How comfortable are people sharing personal thoughts with AI?
Personal disclosure to AI is already meaningful, but it is not universal. In a Human Clarity Institute sample of 501 people, 39% were comfortable sharing thoughts or emotions with AI that they would not share with most people. Comfort rose to 61% among everyday AI users and 82% among people who used AI for emotional support. Yet personal openness did not automatically mean companionship or weak boundaries: most high-disclosure respondents still reported clear limits and did not score highly on AI companionship.
Personal disclosure to AI: the benchmark at a glance
What does this personal disclosure benchmark measure?
Personal disclosure to AI means sharing thoughts or emotions with an AI system that a person would not share with most other people. This benchmark measures respondents' comfort with that behaviour and examines how it relates to use frequency, emotional safety, trust, purpose of use and boundary clarity.
It does not measure how many messages people send, what exact details they reveal, whether they disclose names, health information or workplace data, or whether the system stores or uses the information. It also does not measure whether people tell other humans that they use AI. That separate behaviour is social transparency about AI use.
Personal disclosure is also distinct from companionship and dependence. A person can use AI as a private thinking space without experiencing it as a companion, and can disclose personally while retaining clear emotional boundaries.
How common is personal disclosure to AI?
Comfort is substantial, but the sample is divided
The result describes willingness or comfort, not the sensitivity, accuracy or privacy consequences of what was actually disclosed.
Does frequent AI use change disclosure comfort?
Disclosure comfort was markedly higher among frequent users
Does the reason for using AI matter?
Yes. Disclosure comfort differed sharply according to what respondents used AI for, with the strongest result among people reporting emotional-support use.
Emotional-support use had the strongest association
Why does emotional safety matter for AI disclosure?
Personal disclosure was closely associated with feeling emotionally safe
Perceived emotional safety is a person's reported experience; it is not evidence that a system is confidential, accurate, protective of the user's interests or suitable for every kind of personal information.
Does personal disclosure mean someone sees AI as a companion?
No. Disclosure, emotional support and companionship overlap, but they are not interchangeable. Among respondents with high disclosure comfort, 31% scored highly on AI companionship and 28% scored highly on receiving emotional support from AI. That means most high-disclosure respondents did not score highly on either measure.
Across the full sample, 14% scored highly on AI companionship and 12% on emotional support. Personal disclosure is therefore a broader behaviour than companionship: AI may function as a low-judgment space for reflection, planning or expression without being experienced as a relationship partner.
Can people disclose personally while retaining clear boundaries?
Openness and boundaries frequently coexisted
These findings challenge a simple “more disclosure means fewer boundaries” interpretation. People may experience AI as easier to talk to while remaining aware of risk and retaining limits around the interaction.
At the same time, boundary clarity should not be treated as a guarantee of safe information handling. A user's emotional boundary and a product's technical privacy practices are different things.
What is the HCI Disclosure Boundary Framework?
The HCI Disclosure Boundary Framework is a descriptive model for understanding personal disclosure to AI through four related questions. It is not a set of stages, a diagnosis or a claim that every person has one fixed disclosure style.
Disclosure comfort
How willing is the person to share thoughts or emotions with AI that they would not share with most people?
Perceived emotional safety
Does the interaction feel less judgmental, easier to control or emotionally safer than disclosure to another person?
Trust and purpose
What role is AI serving—reflection, decision support, emotional support or companionship—and how much trust accompanies that role?
Boundaries and risk awareness
Does the person retain clear emotional limits and consider the privacy, accuracy and information-handling risks of disclosure?
What does wider research show about self-disclosure to AI?
People can treat AI as a recipient of personal information
A 2025 experimental study found that participants were similarly willing to choose self-disclosing statements for an AI recipient and a human researcher, and were more willing to do so than in a private condition. The authors also cautioned that the statements were mostly surface-level. The study supports AI's role as a disclosure recipient; it does not show that people disclose every type of sensitive information equally.
Reduced fear of judgment may help, but disclosure findings are mixed
In a randomized study of 286 participants, people reported less fear of judgment when interacting with a chatbot, while trust was higher toward a human. Self-reported disclosure intimacy did not differ between conditions, and perceived anonymity was the only variable in the study to directly affect disclosure intimacy. The result supports a contextual account rather than a simple claim that people always disclose more to either AI or humans.
Disclosure is shaped by the interface, person and context
A systematic review of 26 empirical studies identified five groups of influences on chatbot self-disclosure: interface modality, conversational factors, user characteristics, mediating mechanisms and contextual factors. Perceived anonymity appears repeatedly, but no single mechanism explains every interaction.
Conversational comfort is not the same as technical privacy
NIST identifies privacy risks involving personal data in generative-AI systems, while privacy regulators advise organisations to examine how commercial AI products collect, use, retain and expose personal information. A Stanford HAI review of policies from six frontier companies in 2025 also found meaningful differences and defaults that users could easily misunderstand. Policies and controls can change, so privacy should be checked for the specific product and account settings being used.
Privacy controls are product-specific
For example, OpenAI states that users can turn off the use of conversations for model improvement and that Temporary Chats are not used for training or shown in history, although a safety copy may be retained for up to 30 days. Other products may use different defaults, retention periods and controls.
What do the findings mean together?
Disclosure is substantial, not universal
Thirty-nine percent reported high comfort, while 48% reported low comfort. The benchmark describes a divided behaviour rather than a single social norm.
Use frequency and purpose matter
High disclosure comfort was much more common among everyday users and especially among people using AI for emotional support, although the cross-sectional data cannot determine the direction of influence.
Emotional safety is central
The strongest difference between high- and low-disclosure respondents concerned perceived emotional safety. Feeling less judged may help explain why AI becomes a disclosure space.
Disclosure does not define the relationship
Most high-disclosure respondents did not score highly on companionship, and most still reported clear boundaries. Openness, companionship and boundary clarity should therefore remain separate measures.
Privacy requires a separate check
Feeling safe enough to speak openly says nothing by itself about storage, training, human review, data sharing or legal confidentiality.
What can be concluded from this personal disclosure benchmark?
Personal disclosure to AI is a meaningful part of current human–AI behaviour. In the HCI sample, more than a third were comfortable sharing thoughts or emotions with AI that they would not share with most people, and this comfort was considerably more common among frequent users and people using AI for emotional support.
The evidence does not support treating disclosure as a direct indicator of companionship, dependence or lost boundaries. Most high-disclosure respondents retained clear emotional boundaries and did not score highly on companionship. It is more accurate to understand disclosure through the combination of openness, emotional safety, trust, purpose and risk awareness.
These findings are self-reported and cross-sectional. They do not measure the content or sensitivity of actual conversations, establish causal effects, determine whether a specific disclosure is safe, or diagnose an individual's relationship with AI.
How can I compare my personal disclosure to AI with other people?
Group findings show how disclosure appears across an HCI sample, but they cannot determine where one person sits. Frequency alone also misses whether disclosure occurs alongside emotional safety, trust, companionship, clear boundaries or retained human agency.
The Human Clarity Institute AI Identity & Behaviour Assessment asks 39 questions across nine dimensions of AI behaviour. Free personalised results compare your responses with all HCI participants, people in your age group and people who use AI as often as you do.
The assessment provides a behavioural reference point for reflection and comparison. It is not a clinical diagnosis, a privacy audit or an intelligence test.
Frequently asked questions about personal disclosure to AI
Do people share personal thoughts with AI?
Yes, although the behaviour is not universal. In HCI's AI Companionship & Human Connection sample, 39% reported high comfort sharing thoughts or emotions with AI that they would not share with most people, 13% were neutral and 48% reported low comfort.
Is sharing with AI the same as trusting AI?
No. Disclosure and trust were related in the HCI sample, but they are separate behaviours. A person may disclose because an interaction feels low judgment while remaining cautious about the system's accuracy, motives or data practices.
Does frequent AI use increase personal disclosure?
High disclosure comfort was more common among frequent users: 61% of everyday users reported high comfort, compared with 20% among people using AI occasionally or less. The cross-sectional data cannot show whether use increased disclosure or disclosure comfort encouraged more frequent use.
Does using AI for emotional support change disclosure?
It was strongly associated with disclosure comfort. In the HCI sample, 82% of people using AI for emotional support reported high disclosure comfort, compared with 30% of other respondents. This is an association, not proof of causation.
Does personal disclosure mean AI companionship?
No. Only 31% of high-disclosure respondents scored highly on AI companionship, so most did not. Disclosure may support reflection, expression or decision-making without AI being experienced as a companion.
Can someone share personally and still have clear boundaries?
Yes. Sixty-nine percent of high-disclosure respondents reported clear emotional boundaries, almost the same proportion as among low-disclosure respondents. Emotional boundaries and technical privacy protections remain separate issues.
Are conversations with AI private?
Not automatically. Storage, training, retention, human review and sharing practices vary by product, account type and settings. Users should check the current privacy policy and controls for the specific service rather than assuming conversational comfort means confidentiality.
How can I compare my disclosure behaviour with other people?
The Human Clarity Institute AI Identity & Behaviour Assessment asks 39 questions across nine dimensions of AI behaviour and provides free personalised comparisons with all HCI participants, people in your age group and people who use AI as often as you do.
Related HCI benchmarks
Suggested citation: Human Clarity Institute. (2026). Personal Disclosure to AI Benchmark: Sharing, Emotional Safety and Boundaries. https://humanclarityinstitute.com/benchmarks/ai-disclosure/
Methodology: Learn how HCI builds benchmarks.
How does your personal disclosure to AI compare?
Compare your responses with HCI participant benchmarks across nine dimensions of AI behaviour—including personal disclosure, emotional regulation, trust, reliance and human agency.
See Where I Sit →- 01All participants in the benchmark
- 02People in your age group
- 03People who use AI as often as you do