How to Write an Apology Note that Sounds Like You

Anyone who owes a real apology can write an apology note that sounds like you in 15 minutes: name the harm, own one fact, state one repair, then cut the speech. Run ChatGPT, Claude, or Gemini as a clerk, not as a priest. You still have to mean the sentence.

A model will offer "sincere." Sincere is a costume. Your voice is the words you would say across a table in 30 seconds. Speeches ask the other person to comfort you.

This loop is a skill with a stop. Read what a skill prompt actually is, then lock the rules below.

Draft an apology note that sounds like you

Own the fact first. Draft second. Warmth last, and only if it was yours.

1. Write the harm in your words, before the chat

One sentence, on paper or in Notes. "I missed the Friday file and you had to stall the client." Not "things got away from me."

If you cannot name the harm without a because-clause, you are not ready. The model can only polish a dodge. Treat the sentence as a dump, the same way you turn a messy note into a one-page brief: do not let the model infer the job.

2. Ban the speech before you draft

Paste this list and make it law:

I just wanted to reach out, sorry you feel, if anyone was offended, mistakes were made, that’s not who I am, I hope you can understand, we’re all human, I value this relationship, circling back to apologize.

Instruct Claude, ChatGPT, or Gemini: "If a draft uses a banned phrase, rewrite the whole note. Do not swap in a synonym. Do not add a lesson."

The same ban-first move that keeps captions from sounding like ads keeps an apology from sounding like PR.

3. Force four lines, not a letter

Every note, same bones:

• Harm — the sentence you wrote, in your words.
• Own — one fact you will not hedge. Date, file, words you said.
• Repair — one action with a time. Not "I’ll do better."
• Door — one line that lets them set the next step. No demand for forgiveness.

Tell the model: "60 to 120 words. Use my harm sentence as line one, edited only for grammar. Do not add context unless I pasted it as a fact. Do not thank them for their patience unless I wrote that."

"I’ll make it right" is fog, not a repair.

4. Generate two, pick one, do not blend

Ask for two drafts that share the same harm, own, and repair, but change only the door line. One door offers a call. One door offers silence unless they answer.

Pick one. Do not blend. A blended apology has two voices and zero nerve.

ChatGPT will add warmth. Cut it. Claude will lengthen the door. Shorten it. Gemini will be generic if you let it name the harm.

5. One human pass at table speed

Read the note as if they are sitting there. If you would wince, cut the line.

Check three traps:

• A because-clause that asks for sympathy.
• A repair that is a vibe ("be more present").
• A last line that requests closure from them.

Fix those in your own words. Write the first line yourself if the model polished it into a greeting. Then send, or wait until morning if you are still angry.

Reusable skill card

Trigger: You owe a written apology and the draft already sounds like a speech or a defense.

Input: One harm sentence in your words, one owned fact, one repair with a time or UNKNOWN, plus the banned-phrase list.

Output: Two notes, 60–120 words, in the harm / own / repair / door shape, your harm sentence as line one, zero banned phrases, no unsolicited context.

Stop: After you pick one draft, rewrite the first line if needed, and read it at table speed. No blending. No lesson. No ask for forgiveness.

Takeaways

• Name the harm yourself. The model cannot invent honesty.
• Ban speech phrases before you draft. Synonym swaps still sound like PR.
• Four parts: harm, own, repair, door. Sixty to 120 words.
• ChatGPT, Claude, and Gemini are clerks. You still have to mean it.