Do you have to disclose that you used AI to write something?

Start with who is relying on the writing

The question people ask is 'do I have to disclose'. The more useful question is 'who is relying on this, and what are they relying on it for'. Disclosure norms exist where the reader is making a decision about you or about the content's provenance: grading your work, paying you for original writing, publishing it under a byline, or accepting it as evidence.

That framing resolves most cases quickly. A tutor grading an essay is assessing your reasoning, so how the words were produced is material. A colleague reading a status update wants the information, so it is not. Between those poles, err toward saying something: an unnecessary disclosure costs you a sentence, while a missing one can cost you the relationship.

Education

Assume disclosure is required unless your institution says otherwise, and assume the policy is set at course level rather than nationally. Many institutions now distinguish between using AI to brainstorm, to draft and to polish, and treat those differently, which means a blanket personal policy will be wrong somewhere.

Two things are worth knowing. First, editing does not launder the requirement: if a policy covers AI-assisted drafting, running the text through an editor afterwards does not put you outside it. Second, AI-detection tools used to enforce these policies are unreliable and have been shown to misclassify writing by non-native English speakers. If you are accused on the strength of a detector score, the reliability of that score is a legitimate thing to question.

Work and client projects

Check whether your employer has a policy before assuming the absence of one means permission. Where a policy exists it usually turns on confidentiality rather than authorship: the concern is pasting customer data, source code or unreleased material into an external service, and that concern is well-founded regardless of disclosure.

For client work, the contract governs. Agreements for written deliverables increasingly include originality or AI-use clauses, and some clients require notice or prohibit AI-assisted drafting outright. If your contract is silent and the deliverable is the writing itself, raising it proactively is the safer move.

Publishing and platforms

Publishers vary widely and set their own terms. Many journals and outlets require authors to declare AI assistance, and a common position across scholarly publishing is that AI systems cannot be listed as authors because they cannot take responsibility for the work. Trade and consumer outlets differ enormously; the only reliable approach is to read the specific submission guidelines.

Consumer platforms have converged on labelling synthetic media, driven partly by regulation. Those regimes have focused on audio, images and video rather than ordinary written text, so a blog post is generally not covered the way a synthetic video would be. Do not read that as a general exemption: the direction of travel is toward more disclosure, not less.

How to disclose without making it awkward

A single accurate sentence does the job. Say what the tool did and what you did: 'A first draft was produced with an AI assistant; I researched, edited and verified the content and I am responsible for it.' That is more useful to a reader than a vague admission, and it is honest about where the judgement came from.

Two things not to do. Do not overstate your use to seem transparent, and do not understate it to seem more original: both are misrepresentations of the same fact. And whatever you disclose, verify the content first. Responsibility for accuracy sits with the person who publishes, which is you.

Frequently asked