What we measured, written up in full.
Long notes on what we find while building Who AI Knows: what AI assistants name, what they read and what they skip, with the numbers and how we got them.
· 5 min read
A 700-token cap broke our AI answers and no check noticed
Cheaper engine settings cut Claude from 6.8 brands per answer to 3.1 and stopped seven of ten answers mid-sentence. Every check said nothing changed.