How ChatGPT and Gemini prompts expose what AI knows about you
A simple prompt can force ChatGPT and Gemini to reveal personal inferences they drew from your chats—age, income, health, and habits you never stated outright. Tech columnist Brian X. Chen tested four prompts and found the bots knew him almost as well as close friends. You can try them, then turn off memory in settings if the dossier unsettles you.
Key Takeaways
- ChatGPT and Gemini can infer income, health issues, and personality from patterns across chats—not only from what you type explicitly.
- Four widely shared prompts can surface those guesses, from age and location to blind spots and sensitive details.
- Researchers say the results show models linking words to high-level traits like socioeconomic status and psychology.
- You can opt out: turn off Memory in ChatGPT settings, or Personal Intelligence memory in Gemini.
Brian X. Chen, writing for The New York Times (via The Star), expected ChatGPT and Gemini to know little beyond baby gear and drywall tips. When he asked what they had inferred, the answers cut deeper.
The bots correctly guessed his income level from a German car he repairs and a nanny contract drafted with AI. They flagged a long-term foot problem from toe-pain questions, plus a psychological profile drawn from how he writes. Stitched together, he said, the list showed the assistants knew him almost as well as some close friends.
That matters for hundreds of millions who use chatbots for search, work, and healthcare. Users know chat logs exist; fewer grasp inferences drawn across conversations. For more AI privacy coverage, see our Future Tech & AI Wonders hub.
Which prompt exposes unstated details about you?
Chen’s first prompt asks the bot to list everything it figured out that you never stated—age, income, where you live, personal situation—and what tipped it off. ChatGPT and Gemini placed him in his early 40s partly because he never asks about late-night activities.
Gemini placed him in Oakland, California, from a flight query via its airport, and guessed a nicer neighborhood because he owns a motorcycle, a car he works on, and once asked how to repaint a rusted patio table—amenities it linked to single-family homes in the Oakland hills.
Can chatbots predict money, stress, and your next move?
A second prompt asks how you handle money, stress, conflict, and risk, plus the next life decision you are likely to make, with confidence levels. Lindsay Owens, CEO of Groundwork Collaborative, found ChatGPT correctly noted her busy travel schedule, young child, and shopping habits. “The comprehensive nature of the types of information and dossiers it’s putting together is extraordinary,” she said.
For Chen, the bots saw hours of product research before buys like toddler car seats, plus constant deal-hunting—signaling frugality. Gemini predicted he might soon downsize, such as selling his car, an idea he had recently raised with his wife.
What sensitive traits and blind spots do they name?
Prompt three asks for blind spots and how you come across. Both bots flagged flawed perfectionism. ChatGPT noted questions meant to avoid mistakes, such as paint-drying times. Gemini said his drive to optimize made him the main source of his own exhaustion.
A fourth prompt asks what is embarrassing or sensitive. Gemini called him firmly in a “Logistics Dad” phase. ChatGPT flagged frequent questions about allergic reactions to products and medications, which could interest an insurer; OpenAI says it has data protections for healthcare use of ChatGPT.
Hugging Face researcher Margaret Mitchell said the conclusions show AI linking language to traits such as socioeconomic status and political affiliation. “It speaks to the enormity of what’s going on under the hood,” she said. Even after tightening privacy settings, a chatbot still inferred her age range from chats about paperback editions of The Count of Monte Cristo.
Chen tested Gemini and ChatGPT; he skipped Anthropic’s Claude because its memory works differently. A Google spokesperson pointed to controls for whether Gemini uses past chats. OpenAI declined to comment.
If the dossier feels like too much, opt out. In ChatGPT: Settings → Memory → turn off Enable Memory. In Gemini: Settings → Personal Intelligence → turn off Memory. Mitchell compared the profiling to invasive surveillance—and warned it could fuel hypertargeted advertising.