HOW TO USE AI
How to use AI fairly and safely
AI usually sees only your side and, by default, tends to soothe you. Here are rules for turning it into a mirror (self-reflection), not a judge (an alibi).
0) Mandatory disclaimer (at the start of every conversation)
Copy–paste
“Be fair and critically factual — without a flattering tone and without automatic agreement. You only see my version. First look for my share of responsibility, and only then for the other person’s mistakes.”
1) AI sees only one side
- AI doesn’t know the other person or the context. If you give it a one-sided story, it will often respond one-sidedly.
- Force this: “Write 2–3 credible alternative interpretations of the other person’s behavior (neutral / most charitable / toughest plausible — without diagnoses).”
2) Separate facts, interpretations, and assumptions
- Facts: what objectively happened (who said/did what).
- Interpretations: what you make it mean (“this means that…”).
- Assumptions / projections: what you’re only guessing (“they must have wanted…”).
- Ask AI to state confidence for conclusions (low/medium/high).
3) Priority: your share first
- This isn’t self-flagellation. It’s about control: what you can change.
- Force the question: “Where did I make it worse? What was my mistake in communication style?”
4) A brake on irreversible moves
- In the heat of emotion, don’t do: ultimatums, blocking, “final messages,” “break up right now.”
- AI should first run a STOP-check: risks, long-term consequences, and a reversible alternative.
- Rule: delay 48 hours for irreversible decisions.
5) A reversible step instead of judgments
- A “reversible step” = a small, safe move forward that doesn’t close anything forever.
- Typical forms: set a time to talk, name a need, offer a compromise, apologize for tone, ask for clarification.
- Every answer should end with: 1 concrete step within 24 hours (what you will do).
6) What not to ask AI for
- “Tell me who’s to blame.” (it has only one version)
- “Confirm I’m right and the other person is wrong.” (relief, but it hardens the conflict)
- “Diagnose them.” (labels instead of working with behavior)
- “Write an ultimatum for them” while you’re emotional (AI can write it, but you have to live with it)
7) When to trust AI less
- When you’re in strong emotions (anger, panic, euphoria) — slow down, delay.
- When you want “quick relief” — risk of having your narrative validated.
- When it involves safety, trauma, addictions, repeated destructive cycles — prioritize a real person/therapist.
8) Clear limits of AI
- AI is not a therapist and doesn’t carry responsibility for the consequences of your actions.
- AI is good at: structuring thoughts, offering alternatives, de-escalating, drafting a factual message.
- AI cannot: verify the other side’s reality, read nonverbal cues, know the history of the dynamic.
Recommendation: when in doubt, always ask in a way that leads to self-reflection, not ego validation.