Disclose AI use in your work the right amount
A workable line between labelling everything and hiding it, based on what readers actually care about.
Before you start
- Work you produced with AI assistance
What you will be able to do
- Apply one test that resolves most disclosure questions
- Match the disclosure to what the audience is relying on
- Handle the contested cases honestly rather than confidently
Nobody discloses spellcheck and everybody expects to be told when a photograph of an event is synthetic. The interesting cases are in between, and the rules are still forming.
One test handles most of them, and the remaining ones are genuinely contested rather than merely unclear.
The test: would they feel misled?
Not "did I use AI" but "were they relying on something that is not true".
Imagine telling the audience exactly how the work was made. If a reasonable person would feel they had been misled, disclose. If they would shrug, you do not need to.
Most people shrug at a drafted email and object to a synthetic testimonial from a customer who does not exist. The test tracks that intuition better than any rule about percentages.
Ask what the audience is relying on
Authorship, authenticity or accuracy — different reliance, different duty.
Sometimes people rely on it being your work: a personal essay, an academic submission, a portfolio piece. Sometimes on it being a record of something real: journalism, a photograph, a quote. Sometimes only on it being right: documentation, a recipe, a how-to.
The first two carry a real disclosure duty. The third mostly carries an accuracy duty — readers want it to work, not to know who typed it.
Make the disclosure proportionate and close
A line near the work beats a policy page nobody visits.
Match the prominence to the stakes. A note at the foot of an article is fine for a drafted piece; a synthetic image in a news context needs a label on the image itself.
Say what was actually done — "researched and drafted with AI, edited and fact-checked by the author" — rather than a bare "AI was used", which tells a reader nothing about what to trust.
- Agree the standard once for your team and apply it consistently. Ad hoc disclosure reads as an admission when it appears and as concealment when it does not.
Be honest about the contested cases
Some of these do not have an agreed answer yet.
Heavily edited AI drafts, AI-assisted code in an open-source contribution, synthetic voice on your own content — reasonable people currently disagree, and platform rules are changing quickly.
Where a specific rule exists — a publisher's policy, an academic code, a marketplace term — follow it, because it beats your judgement. Where none does, err toward disclosing: the cost of over-disclosing is mild, and the cost of being found out is not.
Ask whether they would feel misled. Disclose where the answer is yes, in proportion, and near the thing itself.
Common questions
Was this guide useful?
94% of readers found this useful
Read next
Handle personal data in AI tools without breaking the law
Most AI data problems are not exotic. They are a consumer account processing customer records, and they are avoidable with four de…
Spot and reduce hallucinations in your workflow
Fabrication is not random. It clusters in predictable places, which means you can check the high-risk output and stop reading ever…
Connect two apps with an AI step in the middle
The genuinely useful automations are not the clever ones. They are a trigger, one AI step that makes a small judgement, and a writ…
Write ad variants and test them properly
AI removes the cost of writing variants, which makes it very easy to run tests that cannot teach you anything. The discipline is i…
