Field notes on how AI reads names.
We probe how the frontier models read, recognize, and place names — at scale, with our own data. A desk publishes its research; this is ours.
- domain statisticsDomain counts you can cite
Every TLD’s size and every gTLD registrar’s share, month by month back to December 1983, with the month it covers and the name of whoever reported it attached to every figure.
read → - the 2026 roundThe 2026 new TLD round, explained
ICANN just closed the first application window for new top-level domains since 2012: $227,000 per application, the list sealed until Reveal Day in October. How the round works, what happens next, and when you'll actually be able to register names on these things.
read → - answer engine optimizationYou can’t AEO your way out of your name
Answer Engine Optimization assumes AI can already place you. But for about a third of names, the models can’t even agree what the company is for — and that’s set by the name, before you buy. You can optimize what AI says about a name; you can’t optimize your way out of the name.
read → - namingThe names you can think of are already taken
The good names left aren’t the ones you’ll brainstorm — they’re the ones you have to invent. Here’s how the desk finds them.
read → - the tech defaultWe invented 50 startup names. AI thought 38% of them were tech.
We invented 50 names and asked four models what industry each was in. The modal answer was “tech” — even for pure nonsense, even for words with a real non-tech root.
read → - the blank-canvas tailThe 8% of startup names AI hasn’t decided about yet
599 names probed. Most get a confident, unanimous read; about 8% split the models across industries — the highest-leverage band you can pick from.
read →