Use an AI companion for reflection without treating it as an authority

Use an AI companion for a bounded task such as generating a journal question, rewording a plan, or organizing observations you supply. Check factual claims before acting on them, and keep decisions about health and treatment with a qualified professional. A fluent, encouraging answer is not proof that the system understands your situation or that its advice is correct.

NIST identifies confidently incorrect output as a risk of generative AI. Its guidance also discusses risks involving personal data and the way people relate to AI systems. Those are general risks, not an assessment or certification of Unhook. See the NIST Generative AI Profile.

Give the conversation a small job

Choose one output you can evaluate for yourself. These are original prompt examples, not tested therapeutic exercises:

• “Give me three neutral questions about why I keep checking my phone during study time. Do not infer a diagnosis.”
• “Turn this into a specific plan with one fallback: I want to leave my phone outside the kitchen during dinner.”
• “Help me shorten this check-in to two sentences. Keep the facts I gave you and do not add an explanation for my behavior.”
• “List two possible adjustments to this routine. Ask me about practical constraints before recommending one.”

You can reject a suggestion because it does not fit. You can also finish the task without continuing the conversation. A prompt that asks for caution does not guarantee an accurate or appropriate answer, so the review still belongs to you.

Separate a suggestion from a claim

Imagine an AI answer gives you both a proposed evening routine and a statement about how quickly your brain will change. These are different kinds of output. You can inspect the routine for practical fit; the biological statement needs credible evidence.

Use this review before keeping an answer:

1. Underline the facts. Mark percentages, diagnoses, timelines, and claims about research.
2. Look for a real source. Open the source yourself. A convincing title or citation in an AI answer may be wrong.
3. Check the match. Does the source actually study the behavior, people, and outcome being discussed?
4. Remove invented context. Delete any assumption about your history, relationships, beliefs, or intentions that you did not provide.
5. Keep only the useful next step. Rewrite it in your own words. If the suggestion involves a health decision, take that question to an appropriate professional.

The recovery-claim checklist provides a longer worksheet for that source check.

Make a separate privacy choice

An AI chat is not automatically a private diary. Before pasting an entry, decide whether a shorter, non-identifying description would do the job. For a planning prompt, the names of other people, an exact address, or a transcript of an intimate conversation usually add detail you can choose to leave out.

Unhook's Privacy Policy states that messages sent to Lumina are transmitted to OpenAI for processing. Read the current policy for its explanation of service providers and retention. That disclosure describes stated practices; it is not a guarantee that a chat stays only on your device.

Know when the task belongs elsewhere

Unhook describes Lumina as an AI companion, not a therapist, and its tools do not provide medical diagnosis or treatment. Do not rely on a chat to assess an emergency or tell you whether a health concern needs attention. Use human support appropriate to the situation; Unhook Support lists urgent-help options.

For ordinary reflection, the useful result may be one question you answer yourself. Keep your own observations, your boundaries, and your judgment at the center.