Did you use AI? Publishers ask it. Contest judges ask it. Readers ask it in reviews. Writers, staring at a checkbox, are supposed to answer yes or no about a process that rarely fits either box. At YouWrite we hear this from users of our Refine tool every week. They used AI to reorder a chapter or stress-test a metaphor, then rewrote every line themselves, and now they cannot figure out what to declare.
That confusion is not the writer's failure. It is the failure of a debate that collapsed a spectrum into a binary and then started punishing people for landing in the middle.
What disclosure was supposed to do
Strip away the moralizing and disclosure serves three practical goals.
Reader trust: a person buying a memoir wants to know a human lived it. A person buying a technical guide wants to know a human vetted it.
Market fairness: contests, grants, and trad publishing contracts allocate scarce resources. If one entrant spent 400 hours and another spent 40 minutes, judges want to know.
Authorship integrity: copyright, moral rights, and the basic claim of "I wrote this" depend on some defensible account of who did the creative work.
None of those goals is served by a yes/no checkbox. All three depend on knowing what kind of AI use, at what stage, and with what human judgment layered on top.
Where the binary breaks
Consider two writers.
Writer A drafts a novel over two years. Late in revision, she pastes three chapters into a model and asks whether the pacing sags. It tells her chapter nine is thin. She agrees, cuts 6,000 words, writes 4,000 new ones by hand. Every sentence in the finished book is hers.
Writer B types "write a 90,000 word thriller about a Boston detective" into a chatbot, lightly edits the output, and uploads to Amazon.
Under most current disclosure regimes, both check the same box. Or, more perversely, Writer A checks the box honestly and gets flagged, while Writer B says nothing and sails through. The New York Times has been covering this as a genuine authorship crisis (see Alexandra Alter's reporting on AI-generated books flooding Amazon, and the Authors Guild's evolving position), and outlets like American Bazaar have framed the ambiguity as one of the defining publishing problems of the decade. The reporting is not alarmist. The category confusion is real.
The good-faith penalty
Here is the ugly part nobody wants to say out loud. Writers who try to disclose honestly, under a standard that does not exist yet, are routinely punished more than writers who stay silent.
A novelist who writes "I used ChatGPT to brainstorm character names" in her acknowledgments has, in current discourse, said something closer to "I used AI" than "I named my characters." Reviewers will quote that line. Prize committees will hesitate. Meanwhile, the writer down the street who ran her whole first draft through Sudowrite and said nothing faces no consequences at all.
This is what economists call a lemons market. When honest signaling is punished, honest signaling disappears, and the category becomes uninformative. We are watching that happen in real time.
The better question
Stop asking "did you use AI." Start asking: is the creative vision, voice, and judgment yours?
That question actually maps to the goals disclosure was supposed to serve. It distinguishes the writer who used a model as a sparring partner from the writer who used it as a ghostwriter. It gives readers something meaningful to trust or distrust. It gives contests a real basis for eligibility rules. And it forces the writer to describe process rather than tick a box.
It is also the question copyright law is quietly converging on. The US Copyright Office's 2023 guidance on works containing AI-generated material turned on "human authorship" and whether a human exercised "creative control," not on whether any AI tool touched the file. That framing is imperfect and still being litigated, but it is closer to the right question than any checkbox.
How to talk about your process without self-incriminating
The standard does not exist yet. You still have to publish, pitch, and enter contests today. Some practical guidance.
Describe verbs, not tools. "I used an AI assistant to interrogate my outline" tells a reader something. "I used ChatGPT" tells them nothing except a brand name that will trigger their priors. Name what the tool did, not what it was called.
Locate the use in the timeline. Research, outlining, drafting, revising, line editing, and proofreading are different stages with different weights. "I drafted by hand and used AI to catch continuity errors in revision" is a real sentence. "I used AI" is not.
Claim your judgment explicitly. If every decision about voice, structure, and meaning was yours, say so. This is not defensive. It is the actual thing readers want to know.
Do not disclose things you did not do. If you brainstormed with a model and used none of what it said, you did not "use AI to write your book." Writers over-disclose out of anxiety and then get punished for the confession.
Keep a process note for yourself. Not for publication. For the moment, eighteen months from now, when a publisher or platform asks and you need to answer precisely.
Where YouWrite falls short here
We should be honest. Our Refine tool sits squarely inside the ambiguity. A writer who runs a chapter through it for line-level suggestions is doing something meaningfully different from a writer who generates a chapter from scratch, but our product does not currently emit any kind of process log the writer could show a publisher. We have talked internally about a provenance export. We have not built it. Until we do, the burden of describing process falls entirely on the writer, which is exactly the problem this piece is about.
What competitors are getting right and wrong
Sudowrite leans into generation and treats disclosure as the writer's problem, which is at least honest about what the tool does. NovelAI takes a similar posture. Grammarly and ProWritingAid have started labeling their generative features separately from their correction features, a small step toward the verb-not-tool framing. None of them, including us, has shipped a real provenance standard.
The writers who will come through this decade with their reputations intact are not the ones who avoided AI entirely. They are the ones who can describe, in specific verbs and specific stages, what a human did and what a machine did. Start writing that description now, before someone hands you a checkbox and calls it a standard.