Signal
What people ask has become a record worth reading. Labs and newsrooms now study millions of questions put to AI assistants and find rhythms nobody reported: when work stops, when someone starts thinking about a business, what people tell a chatbot but never a search box.
So what?
What people ask says more than what they report. A log of questions is the most honest record of what a group cares about.
Evidence
NBER, September 2025 (study): "How People Use ChatGPT" finds non-work messages grew from 53% to more than 70% of usage, with practical guidance, seeking information and writing making up nearly 80% of conversations.
Library Journal infoDOCKET on the Washington Post, 12 November 2025 (news): the Post's analysis of 47,000 publicly shared ChatGPT conversations finds people disclose what they would never type into a search engine, including hundreds of email addresses and phone numbers.
Anthropic, 26 June 2026 (study): the Economic Index report reads anonymised Claude conversations with privacy-preserving classifiers and finds work queries fall at weekends, business-starting conversations peak on Saturdays and Sundays, and requests for news cluster at 7 a.m.
Worth asking
What would a month of your own questions reveal that a month of your reports never would?
