FAQ: If I Wrote It With an AI Assistant, Do I Have to Disclose It?

Sometimes — whether you have to disclose AI use depends on the rules governing that specific piece of writing: your course’s syllabus, the journal’s policy, your employer’s guidelines, or the platform’s terms, because no general law requires it and no venue’s permission excuses ignoring another’s rules.

So, do you have to disclose AI use? That “sometimes” is the accurate answer, and it’s more useful than either of the confident wrong ones. “You always have to disclose” isn’t true — nobody owes a disclosure statement on a birthday card or most workplace emails. “You never have to unless they ask” is more dangerous, because the venues that do require disclosure treat silence as the offense. Here’s the map — which venues require it, where the mechanical-versus-generative line sits, what non-disclosure actually costs, and how to write a disclosure that protects you.

Key takeaways

• There’s no blanket legal duty to disclose AI assistance in ordinary writing. Obligations come from venues: courses, journals, employers, platforms.
• Where disclosure is required, the concealment — not the AI use — is usually what converts allowed behavior into misconduct.
• Nearly every framework distinguishes mechanical help (spelling, grammar) from generative help (drafting, rewriting). Disclosure rules are about the second.
• When the rules are silent, asking first beats guessing. A two-line email is a permission slip you write yourself.
• Good disclosures are specific: tool, task, and what remained yours. Vague ones imply more than happened.

Do you have to disclose AI use? Where it’s genuinely required

Academia. The most common regime in 2026 is conditional permission: AI tools allowed *with disclosure*. That “with” is load-bearing — the disclosure is the condition, so skipping it revokes the permission retroactively. The MLA-CCCC task force’s guidance pushed programs in exactly this direction: transparency over prohibition. What disclosure looks like varies from a footnote to a formal appendix; we’ve collected working formats in AI disclosure statements: when and how to tell your professor.

Scholarly publishing. This is the most codified space. Nature drew its line in January 2023: no AI tool can be a credited author, and use of such tools must be documented. The ICMJE recommendations require disclosure of AI-assisted technology in the writing process — typically in the acknowledgments or methods — while holding authors fully responsible for everything the tool touched. Most major publishers have since converged on the same two-part rule: disclose the assistance, own the output.

Journalism and media. Newsrooms have moved to explicit AI policies, generally requiring internal disclosure and often audience-facing labels for AI-generated content. The interesting wrinkle is bylines — a human name on machine prose is treated as the serious offense. We’ve tracked the details in how journalism is handling AI standards, disclosure, and bylines.

Work-for-hire. Freelance contracts increasingly carry AI clauses — some banning generative tools outright, some requiring disclosure. A client relationship also runs on expectations that behave like rules: a client paying human rates who discovers undisclosed machine drafting rarely parses the distinction between “broke the contract” and “broke my trust.”

**What’s *not* on this list:** your personal blog, your novel (until your publisher’s contract says otherwise), your emails, your cover letter — venues where no governing rule exists. There, disclosure is a judgment call about your relationships, not an obligation.

The line almost everyone actually cares about

Every framework above, when you read it closely, is regulating the same thing: who generated the prose. None of them require you to announce spell-check. The ICMJE explicitly scopes its rule to AI-assisted *writing* technology; journal policies exempt grammar tools; syllabi that mention Grammarly nearly always allow it.

So the practical line runs through the middle of modern writing tools, not between them:

• Mechanical (generally no disclosure expected): spelling and grammar correction, citation formatting, find-and-replace, readability statistics.
• Generative (this is what the rules are about): drafting paragraphs from a prompt, “rewrite this to be clearer,” tone transformation, summarization you paste in, substantial AI paraphrase.

The same app can do both, which is where honest people get confused. The question isn’t “did I use an AI product?” It’s “did a model produce sentences that survived into my final draft?” If yes, you’re in disclosure territory wherever disclosure rules exist.

What non-disclosure actually costs

The consistent pattern across venues: consequences attach to the concealment at least as much as the use.

A student whose syllabus permits disclosed AI use, and who uses it silently, has converted permitted behavior into an integrity case. A researcher whose undisclosed AI drafting surfaces post-publication is looking at a correction or retraction — journals have issued both. A freelancer’s quiet automation becomes a contract dispute. In each case the AI use itself might have been completely acceptable *if declared*. The cover-up-versus-crime cliché holds remarkably well here, with one addition: in several of these venues, the cover-up *is* the only crime.

It’s worth saying plainly why venues weight it this way. A disclosure rule is a trust mechanism — it lets the evaluator apply the right standard. Concealment doesn’t just break the rule; it defeats the mechanism, and institutions punish attacks on their mechanisms hard.

How to disclose well (and when to ask instead)

When the rules are silent and stakes exist, ask first. “Does your policy permit using ChatGPT for outlining if the writing is my own? Happy to include a note either way.” Two lines, written record, done.

When you do disclose, precision protects you. A good disclosure names three things:

1. The tool — “ChatGPT,” not “AI technologies.”
2. The task — “brainstormed counterarguments and critiqued my second draft.”
3. The boundary — “all final text is my own; I verified every claim and citation.”

Compare that to “AI was used in preparing this work” — a vague formula that implies the model may have written everything, invites the maximum suspicion, and helps nobody. Specific disclosures read like professionalism; vague ones read like confession.

One last connection to the detection world we usually write about: disclosure and detection are complements, not substitutes. A disclosure doesn’t make your prose read as human, and a clean detector score doesn’t satisfy a disclosure rule. If your venue permits assisted writing and you want the result to carry your own voice, check how your draft reads to a detector and edit from there — openly. There’s more of our writing on AI policy and ethics, and our plans if checking becomes part of your regular workflow. The rule of thumb that survives every venue: when a governing document exists, read it; when it doesn’t, the person you’d least like to surprise is the person to ask.

Frequently Asked Questions

Do you have to disclose AI use in your writing? It depends entirely on who governs the writing. There’s no general law requiring individuals to disclose AI assistance in ordinary prose, but specific venues impose real requirements: many course syllabi require disclosure as a condition of permitted use, major journals and publishers require it in manuscripts, some employers and clients require it contractually, and some platforms require labeling. The question is never “must I disclose, in general?” — it’s “what do the rules of this specific venue say?”

Do I have to tell my professor I used AI? If the syllabus or assignment says disclosure is required, yes — and skipping a required disclosure can itself be the violation, even when the AI use was allowed. If the policy bans AI, disclosure isn’t the issue; the use is. If the policy is silent, you’re in the gray zone where asking first is the only safe move: a two-line email creates a record and converts a gamble into a permission. Most integrity cases in this area aren’t about AI use itself; they’re about concealment where openness was expected.

Does using Grammarly or spell-check count as AI use I have to disclose? Almost never. Disclosure norms nearly everywhere distinguish mechanical assistance — spelling, grammar, citation formatting — from generative assistance that produces or rewrites prose. Journals that mandate AI disclosure typically exempt grammar tools explicitly. The line to watch is inside the tools themselves: accepting a red-underline correction is mechanical, while clicking “rewrite this paragraph” is generative. If a tool drafted sentences you kept, that’s the territory disclosure rules are about.

What happens if I don’t disclose and it comes out later? The consequences usually attach to the concealment as much as the use. In academia, undisclosed AI use where disclosure was required is treated as an integrity violation, with the standard ladder of outcomes. In publishing, journals have issued corrections and retractions over undisclosed AI writing. For freelancers, it’s a contract breach and often a lost client. The pattern across venues is consistent: permitted-but-undisclosed converts into misconduct retroactively, which is a bad trade for a sentence you could have written.

How should I word an AI disclosure? Be specific and brief: name the tool, what it did, and what remained yours. For coursework: “I used ChatGPT to brainstorm the outline and critique a draft; all text is my own writing.” For a manuscript: “The author used [tool] to improve readability; all content was reviewed and verified by the author, who takes full responsibility.” Vague disclosures (“AI was used in preparing this work”) protect you less because they imply more than may have happened. Precision is both more honest and safer.