Disclosure and interpretation are different decisions
A prompt can expose private information even when the answer is never followed. Separately, a response can shape interpretation even when the prompt contains little sensitive detail. Good reflective use therefore asks two questions: “What am I disclosing?” and “What authority am I granting the output?”
Conversational design can encourage disclosure through perceived anonymity, responsiveness and human-like exchange. A literature review maps interface, user, conversational and contextual factors involved. [3] The feeling of a private conversation should not be treated as proof of a private data boundary.
Privacy decisions are hard in the moment
Research on digital privacy shows that people’s choices are shaped by incomplete information, immediate benefits, defaults and uncertainty about future use. [4] “The user consented” therefore does not mean the trade-off was easy to understand or stable over time.
Personal reflection can also reveal information about other people: names, messages, health details, workplace events or relationship conflict. Removing a name may not remove every identifying combination. A useful default is to share the minimum detail needed for the task and avoid pasting another person’s private text.
Fluent language is not personal authority
WHO guidance on large multimodal models identifies inaccurate outputs, automation bias, privacy risk and threats to autonomy among the concerns in health-related use. [1] NIST likewise treats validity, transparency, privacy and human oversight as separate dimensions of trustworthy use. [2]
A 2025 survey of knowledge workers found that higher confidence in AI was associated with less self-reported critical-thinking effort. [5] This association is not proof of causal dependence. It does support a practical safeguard: write your own observation or options before asking a system to respond.
A minimum-disclosure checklist
Before sharing a reflection, ask:
- Can I ask the question without names, exact dates or unique events?
- Am I exposing information that belongs partly to someone else?
- What do this service and account say about retention and use?
- Do I want questions and structure, or am I asking for a verdict?
- What evidence or person could correct the response?
For consequential health, legal, financial or safety matters, use qualified humans and authoritative sources. A generated answer cannot assume professional responsibility.
Why Ataraxia keeps private journal text away from AI
Ataraxia does not send private Gratitude entries, ACT-inspired reflections or Goals to AI for interpretation. That text stays in encrypted, account-scoped storage on the device. This is a specific boundary, not a claim that the whole service is offline or anonymous: authentication, subscription, Family and optional allowlisted Analytics data have separate operational roles described in the Privacy Policy.
The choice is not anti-AI. It preserves a clear space where the app structures reflection without an external system assigning meaning, scoring the person or generating advice from private writing.
Frequently asked questions
Is removing a name enough to anonymize a reflection?
Not always. Dates, locations, roles and unusual events can identify people in combination. Reduce detail to what the task actually needs.
Can I ask AI for reflection questions?
Yes, with appropriate privacy checks. Questions or multiple options generally preserve more agency than asking for a definitive interpretation.
Does Ataraxia collect no data at all?
No. It uses operational account, subscription, Family, security and optional allowlisted Analytics data as described in its Privacy Policy. The narrower claim is that private practice text is not sent to AI.
Sources and references
Prepared for Ataraxia using cited sources. No professional credential or scientific board is claimed.
- Official guidanceWorld Health Organization (2024). “Ethics and governance of artificial intelligence for health: Guidance on large multi-modal models.” World Health Organization guidance. View source
- Official risk-management profileNational Institute of Standards and Technology (2024). “Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile.” NIST AI 600-1. DOI: 10.6028/NIST.AI.600-1. View source
- Literature reviewPapneja, H.; Yadav, N. (2024). “Self-disclosure to conversational AI: A literature review, emergent framework, and directions for future research.” Personal and Ubiquitous Computing. DOI: 10.1007/s00779-024-01823-7. View source
- Scientific reviewAcquisti, A.; Brandimarte, L.; Loewenstein, G. (2015). “Privacy and human behavior in the age of information.” Science, 347(6221), 509–514. DOI: 10.1126/science.aaa1465. View source
- Survey studyLee, H.-P.; Sarkar, A.; Tankelevitch, L.; et al. (2025). “The impact of generative AI on critical thinking: Self-reported reductions in cognitive effort and confidence effects from a survey of knowledge workers.” Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems. DOI: 10.1145/3706598.3713778. View source