the 13 cities of our US study, the median count of distinct names per trade runs from 19 to 22.5 - with no relationship to city size at all. New York, twenty times Miami, gets 20 seats where Miami gets 20.5. The shelf is not sized to the market: it is sized to the answer.

Shelf size is a property of the answer format, not of the market.

What was measured

The US study of September 13, 2026, re-read city by city: 13 cities, 16 to 18 trades each, 2 assistants (ChatGPT and Gemini) × 2 phrasings, for 872 answers and 218 city × trade pairs, each with exactly 4 answers. For each pair we count the distinct names kept by the cleaning rules of the previous editions; per city, we take the median across trades. Against it, municipal population from the 2020 federal census. The medians run from 19 (Houston) to 22.5 (Seattle, second-smallest city in the set), and the rank correlation between population and shelf is zero: Spearman’s rho = −0.09. The overall median is still 21 seats, yesterday’s figure.

The rationing comes from the format, not from knowledge

You might think the assistant knows too few New York businesses to make the list longer. The opposite is true: it knows far more than in Miami, and it seats no more of them. Read straight from the text of the 872 answers, bypassing our extractor: the median answer contains 10 list items, in every one of the 13 cities, and the 90th percentile runs from 14 to 18 items with no city-size gradient. An assistant’s answer is a format: a short list, roughly the same length everywhere. That format is what sets the number of seats - not local supply, not demand, not the city.

What this changes if you are the business

The arithmetic is brutal, and it is local. In Miami or Atlanta, one median seat “stands for” 21,500 to 23,700 residents; in New York, 440,000. With equal shelves, a seat is contested twenty times harder in New York than in Miami. Yesterday’s map said the same thing from the other end: 171 mapped New York salons with a website for 19 seats, 27 Houston garages for 22. In a very large city, generic fame wins; in a mid-size city, a well-documented local business can genuinely aim for a seat. Knowing which side you are on - and what the assistants already say about you - is exactly what the audit measures.

What this measurement does not prove

• Population is city proper, not the metro area. On metro areas the New York/Miami gap falls to around ×3 - the shelf stays equal and the conclusion holds, but the headline ×19.9 is a city-proper ratio, the geography the question actually names.
• Our extractor caps at 8 names per answer (an audit-product choice): distinct-name counts are FLOORS. Flatness was therefore re-verified without it, in the raw answer text: a median of 10 list items per answer, everywhere. The cap also applies to previous editions’ numerators: floors, never ceilings.
• Residents are not businesses: “residents per seat” is a shorthand. Yesterday’s OpenStreetMap floor points the same way (171 salons for 19 seats in New York), without being a census either.
• US only, 13 cities, one pass, September 13, 2026, 4 answers per pair. Stability over time is not measured here.
• An equal-sized shelf is not an equal shelf: yesterday’s edition shows the seats go first to the names that travel from city to city.
• Nothing here says why a list is about 10 items long: formatting habits, training data and fine-tuning mix, and this measurement does not separate them.