Free to read · no account needed

9 min · safety

When and how to disclose AI assistance

Goal: Disclose AI help in one calm ≤25-word line—neither more nor less than the truth.

Best for: Students, employees, indie publishers, and anyone shipping AI-shaped drafts.

Policies differ. Some classes ban AI; some require a footnote; some workplaces require internal tags. Default ethically: if a reader would reasonably want to know AI helped shape the words or analysis, disclose in proportion.

Disclosure is a trust habit, not a confession booth. Say what the AI did and what you checked—especially when the reader might think a human alone wrote every word.

Disclosure is not shame—it is calibration. Readers can trust you more when they know which parts a model drafted and which parts you verified.

Why beginners skip this

Skipping disclosure feels tidy—readers assume all human hands, and trust collapses later. Write the disclosure line first.

Step 1 — Quick decision

Use this table first; when unsure, disclose on the stricter row.

ContextUsually
Graded work with ban/limitsFollow syllabus; when allowed, disclose method
Client deliverablesAsk; many want a note in process docs
Public essays / newslettersShort footnote if AI drafted structure/prose
Private brainstorm only you seeOptional
Draft a one-sentence disclosure for [context].
Facts: AI helped with [outline / wording / translation]; I verified [facts]; final edits by me.
No need to paste my prompts. Keep under 30 words.

Step 2 — Example footnotes

Example

Filled prompt

I used ChatGPT to polish a client proposal (structure only; all numbers mine).
Draft a one-line disclosure suitable for email footer AND a longer note if client asks.
No overclaiming "AI wrote this."

What good output looks like

  • Opens as one calm sentence under ~25 words—no confession essay
  • Names AI help for outline and that you rewrote every paragraph
  • No ChatGPT/Claude brand parade or model-version dump
  • Tone stays matter-of-fact—not defensive, cute, or oversharing
  • Fits a blog footer without reading like a confession booth

Step 3 — Honesty > clever labels

“Inspired by AI” while the draft is mostly model prose is still misleading—say so plainly, or revise until the label matches reality.

Example

Filled prompt

Live artifact this week for: Know when transparency is required or wise—school, work, publishing—without oversharing pr
Name real files/people. No invented facts; label Unknowns; stop before send/delete.
Return working prompt + one pass/fail check I can mark in a minute.

What good output looks like

  • One sentence; drops cousin name and multi-tool brand spam
  • Keeps “not legal advice” as a separate boundary if the wiki needs it
  • Still says AI helped with wording and that final edits are yours
  • No private prompts, cousin jokes, or model-version dump
  • Fits an internal wiki footnote without looking like a confession

Save this card

Card: Disclose in proportion: one calm line for what AI did and what you checked. Keep this beside the tool; read before you start.

Before you start

Put the scrub checklist on your desk. For “Know when transparency is required or wise—school, work, pub,” decide placeholders before the paste window opens Guide: safety-disclose-ai-assistance Keep the live artifact visible the whole time. If you only have a blank box, stop and fetch the real job first

How to use this week

This week add a disclosure line to a real external draft: which parts used AI, what a human changed; no send without it.

Close the week by quoting aloud the refuse line you actually used once. If you cannot, rerun Practice now and date the pass/fail (Asia/Shanghai).

Under 25 words

One calm sentence: what AI did + what you checked. No tool-brand dump.

Match the label to reality

“Inspired by AI” while most sentences are model prose is still misleading. Either rewrite until the label fits, or say AI drafted and you edited. Never put keys or PII inside a disclosure footnote.

Example

Filled prompt

AI outlined a proposal; numbers all mine.
Footer one-liner + longer answer if client asks. No "AI wrote this." overclaim.
≤25 words; what AI did + what you checked—no tool brands.

What good output looks like

  • No raw PII/secrets in the paste for “Know when transparency is required or wise—school,”
  • Placeholders used; scrub checklist visible
  • Disclosure matches how much AI actually did
  • Pass/fail: desk card updated today

Red flag

Silent AI on high-stakes outbound; disclosure missing where readers expect it—add one calm line.

Pitfalls

PitfallFix
Ignoring syllabus bansRead policy first
Oversharing secrets in disclosureNo keys/PII in footnotes
Using disclosure to excuse fabricationNever; verify anyway
Example

Filled prompt

Temptation: skip the habit for safety-disclose-ai-assistance to save five minutes.
Job still: Know when transparency is required or
Smallest correct pass: Done visible · checkable artifact · Unknowns [?].
Boundaries: draft only; no silent send; no invented metrics.
Return: artifact + refuse-line said aloud.

What good output looks like

  • Names the shortcut refused (specific to safety-disclose-ai-assistance)
  • Still delivers checkable output for “When and how to disclose AI assistance”
  • Unknowns labeled; draft-only
  • Pass/fail: refuse-line dated (safety-disclose-ai-assistance #4)

Practice now

Policy link: if org policy exists, link it beside disclosure (Asia/Shanghai).

Audience variant: short vs long disclosure for two channels (Asia/Shanghai).

Where pasted: note which channel got the disclosure (Asia/Shanghai).

Channel fit: tailor disclosure length to email vs doc vs chat (Asia/Shanghai).

Paste-ready: disclosure line lives in a snippet tool, not only in memory (Asia/Shanghai).

Disclose drill: write the disclosure line you will actually paste; put it on Save (Asia/Shanghai).

Thin-spot drill: rewrite the Save card with real nouns from your work (repo, doc, inbox). Generic placeholders fail the drill.

Tonight add one AI-assistance disclosure line to a real external draft before send.

Close the loop

Write one line: next time I send an external draft with no disclosure, I refuse.

One sentence takeaway

Disclosure line: which parts used AI, what a human changed; no send without it.

Next

Copyright & voice · Study honesty · Eval outputs · Secrets and pasting

← Back to home · Learn · Tools · Compare · Start