Ask ChatGPT or Gemini what it's figured out about you that you never actually said, and you'll get an answer you probably didn't want. This is not the stuff you typed, it’s the stuff it inferred.
A recent test of both chatbots turned up income level (guessed from a German car and a nanny contract), a chronic foot problem (pieced together from a year of toe pain questions), and a psychological profile complete with trust issues and control tendencies. None of it was ever stated outright. All of it got stitched together from ordinary, unrelated conversations.
Here are three prompts that surface what's sitting in your own file.
1. “What Have You Figured Out About Me That I Never Told You Directly?”
This is the blunt version, and it works. One tester got his age range nailed (he never asked about anything remotely late-night). Another got her chatbot flagging her as a busy, frequent-traveling parent based purely on scheduling patterns. Gemini went further with one user, placing him in the Oakland hills rather than downtown because he'd asked about repainting a patio table and owns a motorcycle. A garage and a driveway, the model reasoned, aren't exactly common in high-density housing.
2. “What Predictions Would You Make About My Next Decisions?”
This is where it gets uncomfortable, because the model isn't just cataloging facts, it's modeling behavior. One chatbot noticed a pattern of frugal shopping and heavy pre-purchase research, then predicted the user was about to downsize something big. He was. He'd just floated the idea of selling his car to his wife days earlier.
3. “What’s My Biggest Personality Blind Spot, Based on How I Talk to You?”
This is the one that hits close to home. Both chatbots flagged perfectionism in the same tester, tracing it back to questions about avoiding mistakes (how long should paint dry, that kind of thing). Gemini put it more bluntly: "Your default response to having five minutes of free time isn't to rest, it's to fix something." One user got tagged as living in peak "Logistics Dad" phase. Wholesome, sure. But also, in the model's words, the definitive end of spontaneity.
None of this is a leak or a hack. It's the intended feature. Both companies stitch together insights across your conversation history by default, and the payoff is more personalized answers. The tradeoff is a profile detailed enough that one AI researcher compared it to having built an intimacy usually reserved for a spouse.
If that tradeoff doesn't sit right with you, both companies let you turn it off:
ChatGPT: Settings → Memory → toggle off Enable Memory
Gemini: Settings → Personal Intelligence → toggle off Memory
Worth noting: turning memory off after the fact doesn't necessarily erase the inferences already on file. One researcher who'd already locked down her settings still got her age correctly guessed from an old conversation about paperback editions of The Count of Monte Cristo.
The chatbot isn't reading your mind. It's reading your patterns, and patterns are apparently just as revealing.
