MIT Technology Review published a piece on August 18 about a hole in the AI discourse: nearly everything we know about how people actually use AI comes from the vendors themselves.
Anthropic and OpenAI both put out regular reports on how people use Claude and ChatGPT. Researchers quoted in the piece note the obvious limitation: the companies release the data they want released, and nobody outside can check it.
"There is no independent source to corroborate it" — Anka Reuel, computer science PhD candidate, Stanford Trustworthy AI Research group.
Those reports are load-bearing. They get cited in labor-impact arguments, safety claims and policy hearings, and they are first-party marketing documents from firms with a direct interest in what the numbers say. Until someone builds an independent measurement panel, the field is reasoning about AI adoption from telemetry it cannot audit.
Sources: MIT Technology Review