OpenAI
OpenAIBack-to-school report

Learning with ChatGPT continues after the school day

At the school-year peak, U.S. users sent more than 460 million coursework-related ChatGPT messages a week; summer activity still exceeded 180 million. Separately, OpenAI identified up to 70 million weekly self-testing conversations across all ages. The populations and units differ, so the figures cannot be combined into a count of students.

An open notebook and laptop beneath a classroom window opening onto a night sky.Original AI-generated illustration: learning beyond classroom hours.

Self-testing conversations: answer-checking 63%, iterative learning 59%, learner attempt 58%, practice requests 51%, misconception checks 36%.OpenAI, Learning Never Stops, p. 5. Internal estimates; overlapping behaviors, not learning outcomes.

In self-testing conversations, answer-checking appeared in 63%, iterative learning in 59%, the learner’s own attempt in 58%, practice requests in 51% and misconception checks in 36%. Those categories overlap.

The report includes English teacher Casey Cuny’s use of AI feedback to help students develop arguments while retaining responsibility for drafting and final writing. His assessment comparison is observational, rather than a randomized estimate of the tool’s effect.

The usage shows people seeking help beyond classroom hours. It does not establish how much they retained or could apply later without assistance. Checking an answer and learning to solve the next problem remain different outcomes.