Survey Reveals How Deeply People Trust Chatbots with Personal Information

A new DuckDuckGo study shows that more and more people are using artificial intelligence not only for work, searching for information, or entertainment, but also as a trusted friend they can tell everything to. For many, a chatbot seems like a calm and impartial space where they can share their thoughts, worries, and personal experiences without fear of being judged.

According to the study, which was conducted among nearly 2,000 American adults who use AI, almost one in three—32%—admitted that they had told a chatbot something they had not shared with a close friend, parent, colleague, doctor, or therapist. Among people who describe themselves as AI enthusiasts, the figure is even higher, at 56%.

The difference between people with children in their household and those without children is particularly striking. Among parents who use artificial intelligence, 43% have told a chatbot something they had never shared with another person. Among users without children at home, the figure is just 25%. One possible reason is that parenthood often brings questions and worries that people may feel uncomfortable discussing openly, yet find it easier to share with a digital “advisor.”

The problem is that the perception of privacy does not always correspond to reality. The study shows that many users do not know what happens to their conversations after they send them to an AI tool. Fifty-three percent of respondents were unsure or unaware that, with many chatbots, conversations may be used to train AI models by default.

Even fewer people are familiar with the legal risks. Only 25% knew that AI conversations can be requested by courts or law-enforcement authorities. Users are also often unaware that conversations may become public as part of a legal case, that employers may have access to work-related AI accounts, or that storage and deletion settings vary significantly between services.

According to the results, 43% of all participants were unaware of any of the six basic facts about the way AI platforms store, review, or disclose conversations. This was also true of a significant proportion of active artificial intelligence users.

When people learn more about these practices, their attitudes change. Fifty-eight percent said they felt uncomfortable with their conversations being used to train AI, while 30% described their discomfort as “very strong.” Trust in major AI companies is also limited: only 14% of respondents have significant confidence that these companies will protect their data, while 39% said they have no trust in them at all.

Distrust of institutions is even greater. Nearly half of the participants—48%—said they have no trust at all in the government to protect their personal data.

The study highlights an important change in the way we communicate with technology. Traditional internet searches usually consist of a few brief words. A conversation with a chatbot, however, can reveal much more—personal fears, ways of thinking, relationships, health concerns, and details about everyday life.

That is precisely why the feeling that a conversation with a chatbot is “just between us” can be misleading. Depending on the service, conversations may be stored, linked to user profiles, reviewed by employees, or used to improve AI models. In some cases, they may also be disclosed through legal proceedings. There is also a risk of data breaches, in which personal conversations could become accessible to other people.

The study’s conclusion is that users should be more careful about the information they share with AI. While regulations and industry practices continue to evolve, the most sensible approach is to choose services with clear data policies and settings that limit the storage and use of conversations.

The survey was conducted online between June 17 and June 27, 2026, among 1,944 adults living in the United States. The results were balanced according to the demographic characteristics of the population by age, gender, and region. The approximate margin of error for the full sample is ±2.2 percentage points at a 95% confidence level.

Andrey Hristov