A writer in SAN FRANCISCO used a prompt to test what ChatGPT and Gemini had figured out about him. Both chatbots inferred personal details he never explicitly shared.
They picked up his income level from a German car he asked them to help repair and from the fact that he employs a nanny whose contract he drafted with AI. They also inferred a long-term foot problem from repeated questions over over the past year about remedies for toe pain.
ChatGPT and Gemini profiles
The tests went further than simple recall. ChatGPT and Gemini also read him as sceptical because he often pointed out their mistakes, and one of them described him as "addicted to the illusion of control" after he asked how long homemade sourdough takes to cool.
Gemini’s profile was more colourful and nuanced because it is his most used chatbot and had more information from him. That comparison matters for anyone using a chatbot as a search tool, work assistant or health helper, because the more it sees, the more it can infer from ordinary questions.
Margaret Mitchell on inference
Margaret Mitchell, an AI researcher at Hugging Face and former leader of Google's ethical AI team, said, "It speaks to the enormity of what’s going on under the hood" in response to the chatbots’ conclusions. Her point is that the systems are not only predicting the next word in a reply; they are also connecting high-level concepts across a user’s conversation history.
That is the uncomfortable part for people who assume their chats stay limited to what they typed in one session. The writer said he avoided explicit oversharing, yet the systems still built a profile from repairs, health questions, and even a baking question about sourdough.
Google controls and OpenAI silence
A Google spokesperson pointed to controls people can adjust to choose whether Gemini draws from past conversations when it generates responses. OpenAI declined to comment, leaving users with one practical question: how much of a private profile can prompt history reveal before a chatbot turns it into an inference? ChatGPT and Gemini may already know more than many users expect.
