Disclose where AI involvement was material — where knowing would change how a reasonable person reads, trusts or acts on the work.
Do now: Write a one-sentence disclosure for one real piece of work, saying what AI did and who is responsible.
Disclosure tells people what happened. It does not transfer your responsibility to a tool.Disclosure norms are still forming, and the people who get this right early build the trust that others will later have to buy back. It also protects you: a stated position is far easier to defend than a discovered one.
The materiality test
The question is not whether AI touched the work. It is whether knowing that would change how a reasonable person reads it, trusts it, or acts on it. Spell-checking is not material. A photorealistic image of an event that did not happen certainly is.
Material involvement generally includes: synthetic media that could be mistaken for a recording of reality; a synthetic voice or likeness; substantive content in academic, journalistic or professional work where authorship carries weight; and anything where a person is choosing to rely on your expertise specifically.
Non-material involvement generally includes: grammar and spelling, formatting, translation assistance you have reviewed, research leads you then verified yourself, and idea generation where you did the work. Labelling all of these floods the signal, and a label that appears everywhere stops meaning anything anywhere.
Disclose what changes the reading. Labelling everything is the same as labelling nothing.
Write it plainly and specifically
'This content was created with AI' tells a reader almost nothing. Was the data generated? The images? The conclusions? Was any of it checked? Vague labels can conceal more than they reveal, and readers have learned to discount them.
Better: name what AI did, what a human did, and who is accountable. 'Images illustrating historical scenes are AI-generated; all photographs are original. Figures verified against published statistics. Written and checked by [name].' Three sentences, and a reader knows exactly where they stand.
Place it where it will be seen at the point of judgement, not in a footer nobody reaches. Platform and institutional rules may set the format and the position; follow the stricter of their requirement and your own standard.
Institutions, and the shape of an honest policy
Where you study or work, the rules exist and they usually differ by task. Academic integrity policies commonly permit AI for some purposes and prohibit it for assessed work. Employers may permit drafting but not decisions. Learn the actual rule rather than assuming permission or assuming prohibition.
A good organisational disclosure policy states what must always be disclosed, what never needs to be, and who decides the middle. It also names the person accountable for the output regardless of what tool made it — which is the sentence that stops 'the AI produced it' from becoming an explanation.
Keep the process evidence. Drafts, prompts, sources and revision history are what let you demonstrate the work, and they matter far more than any label when a question is asked seriously. Detection tools are unreliable in both directions; showing your process is the answer to a good-faith challenge.
The tool is never the author. Someone is accountable, and disclosure names them rather than replacing them.