How People Use ChatGPT[pdf] (www.nber.org)

An NBER working paper (Sept 2025) analyzes how consumers used ChatGPT from launch (Nov 2022) through mid‑2025, using a privacy‑preserving automated pipeline to classify a random sample of user-to-bot messages from consumer plans (May 2024–June 2025). By July 2025 ChatGPT reached roughly 700 million users (≈10% of adults) sending ~18 billion messages per week. The authors map messages across taxonomies (work vs non‑work, conversation topic, interaction type, and O*NET task categories) by passing classification prompts to an LLM in a secure data clean room, enabling large‑scale labeling without humans seeing raw content and linking aggregated employment/education data. Key findings: non‑work use grew faster than work use (share rose from 53% to >70% of messages), and nearly 80% of conversations fall into three topics—Practical Guidance, Seeking Information, and Writing. Writing drives most work usage (≈40% of work messages), and two‑thirds of writing requests are edits or revisions rather than from‑scratch generation. Surprisingly, programming accounts for only ~4% of messages and companionship/therapy is <2%. Usage skewed early male then equalized; younger users (<26) send nearly half of messages; usage grew faster in lower‑income countries. Implication: chatbots are primarily decision‑support and content‑generation tools that extend beyond workplace productivity into broad consumer and educational uses, with important consequences for task redesign in knowledge‑intensive jobs and for measuring economic value beyond traditional productivity metrics.