If a model wrote part of an article, the reader should be able to find that out without hunting for it. Everything else is detail, but the detail is where this gets decided.
Three things worth stating plainly
Who wrote it. Not “this newsroom uses AI tools”, which is true of everyone and tells a reader nothing. Which model, on which piece.
What a person did about it. “Reviewed by” is a meaningful claim and an empty one depending on whether anybody checked the sources. If a human read it and signed off, say who. If nobody did, say that instead.
What it was based on. A list of sources with links is worth more than a paragraph of methodology. It is also checkable, which is the point.
What not to do
Do not put a model’s name in the byline position where a person’s name goes. A byline is a claim about responsibility, and a model cannot hold any.
Do not bury the disclosure in a site-wide policy page. A reader deciding whether to trust one article is not going to go looking.
Do not use “AI-assisted” as a shield. It covers everything from a spellcheck to a first draft, and a phrase that covers everything discloses nothing.
The uncomfortable part
Some of this costs you. An honest attribution line on a piece a model drafted will lose some readers who would otherwise have read it. That is not a bug in the disclosure; it is the disclosure working. Readers are entitled to that preference and it is not the newsroom’s business to route around it.