We analyzed 47,000 ChatGPT conversations. Here's what people use it for (www.washingtonpost.com)

Researchers analyzed 47,000 ChatGPT conversations to map what real users actually ask of large conversational models. The queries run the gamut — from everyday how-tos (product and beauty advice) and factual questions (drug overdose survival rates) to relationship coaching, text analysis, niche ideological prompts (“woke mind virus”), and even existential probes (“are you feeling conscious?”). The scale and variety in the sample make it clear people treat chatbots as multitool assistants: search engines, personal therapists, fact-checkers, creative collaborators and debate partners. For the AI/ML community this matters because usage drives both risk and research priorities. A diverse corpus of sensitive, medical, legal and emotionally charged queries underscores needs for stronger grounding, calibrated uncertainty, domain-specific guardrails and robust safety disclaimers. It also highlights privacy concerns and the importance of logging/consent practices. Technically, the findings argue for evaluation metrics and datasets that reflect real-world conversational intents, hybrid systems that combine retrieval and human oversight for high-stakes queries, and improved prompt-handling to reduce hallucinations and manipulative behaviors. In short: building better chat models requires optimizing for the messy, human-centric ways people actually use them.