You are not legally required to disclose AI use in most everyday writing, but you should disclose it whenever the client or reader is relying on the content as your own human judgment, expertise, or original work — and you must disclose whenever a contract, platform rule, or publisher policy requires it.
In practice, that means a ghostwriter producing a personal essay for a named author should not announce "written by AI" in the byline, but should tell the client privately; a freelancer delivering a research summary or legal explainer should say so plainly, because the reader is trusting the accuracy of the claims, not the prose style.
The mechanism behind this rule is trust transfer. When a reader opens an article, they are not just reading words — they are assigning credibility to a source. A byline that says "by Jane Smith" transfers Jane's reputation, her training, and her accountability onto the text.
If a machine wrote the draft, that transfer is happening without the reader's knowledge. The same logic applies to a client paying for "your expertise" rather than "a document." The question is not "did a tool help?"
— spellcheck and search engines have always helped — but "is the reader relying on a human judgment that didn't actually happen?" A useful test: if the client knew exactly which sentences came from a model and which came from you, would they still pay the same rate and feel the same way about the deliverable?
If the answer is no, disclose. If the answer is yes, disclosure is optional but still often wise.
Here is what disclosure looks like in practice. For a freelance proposal, a single line near the scope section works: "Drafting may use AI writing tools for research summarisation and first-pass structure; all factual claims are verified by me, and final copy is edited and approved by me before delivery."
For a published article where AI assisted but a human wrote the substance, a footer note is enough: "AI tools were used for background research and copyediting. All reporting, analysis, and conclusions are the author's." For a ghostwritten book where the client's name is on the cover, the disclosure belongs in the contract, not the book: a clause stating that AI-assisted drafting was used and that the client reviewed and approved all content.
A concrete example: a marketing consultant writing a client's LinkedIn posts might add to the contract, "AI tools may be used to generate draft variations; the client approves every post before publishing." That single sentence protects both sides and takes thirty seconds to write.
The limits are real. Some jurisdictions and sectors do have hard rules: financial advice, medical information, legal documents, and certain government communications often carry explicit disclosure or accuracy requirements that vary by country and by regulator, so check the specific rule that applies to your work rather than assuming a general norm covers you.
Some clients will explicitly ask you not to mention AI at all — for example, a brand that wants its audience to believe every word came from its founder. You can agree to that, but you should get it in writing, because if the arrangement later becomes public, the reputational damage lands on you as well as them.
There is also a practical cost: disclosure can invite scrutiny or lower perceived value, especially in markets where "AI-written" is treated as a quality warning rather than a production note. And no disclosure statement fixes a factual error — if the model invented a statistic and you shipped it, saying "AI-assisted" does not undo the harm.
Disclosure is about honesty, not about shifting responsibility. If you want a fuller picture of how AI tools handle the material you feed them, it is worth reading Is the best AI email writing assistant safe to use with confidential work emails? before you decide what to share with a tool in the first place.