Consent and Disclosure: Telling Your Co-Authors You Used AI

Telling your co-authors you used AI isn’t a courtesy — it’s consent: their names share responsibility for every word, so a tool that touches shared text is a decision the whole byline owns, made before the drafting, not confessed after it. That single reframe — from etiquette to consent — answers most of the practical questions, and the rest is scripts and timing.

Key takeaways

• Co-authorship pools accountability: collaborators answer for your AI decisions whether or not they knew, which is what makes silent use a consent violation rather than a style choice.
• Research publishing has made the duty explicit — ICMJE and COPE bar AI authorship and require disclosure, with humans owning every AI-touched claim — but the logic binds informal collaborations too.
• Timing converts the conversation: raised early it’s a preferences discussion; raised late it’s a confession with rework attached.
• An objecting co-author controls the shared work; your options are narrowing the AI’s role or keeping it to solo projects — not private defection.
• The external disclosure (methods note, acknowledgment) needs co-author sign-off too: it speaks for everyone whose name is on the paper.

Why this is consent, not courtesy

Start with what a shared byline actually does. When two names sit on a paper, a report, or a book chapter, readers — and editors, and integrity panels — treat every claim as jointly owned. There is no asterisk system for “section 3 was Priya’s problem.” If AI-generated text in your sections contains a fabricated citation, a subtly wrong claim, or prose that later trips a journal’s screening, the consequences arrive addressed to *all* authors.

That’s the structural fact silent AI use ignores. Your collaborator has staked their reputation on a process they believe they understand. Using a generative tool without telling them changes the risk profile of their signature without their knowledge — which is why the right frame is consent. You wouldn’t add a third author unilaterally, or submit to a journal without asking. Tool provenance that affects accountability belongs on the same list.

Research publishing has already formalized this. The ICMJE recommendations and COPE’s position statement converge on three rules: AI tools cannot be authors (they can’t be accountable), AI use must be disclosed to the venue, and the human authors take responsibility for all content, including whatever the model contributed. Notice that the disclosure duty runs to the *journal* — but you cannot honestly sign a joint disclosure your co-authors have never seen. External disclosure presupposes internal disclosure.

Outside academia, few contracts spell it out, but the logic is identical anywhere names are shared: co-founders on a whitepaper, consultants on a joint report, friends writing a book. Shared byline, shared risk, shared decision.

Timing is most of the ethics

The same information lands completely differently depending on when it arrives.

Raised at the start, AI use is a preferences conversation: “I normally use Claude to tighten my drafts — are we all comfortable with that here, and does this venue have a policy?” Nobody has sunk work; nobody feels deceived; the team can set rules in five minutes. Objections at this stage cost almost nothing to accommodate.

Raised after drafting, it’s a confession: the text exists, your collaborator has perhaps edited it believing it was yours, and their options have collapsed to acquiescence or conflict. Worse, if a venue’s policy forbids what you already did, the team now faces rework or an uncomfortable submission.

Discovered rather than raised — a co-author noticing the prose doesn’t sound like you, or a reviewer flagging it — is the disaster case. The material issue (AI use, usually survivable) is now fused with a trust issue (concealment, often not). Students hit the same dynamic in team assignments, which we’ve covered in group project AI etiquette for students; the stakes just grow with the byline.

So the operational rule is boring and absolute: *the AI conversation happens before the AI use.* Put it next to the authorship-order conversation, which every experienced collaborator already knows to have early.

A working agreement in four questions

You don’t need a policy document; you need four answers, noted in the project channel:

1. Which uses are fine without further discussion? Typical consensus: grammar and spell checking, reference formatting, AI critique of finished drafts. Most teams wave these through.
2. Which uses need a heads-up? Drafting assistance, substantial rewriting, summarizing sources. The middle band where preferences genuinely differ.
3. Which uses are off the table? Common candidates: generating analysis or results text, literature summaries nobody verifies, anything touching unpublished data in consumer tools (confidentiality lives here too — pasting a co-author’s unpublished work into a chatbot is its own consent problem).
4. What do we disclose externally, and in what words? Draft the methods/acknowledgment note together, per the venue’s policy, and get explicit sign-off — that note speaks for every author. Graduate students coordinating with advisors face a version of this we’ve detailed in disclosing AI assistance in a thesis or dissertation.

Two scripts to make the awkward parts easy. Opening the topic: *”Before we divide sections — I sometimes use AI to tighten drafts. Want to agree on what we’re each comfortable with and check the journal’s policy?”* Coming clean late (because the second-best time is now): *”I should have raised this earlier — I used [tool] to draft parts of section 2. Here’s exactly what it did. Happy to rework anything, and we should align our disclosure before submission.”*

One practical addition: teams increasingly sanity-check their own drafts the way editors will. It’s worth a shared pass to check how a draft reads to an AI detector before submission — not because scores are proof of anything, but because knowing what a screener will see beats being surprised by it; frequent collaborators can look at PaperBleach plans. There’s more on AI ethics and disclosure across the blog.

Frequently Asked Questions

Do I really have to tell co-authors I used AI on my sections?

Yes, and the reason is structural, not polite: co-authorship means every name on the work shares responsibility for all of it. Your collaborators are signing their reputations to text they believe reflects the team’s stated process. If a journal, client, or committee later questions the AI use, your co-authors will answer for a decision they never got to make. In research the duty is explicit — ICMJE-style guidelines require disclosing AI use, bar listing AI as an author, and make humans accountable for AI-touched text. Everywhere else the logic is identical even when no rule spells it out: shared byline, shared risk, so shared decision.

When is the right time to raise AI use with collaborators?

Before you use it, not after the draft lands. Disclosure after the fact converts a preference conversation into a confession, and it can force painful rework if a co-author objects or a venue forbids what you already did. The clean pattern is one agenda item at the project’s start: what tools are we comfortable with, for which tasks, and what will we disclose externally? Five minutes then prevents the worst version — a collaborator discovering mid-review that ‘their’ methods section was machine-drafted, in front of an editor asking questions.

What if my co-author strongly objects to any AI use?

Their objection controls the shared work, even if you think it’s overcautious. Consent is the operative concept: you can’t unilaterally impose a provenance on text that carries someone else’s name, any more than you could unilaterally add a co-author. Practical options short of abandoning the tool exist — agree to keep AI to non-substantive uses like grammar checking, confine it to your solo publications, or split responsibilities so your drafting stays outside the shared document. What you can’t do is comply in public and defect in private; if that surfaces later, the discovered concealment will damage the collaboration far more than the disagreement would have.

How do I disclose AI use in a paper’s acknowledgments or methods?

Follow your venue’s stated policy first — many journals now specify where and how. The common pattern: state the tool and version, what it was used for, and affirm that authors reviewed and take responsibility for the content. Something like: ‘The authors used [tool] to improve the readability of an early draft; all content was subsequently reviewed and edited by the authors, who take full responsibility for the final text.’ Never list an AI as an author — major editorial bodies including ICMJE and COPE bar it because tools can’t be accountable. And make sure every co-author has seen and approved the disclosure language itself before submission.

Does AI-checking or grammar software count as disclosable AI use?

Usually not — spell-checkers, classic grammar tools, and reference managers are standard writing infrastructure, and most policies explicitly exempt them. The line most venues and sensible teams draw is substance: tools that generate or substantially rewrite content are disclosable; tools that flag your own errors are not. The gray zone is heavy AI rewriting that changes structure and phrasing wholesale, which leans disclosable. When you’re unsure, the co-author version of the disclosure test works: if you’d feel awkward mentioning the use casually, that awkwardness is your answer — mention it. A one-line Slack message is cheap; discovered concealment is not.

The bottom line

A shared byline is a small partnership, and partners don’t get surprised by each other’s risk decisions. Raise AI at the start like you’d raise authorship order, agree on the three tiers of use, and draft the external disclosure together. The conversation costs five minutes when it’s consent — and a collaboration when it’s discovery.