Who is responsible for AI-edited text?

Who is responsible for AI-edited text?

A language model can rewrite your methods section in 4 seconds. It cannot sign the paper. Someone has to, and that someone is you.

The committees have already answered the liability question

COPE and ICMJE agree that an AI tool cannot be an author, because it cannot answer for the work. You stay fully responsible for everything in your manuscript, including the parts a model produced, and liable for any breach of publication ethics. What remains open is how much you have to declare.

COPE’s position statement on authorship and AI tools is blunt. AI tools “cannot be listed as an author of a paper” because they “cannot meet the requirements for authorship as they cannot take responsibility for the submitted work.” COPE then draws the consequence: “Authors are fully responsible for the content of their manuscript, even those parts produced by an AI tool, and are thus liable for any breach of publication ethics.”¹

ICMJE says the same in its own words. Chatbots “should not be listed as authors because they cannot be responsible for the accuracy, integrity, and originality of the work,” and “humans are responsible for any submitted material that included the use of AI-assisted technologies.”²

So liability is settled. Disclosure is the part still moving.

What the big 4 publishers require you to declare

Elsevier and Wiley draw the line in about the same place: a spelling or grammar check stays below it. Springer Nature and Taylor & Francis ask for a declaration whatever the tool did. Having a model rewrite your discussion or restructure your argument sits above the line everywhere, and Taylor & Francis wants the tool named with its version number.

Publisher	What triggers a declaration	Where it goes
Elsevier	Generative AI used to improve language, clarity and readability. “Basic checks of grammar, spelling and punctuation do not need a declaration statement.”3	A declaration in the manuscript, published with the article
Springer Nature	Graded by how deep the tool reaches. Polishing language sits in the low-risk band, with a declaration required; “extensive copy editing or writing support” moves up a band; naming an AI system as an author is barred outright4	In the manuscript, usually in the introduction or the acknowledgements
Taylor & Francis	Any use, declared with “the full name of the tool used (with version number), how it was used and the reason for use”5	A statement of AI usage disclosure in articles; the preface or introduction in books
Wiley	Substantial editing, development or translation. “AI Tools used solely for spelling, grammar, and general editing are not included in these disclosure requirements.”6	At submission

The threshold is roughly the same everywhere. Fixing a typo sits below it. Having a model rewrite your discussion, restructure an argument or translate the paper sits above it. Most authors assume they are in the first group. A good number are in the second.

The line is easier to cross in a second language

If you write in English as an additional language, you push these tools harder. You are asking the model to make a paragraph sound like a journal, which is a long way past a comma check.

Ariane Peveto, reviewing AI copyediting tools, concluded that “in their current forms, AI copyediting tools are not a fine-tooth comb and should not be treated as such.” Her warning was aimed at native speakers: “If you are a creative writer and a native speaker of English, these tools are more likely to damage rather than polish your prose.”⁷

For a second-language author the risk runs the other way. The output usually sounds better than what you wrote. You have less ability to hear where it drifted, more of the paper now belongs to the tool, and your disclosure statement quietly stopped being accurate.

What a human editor puts on the line

A copyeditor answers for every change they propose. If a term is wrong, if a sentence now claims more than your data support, that lands on them, and they will hear about it from you. Trust is the entire business. An algorithm has no reputation to lose and no stake in whether the paper works.

A copyeditor is also your first engaged reader, often the only person who reads the paper before a reviewer does. That reading catches what spelling tools never see: the sentence that is perfectly grammatical and still says the wrong thing. And if a screening tool later flags a passage, you want to be able to say who wrote it and why.

We take up the broader question, can AI replace human copyediting, in a separate piece. This one is narrower. Whoever helped, your name is on the byline.

A working rule

Keep a log while you write: tool, version, what you asked it to do, which section. Reconstructing that at submission is guesswork, and guesswork inside a disclosure statement creates a second problem on top of the first.

Check the journal, not only the publisher. Imprints under one publisher differ, and every policy above has been revised since 2023.

Then read every accepted change against your own data. Author accountability is the one point on which all 4 publishers and both committees have converged, and it is the only part of this that no update is going to soften.

If you would rather have a person answer for the language in your paper, request a quote.