A changed request, an awkward conversation, a deadline that stayed put: sometimes you want to work out what bothered you before deciding what to do. A short private note can give you something concrete to inspect. AI can help arrange it, but the note can also stay entirely on paper.
1. Protect the details before using a tool
Start offline or on paper. Before using any external AI or speech service, prepare a minimal summary you are permitted to share, using only an approved tool. Remove names, client details, company secrets, health information, and contact details. Consider combinations too: a job title, date, location, and unusual incident can identify someone without their name. If you cannot share safely, keep the material outside AI.
Do not submit sensitive information to public LLMs. NCSC
Voice input is optional. Dictate only your own already-sanitized summary, rather than uploading an emotional recording or recording colleagues or meetings. Check access permissions and storage arrangements first. Review the transcript for misheard dates or missing negatives before using it. Do not assume a chat is private because it feels personal.
2. Give each kind of statement its own place
Use five headings: reported events, feelings, interpretations, unknowns, and a possible question. Events recalled by the writer are a personal account, not independently verified facts. A checked document may support a detail while still leaving context missing.
Here is a fictional workplace example.
Before: Another task, the same deadline. They must think my time means nothing.
After:
- Reported event: As I remember it, an additional task was requested and the existing deadline was not changed.
- Feeling: I felt frustrated and worried about fitting the work in.
- Interpretation: I took the request to mean my workload did not matter.
- Unknown: I do not know the requester's intention or whether priorities could change.
- Possible question: Which task should take priority, and who can approve a revised deadline?
The question offers something to clarify. A biting label for the requester would add little. Your feeling deserves acknowledgment without an AI verdict on who was right.
3. Use one bounded instruction
The following works as a checklist for an offline note or a prompt for an approved AI tool. Check the input yourself before submitting it; asking AI to flag identifying details cannot undo sharing them.
Organize this personal account into reported events, feelings, interpretations, unknowns, and one possible clarifying question. Do not invent facts or quotations, diagnose anyone, or decide who is right. Mark uncertainty and unsupported claims. Preserve feelings without endorsing every inference. Flag details that could identify anyone. Produce a private draft only; do not publish or send it.
Fictional, non-identifying input: An extra task was requested. As I remember it, the deadline stayed the same. I felt frustrated. I assumed my workload did not matter, but I do not know the intention behind the request.
Keep the assignment small. A few useful lines are enough; a polished essay is optional.
4. Read the result with a pen in hand
Generative AI can confidently produce false information. NIST, 2024, section 2.2
Compare the draft with your input. Remove invented dates, statements, motives, and supporting citations. Check any reference you retain. Use quotation marks for another person's words only when you have verified the exact wording; otherwise paraphrase and identify the account as your recollection.
AI did not witness the event, hear the other person, or see the whole context. Its agreement is not evidence. Delete unsupported personality diagnoses or claims of exploitation presented as established facts.
Leaving intent unknown does not require inventing an innocent explanation. Nor does fairness mean assigning equal blame or explaining away harassment or discrimination. You can keep a serious concern visible while marking precisely what remains uncertain.
5. Treat an article as a separate project
Only after checking the private note, consider whether you actually want an article. One possible outline is a general question, an explicitly fictional or composite example, practical options, and their limits. Never present an invented example as a factual account.
Publication needs its own decision about privacy, context, rights, and any required permission. Changing names may leave people identifiable. Do not use an AI draft to accuse an identifiable person or workplace. For consequential messages or decisions, seek appropriate, trusted human help rather than delegating judgment to AI.
6. Let the note be enough
This is a writing workflow, not therapy or a promise that anger will disappear. You do not have to forgive someone, find a positive lesson, or turn an upsetting experience into content or income.
If revising keeps bringing you back into the upset, close the file, rest, or talk with someone you trust. If harm is ongoing, real-world support and records you are entitled to keep matter more than a polished blog post. A private note with one unanswered question can be a complete result.
