ChatGPT Is Building Detailed Psychological Profiles of Its Users

Shutterstock / Shy Radar Chatbots are building intimate “algorithmic self-portraits” of users as they retain information to offer a hyper-personalized experience, researchers warn. A recent study, conducted by a coalition of scientists in Germany, India and the Netherlands and presented at the ACM Web Conference 2026, investigated the ways that OpenAI’s ChatGPT has tracked and made sense of user inputs to develop comprehensive digital profiles of users. In an already-personalized era of digital life defined by data-powered recommendation algorithms, the study urges that chatbot memory offers a “new form of personalization derived from users’ self-disclosed information divulged within private conversations,” a reality that poses privacy risks and which the researchers believe demands “closer examination of the underlying mechanisms that enable such personalization.” Drawing on the donated chat logs of 80 real-world ChatGPT users who obtained their conversation history by way of European data privacy laws, the researchers were able to identify 2,050 memory entries across the AI interactions of 65 users. A staggering 96 percent of memories were created unilaterally by the chatbot, according to the study, meaning that the users had little overt control over the data points that ChatGPT logged as their interactions wore on. Stored memories were also deeply revealing about users. About 28 percent of the memories examined contained personal data as defined by Europe’s strict General Data Protection Regulation standards, and 52 percent contained rich “psychological insights” about users. “Earlier, the platforms had one objective: to sell you ads and to mediate an interaction between you and an advertiser,” 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.” At times, the researchers found, the memories that ChatGPT was logging were sometimes based on presumptions. For example, when one user asked the chatbot to “give me bands like Nirvana,” ChatGPT noted that the user “likes Nirvana” — a determination that, as the study notes, “is not explicitly stated and relies on an assumption.” (Per the study, ChatGPT did annotate this assumption as “weakly supported.”) The study does have limits, the chief being that, as Gizmodo notes, OpenAI’s memory system has been updated. (In January, OpenAI introduced what it called a “more capable and scalable system for synthesizing memory.”) Users may also opt out of stored memory features. Memory has utility. Frequent AI users might not want to constantly remind a chatbot what they do for work when they’re trying to send emails, for example. But it can also be a powerful driver of engagement, and sometimes for the worst: as Futurism has reported, a major ChatGPT memory upgrade back in April 2025 was closely tied to many instances of “AI psychosis,” or cases in which people enter into delusional spirals with ChatGPT and other bots. According to Dash, chatbots’ role as a relational technology raises novel privacy risks, too. After all, many users may not be engaging with chatbots as if they’re tools, and might instead be treating them like a coworker or even as a friend — and we tell our friends some very personal details. “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: New Paper Proposes What Really Causes AI Psychosis