When AI Feels Easy to Confide In

Imagine a creator trying to make sense of difficult feedback. They open an AI conversation and begin describing what happened. Writing feels easier than arranging a call or explaining the whole situation to a colleague. Soon the account includes names, private messages and details about an unfinished project.

This is a hypothetical situation, but it poses a useful question: what made those details feel appropriate to share? The answer may involve the experience of the conversation. A separate answer is needed about how the service handles the information. Both deserve attention.

The audience we imagine can shape disclosure

In a study published online on 21 May 2014, Lucas and colleagues investigated interactions with a virtual human interviewer in a health-screening context. Participants were led to believe the interviewer was controlled either by a person or by a computer. Those who believed it was computer-controlled reported less fear of disclosure and less effort to manage the impression they made.

This was an experiment with a virtual interviewer, rather than a test of today’s general-purpose AI assistants. Its relevance is specific: beliefs about who or what is receiving our words can influence how comfortable we feel expressing them. It does not establish that every person will disclose more to every chatbot.

For ordinary creative work, this suggests a question worth noticing: am I choosing these details because they help explain the task, or because this particular setting makes it easier to keep talking? There is room to value the ease while making the choice deliberately.

Comfort and privacy answer different questions

Comfort concerns the experience: whether you feel rushed, embarrassed, understood or able to express an unfinished thought. Privacy concerns what information is collected, how it is used, who can access it and what controls apply. A reassuring conversational style cannot answer those operational questions.

The distinction is relevant to current scrutiny of AI services. In an update dated 8 October 2026, the UK’s ICO identified increasing personalisation in consumer-facing AI, including general-purpose, role-play and companion chatbots, as a priority area for its work. It emphasised transparency and people’s understanding and control of their data.

For the service you use, look for current information about retention, access and use, together with the controls available in your account. If the task involves client material, MaryChuks’ guide to checking an AI tool before sharing client data provides a more detailed starting point.

A useful pause can preserve the choice

An interesting design experiment appears in Disclose with Care, a preprint submitted on 1 February 2026. Li, Xiao and Li studied chatbot interviews about using AI in job interviews. They compared no transcript editing, free editing and AI-aided editing after the interview. The authors report that AI-aided editing reduced personally identifiable information in the revised text.

The outcome needs a clear boundary. This was a US study on one interview topic, not evidence about every conversational setting. The researchers collected original and edited records for analysis, with a separate choice about sharing the original. Reduced identifying detail in a revised transcript is a narrower result than proving that information already processed has been removed everywhere.

My practical interpretation is to create a review moment before sharing an account. Read it once for meaning, then once for identifying details. Decide which facts the task needs. This is a proposed habit, rather than a claim that ordinary chat products offer the study’s privacy-control feature.

Keep the meaning without every identifier

Return to the hypothetical creator. Their goal might be to prepare a calm response to feedback. The relevant facts are that a collaborator criticised the draft, the creator felt discouraged, and the next meeting needs a clearer agenda.

They could describe that situation without supplying the collaborator’s name, company, private email address or complete message history. A short account might say: “A collaborator says the opening of my draft is unclear. I want to ask which section confused them and agree what to revise before our next meeting.”

That version preserves the practical issue. It also leaves the creator free to reflect on embarrassment, frustration or uncertainty in their own terms. Personal meaning and identifying detail are not identical: an account can convey why something matters without documenting every person involved.

Sometimes a task genuinely requires precise information. At that point, make a fresh decision about the appropriate tool and setting. Treat that need as a reason to examine the data conditions, rather than an invitation to continue automatically.

Give disclosure a purpose

Before beginning, name the kind of help you want. For example:

• Organisation: separate events, interpretations and unanswered questions.
• Preparation: draft questions for a conversation you intend to have.
• Exploration: consider several possible explanations without settling on one.
• Editing: make your own account clearer while preserving uncertainty.

These are suggested uses, not promises about a system’s accuracy. Review the result for additions and assumptions. If an answer introduces a motive, diagnosis or event that you did not provide, keep it separate from your account of what happened.

At the end, ask what you want to take away. A short list of questions or a revised paragraph may be enough. You can decide that the task is complete even when further conversation remains possible. The amount disclosed should serve your purpose.

Warm interaction can leave room for boundaries

For people designing conversational products, I would make optional questions visibly optional, explain why sensitive details are being requested and let users continue with a more general description where the task permits it. Place explanations near the relevant choice, in language the intended audience can understand.

A warm response can acknowledge a concern while respecting a decision to withhold detail. For instance, a planning assistant could help organise the next step without asking for a complete private history. These are design proposals: their usefulness and effects need testing with the people who will use the product.

The value of an easy conversation does not depend on surrendering every boundary. A thoughtful approach leaves room for both expression and restraint: understand the service, choose the purpose, share the detail that serves it and keep the right to pause. Feeling comfortable is worth recognising. Deliberate choice is worth preserving alongside it.

Sources

• Lucas et al.: It’s only a computer: Virtual humans increase willingness to disclose, Computers in Human Behavior (2014).
• Li, Xiao and Li: Disclose with Care: Designing Privacy Controls in Interview Chatbots, preprint (1 February 2026).
• ICO: AI developer supervision update and priorities (8 October 2026).