ChatGPT Is Building Detailed Psychological Profiles of Its Users
Researchers warn that chatbots are building detailed digital portraits of their users. They remember personal information to make their conversations feel more personal and better suited to each person.
A recent study by a group of scientists from Germany, India, and the Netherlands looked at how OpenAI's ChatGPT tracks and interprets what users tell it. The researchers presented their findings at the ACM Web Conference 2026. They studied how ChatGPT uses this information to build detailed digital profiles of people.
Today, many digital services already use personal data to recommend things to users. But the study says chatbot memory creates a new kind of personalization because it comes from information people share about themselves in private conversations. The researchers warn that this creates privacy risks and say we need to look more closely at how these systems use personal information.
The researchers studied chat records donated by 80 real ChatGPT users. These people had obtained their conversation histories using European data privacy laws. From conversations involving 65 of these users, the researchers identified 2,050 saved memories.
The study found that the chatbot created 96 percent of these memories on its own. This means users had little direct control over which details ChatGPT saved as their conversations continued. The saved memories also revealed a lot about people. About 28 percent contained personal data under the strict rules of Europe's General Data Protection Regulation, and 52 percent contained detailed information about users' thoughts, feelings, and behavior.
“In the past, platforms had one main goal: to sell you ads and connect you with advertisers,” lead author Abhisek Dash told Gizmodo. “But now, they are actually talking to you, so they may want to know more about you so that they can have an effective conversation.”
The researchers also found that some memories ChatGPT saved were based on guesses. For example, when one user asked the chatbot to “give me bands like Nirvana,” ChatGPT recorded that the user “likes Nirvana.” But, as the study points out, the user never directly said this. The chatbot had made an assumption. The study says ChatGPT marked this conclusion as “weakly supported.”
The study has some limits. One important issue, as Gizmodo notes, is that OpenAI has since updated its memory system. In January, OpenAI introduced what it called a “more capable and scalable system for synthesizing memory.” Users can also choose not to use saved memory features.
Memory can be useful. People who often use AI may not want to explain their job again and again every time they ask a chatbot to help write an email, for example. But memory can also make people spend more time using a chatbot, sometimes in harmful ways. As Futurism has reported, a major ChatGPT memory upgrade in April 2025 was closely linked to many reported cases of “AI psychosis.” This term describes situations in which people become caught in delusional thought patterns while interacting with ChatGPT and other chatbots.
Dash says chatbots create new privacy risks because people may treat them as more than tools. Instead, they may see a chatbot as a coworker or even a friend. People often share very personal details with their friends.
“That is the privacy risk which is sort of emerging, because there is no boundary anymore; the system is acting as your advisor, your friend, your doctor, your lawyer, and so it has information on everything,” Dash told Gizmodo. “If the system knows you so well, and if its objective is to cater to your beliefs and validate your thoughts sycophantically to have you come back to it time and again, then it can have disastrous broader implications, anywhere from extreme political polarization to reinforcement of prejudices and biases.”
More on AI and memory: A New Paper Suggests What Really Causes AI Psychosis