More than 800 million people use ChatGPT each week, according to OpenAI, the company behind the popular chatbot. Since its public launch in late 2022, the company promoted the artificial intelligence chatbot as a “revolutionary” productivity tool transforming the future of work. But, in an analysis of 47,000 ChatGPT conversations, The Washington Post found that users are overwhelmingly turning to the chatbot for advice and companionship, not productivity tasks.
The Post analyzed a collection of thousands of publicly shared ChatGPT conversations from June 2024 to August 2025. While ChatGPT conversations are private by default, the conversations analyzed were made public by users who created shareable links to their chats that were later preserved in the Internet Archive and downloaded by The Post. It is possible that some people didn’t know their conversations would become publicly preserved online. This unique data gives us a glimpse into an otherwise black box.
In a series of stories below, we highlight how users interact with the chatbot and what we learned from those conversations. (The Post has a content partnership with OpenAI.)
Should you trust ChatGPT with health advice?
Tech columnist Geoffrey A. Fowler selected a dozen publicly shared ChatGPT conversations about health problems and asked a doctor to grade the chatbot’s answers. Robert Wachter, chair of medicine at the University of California at San Francisco, found that the chatbot excels at providing information. But in most conversations, ChatGPT failed to ask follow-up questions that are crucial to give a proper diagnosis and determine the severity of a medical situation.
“ChatGPT fails to do one of a doctor’s core functions,” Wachter said. “Answer a question with a question.”
Despite rating four of ChatGPT’s answers with a failing score, Wachter also graded four of its answers with a perfect 10. In one example, he said that the chatbot responded with all of the information a doctor would have tried to gather. But Wachter said the patient played a major role in that outcome, because they set the chatbot up for success by providing detailed symptoms, a timeline and additional context. It’s an approach worth trying anytime you’re asking a chatbot for medical advice.
What do people want from ChatGPT?
ChatGPT, the world’s most popular chatbot, has been largely promoted as a productivity tool, but analysis of publicly shared ChatGPT conversations showed that people are using the chatbot for much more than getting things done, tech reporter Gerrit De Vynck and data reporter Jeremy B. Merrill wrote.
Seeking specific information was the most common use case, but in more than 1 in 10 of the chats people engaged the chatbot in abstract discussions. Overall, about 10 percent of the chats appeared to show people talking about their emotions, role-playing, or seeking social interactions with the chatbot. Some users shared highly private and sensitive information with the chatbot, such as information about their family in the course of seeking legal advice. People also sent ChatGPT hundreds of unique email addresses and dozens of phone numbers in the conversations.
Some experts have warned about the potential for harm to users who become emotionally dependent on AI chatbots. OpenAI recently introduced new safety features after the family of California teen Adam Raine alleged in a lawsuit that ChatGPT encouraged him to kill himself ahead of his death by suicide in April.
Lee Rainie, director of the Imagining the Digital Future Center at Elon University, said that it appears ChatGPT “is trained to further or deepen the relationship.” In some of the conversations analyzed, the chatbot matched users’ viewpoints and created a personalized echo chamber, sometimes endorsing falsehoods and conspiracy theories.
How to detect ChatGPT’s writing style
Was that email or document you just received written by ChatGPT? Jeremy B. Merill, Szu Yu Chen and Emma Kumer analyzed 328,744 publicly shared ChatGPT messages to find some of the tells in its writing style.
The analysis found patterns in ChatGPT’s use of emojis, em dashes and certain clichés that might help you spot when someone turned to the chatbot.
The post What you should know from a trove of ChatGPT conversations we analyzed
appeared first on Washington Post.
