AI Chatbots Draw Secrets Users Hide From Doctors and Family

32% of AI users have told a chatbot something they kept from close friends, parents or medical professionals, according to a DuckDuckGo survey of 1,944 U.S. adults. Among self-described AI enthusiasts the figure climbs to 56%. Parents confide at nearly twice the rate of non-parents, even as most respondents remain unaware that the same conversations train models by default and can be subpoenaed.

The findings, released August 25, 2026, capture a quiet shift: chatbots have become judgment-free outlets for material many people still withhold from humans. That appeal collides with how most major services handle the data.

Nearly One in Three Users Share Secrets They Keep From Everyone Else

DuckDuckGo commissioned the survey through the PureSpectrum panel between June 17 and June 27, 2026. Quotas matched U.S. Census age, gender and region. Results are unweighted; the full-sample margin of error is about ±2.2 points.

Among AI users, 32% said they had shared information withheld from close friends, parents or medical professionals. For enthusiasts who use chatbots regularly, the share reached 56%. The DuckDuckGo AI privacy survey results frame these chats as longer and more revealing than typical search queries. They can expose thought processes and communication styles, not just interests.

• 32% of AI users shared a secret kept from trusted people
• 56% of self-described AI enthusiasts did the same
• 43% of parents who use AI told a chatbot something never shared with another person
• 25% of AI users without children at home reported the same behavior

The parent gap stands out. DuckDuckGo notes that even some AI executives have spoken publicly about leaning on chatbots for parenting questions. The survey numbers show the pattern is widespread among households with children.

Parents Confide at Almost Double the Rate

Among parents or guardians with children in the household who use AI, 43% reported telling a chatbot something they never told a doctor, therapist, close friend, parent or child. That is almost twice the 25% rate among AI users with no children at home.

The company links the difference to the daily pressure of parenting decisions and the ready availability of an always-on, non-judgmental listener. The same availability that draws the confidences also feeds systems that store and process them under default rules many users never check.

The Judgment-Free Space AI Companies Market

DuckDuckGo’s report states that for many people AI chatbots serve as a judgment-free space to share things they will not tell anyone else. Conventional AI companies have encouraged the confidant framing in advertising campaigns and media appearances.

Those messages rest on an assumption of privacy that does not match how most services operate. Conversations are typically tied to user profiles, stored, and used for model training unless the user actively opts out. The report cites prior incidents in which personal chats became public through leaks or indexing, exposing names, addresses and other identifiers linked to chat histories.

Crowd reaction on X after the survey release pointed to the same mismatch. One reply to DuckDuckGo’s announcement called the pattern “a huge opportunity for data mining by unethical tech companies.” Earlier posts about shared Claude and ChatGPT links that were indexed by search engines had already surfaced medical details, credentials, internal work notes and deeply personal exchanges that users never expected to become public. The survey numbers quantify how common the underlying confiding habit has become.

What Most Users Still Do Not Know About Their Chats

Awareness lags far behind behavior. 53% of respondents did not know or were unsure that most chatbots train on their conversations by default. Only 25% realized AI chats can be subpoenaed by law enforcement.

DuckDuckGo listed six basic facts about how conventional AI chats are stored, reviewed and disclosed, then measured awareness:

• AI chats can be subpoenaed by courts or law enforcement
• AI chats can be made public in a lawsuit
• Employers can access work or enterprise AI accounts
• Settings to delete or stop storage vary by provider
• Free accounts often carry weaker privacy than paid ones
• Human employees can review chats

Across users and non-users, 43% did not know any of the six. That included a third of self-described active AI users. Once people learned how most companies use the chats, 58% said they were uncomfortable with training on their conversations; 30% specified “very uncomfortable.”

Trust Scores Barely Beat the Government

Only 14% of respondents said they mostly or completely trust large AI companies to protect their data. 39% reported no trust at all. The U.S. government scored worse: 48% had no trust at all. People still place slightly more confidence in the AI firms than in government, yet both numbers show deep skepticism.

That gap between low trust and high confiding volume is the core irony. Users withhold material from humans precisely because of judgment risk, then hand the same material to systems they largely distrust on privacy.

How Default Training and Subpoenas Work

Most major consumer chatbots train on conversations by default unless the user finds and flips an opt-out. OpenAI’s ChatGPT, Google Gemini and, since a 2025 terms update, Anthropic’s Claude all follow versions of that pattern for personal accounts. Temporary or incognito modes often escape training, but ordinary chat history does not. Retention windows and human-review samples differ by provider and can stretch years in some cases.

Legal exposure is real. Courts and law enforcement have already sought AI chat records. Employers can reach enterprise accounts. Shared conversation links have previously been crawled and indexed, turning private-seeming exchanges into searchable pages.

When we talk about online privacy in search engines and browsers, many people think “it’s too late, my information is already out there.” But the most valuable information you could share with Big Tech is your next question, not the questions you asked days or months or years ago. So, it’s never too late to make the switch to more private services online.

DuckDuckGo research report, August 2026

The company stresses that AI chat habits are still relatively new for most people, which leaves room to change course before the pattern hardens.

Private Options Still Exist While Habits Form

DuckDuckGo built Duck.ai as one alternative. By default it does not record or store chats and does not use them to train models. Prompts are anonymized; identifying metadata such as IP address is stripped before the request reaches model providers. Providers agree to zero data retention after generating a reply, no training on the Duck.ai traffic, and no sharing for third-party commercial use, with limited documented exceptions.

Details appear in the Duck.ai zero data retention agreements and the full Duck.ai privacy policy terms. Some models add trusted-execution environments so even the hosting provider cannot read the prompt. Local recent-chat storage on the device is optional and can be cleared.

Other practical steps include checking each chatbot’s data-controls menu for the training toggle, using temporary chat modes for sensitive topics, and preferring services that publish clear retention and review policies. Users already familiar with DuckDuckGo’s other tools, such as its DuckDuckGo Clean Link tracker stripping and the browser that blocks YouTube ads by default, will recognize the same privacy-first defaults applied to chat.

Until regulations and industry norms tighten, the survey argues the burden still falls on individuals to choose tools that match the sensitivity of what they type. The numbers show many are already typing far more than they realize.

Frequently Asked Questions

How many people did DuckDuckGo survey about AI privacy?

DuckDuckGo collected responses from 1,944 U.S. adults aged 18 and older between June 17 and June 27, 2026, via the PureSpectrum panel, with quotas for age, gender and region matching Census demographics; results are unweighted and carry a full-sample margin of error of approximately ±2.2 percentage points.

What share of AI users have shared secrets with chatbots they keep from humans?

32% of AI users reported telling a chatbot something they kept from close friends, parents or medical professionals; among self-described AI enthusiasts the figure rises to 56%, and among parents who use AI it is 43% versus 25% for users without children at home.

Do most AI chatbots train on conversations by default?

Yes for the largest consumer services: ChatGPT, Gemini and Claude (after Anthropic’s 2025 consumer-terms change) use chats for model improvement unless the user actively opts out via a settings toggle; temporary or incognito modes typically escape training.

Can law enforcement obtain AI chat records?

Yes; only 25% of survey respondents knew that AI chats can be subpoenaed by courts or law enforcement, and the report cites prior cases in which chat logs have been sought in legal proceedings.

How does Duck.ai handle chat privacy differently?

Duck.ai by default does not store chats, strips identifying metadata before forwarding prompts, requires model providers to delete data immediately after reply generation under zero-data-retention terms, and prohibits training on those conversations, with additional technical protections such as trusted execution environments on selected models.