One in three users of artificial intelligence chatbots has already shared intimate information with these tools that they would never reveal to friends, family, or healthcare professionals. This finding is part of an unprecedented investigation released this Tuesday, August 25th, by DuckDuckGo, the company responsible for the browser of the same name.
The survey aimed to analyze how people manage their privacy when interacting with AIs in the United States. The results indicate that a large portion of users places trust in chatbots, believing that their conversations are kept in complete confidence. Among enthusiasts of this technology, the rate of those who have shared such information reaches 56%.
The research points out that factors such as family dynamics and daily routines directly influence the degree of trust granted to the technology. For example, parents or guardians living with children show almost twice the trust in artificial intelligence compared to those living alone in the same residence.
Among AI users who are parents, 43% admitted to having told a chatbot something they would not share with anyone else, including spouses, children, doctors, or therapists. In contrast, this percentage drops to 25% among individuals without children at home. According to the study, this disparity is due to the inherent difficulties of raising children, where doubts, insecurities, and daily pressures motivate many guardians to seek guidance without fear of judgment. Technology companies, moreover, take advantage of this situation in their advertising campaigns, promoting AI chatbots as a safe space for venting and advice.
Additionally, the study reveals that most users are unaware of the final destination of their data. In traditional services, interactions are not always private; they are automatically associated with user profiles, stored on third-party servers, and used as a database for training new language models, unless the user disables this feature.
The data details several blind spots for the public, such as employer access to corporate accounts, the fact that free accounts have less rigorous security standards than paid ones, and the possibility of human employees of technology companies reading transcripts for review purposes. When users are informed about the real functioning of the sector, dissatisfaction is immediate. The research recorded that 58% of people expressed discomfort with using their dialogues to train algorithms, with 30% classifying this feeling as 'very uncomfortable'. Additional concerns are reinforced by past data leakage incidents in the sector, which exposed conversation histories.
Data collection was conducted by DuckDuckGo between June 17 and 27, 2026, using an online sample composed of 1,944 adults residing in the United States.