The Authors Guild's 2023 survey found that roughly 13% of published authors reported using generative AI for brainstorming, outlining, and character concepts. That number is neither scandal nor vindication. The argument about whether writers use AI is over. The argument about what counts as using it has barely started. At YouWrite, where the Story tool sits somewhere in the middle of that spectrum, this fight lands in our inbox every week.
The spectrum is longer than the discourse admits
Most public arguments treat AI use as binary: you either wrote it or you didn't. The actual practice looks more like a dimmer switch with at least six positions.
- Idea sparking. You ask a model for twenty premises about a lighthouse keeper. You pick one. You write everything else.
- Outlining. You feed the model your premise and let it propose a structure, then rearrange it.
- Drafting. The model produces prose. You revise heavily, lightly, or not at all.
- Line editing. You wrote the sentences. The model tightens them.
- Paraphrasing. You wrote a passage, disliked it, and asked for three rewrites.
- Structural revision. The model reads your draft and suggests cuts, reorderings, or a new ending.
A writer at position one feels nothing like a writer at position three, and both bristle at being lumped with position five. The discourse collapses them because arguing about a category is easier than arguing about a gradient.
Why the discomfort is real, not performative
A novelist admits she used ChatGPT to name her villain. Readers accuse her of fraud. Something honest is happening on both sides. She used a tool. They noticed that naming a villain is not a mechanical task, it is a small act of authorship. Nobody is lying. The disagreement is about which acts count.
Publishing had this argument about ghostwriters in the 1920s, about editors like Maxwell Perkins reshaping Thomas Wolfe's manuscripts in the 1930s, and about developmental editors in the 1990s. Each time, the field absorbed the practice by naming it. Acknowledged ghostwriting is legal and common. Perkins is celebrated, not shamed. Developmental editing has a rate sheet. A new form of assistance appears, hides for a while, then gets normalized through disclosure norms rather than prohibition.
AI is in the hiding phase.
What the honest writers are actually doing
Talk to working authors off the record and you find personal rules no publisher has asked them to articulate. One thriller writer I know uses AI to generate police procedure details but refuses to let it write dialogue. A literary novelist runs finished chapters through Claude for line-level tightening but will not show it a blank page. A memoirist uses AI to draft transitional scenes she finds tedious, then rewrites every sentence.
None of these writers thinks the others are cheating. Each has drawn a line based on which part of the work they consider constitutive of their authorship. The line is idiosyncratic, defensible, and mostly private.
Private lines become public liabilities the moment a reader, agent, or awards jury asks. Writers who have not thought their position through in advance tend to overclaim ("I didn't use AI at all") or overconfess ("I guess the book isn't really mine"). Both responses corrode trust.
Write your threshold down before you need it
Before your next project, answer these on paper:
- Which stages of my process am I willing to delegate, and which define the work as mine?
- What would I be comfortable saying, unprompted, in an interview about how this book was made?
- If a reader saw my full prompt history, would anything embarrass me? Why that thing specifically?
The third question is the useful one. Embarrassment is a signal. It usually points to a moment where you suspected the tool was doing work you wanted credit for. That is your line, discovered by discomfort rather than argued into existence.
Where YouWrite fits, and where it doesn't
Our Story feature generates long-form fiction from a prompt. That places it in the drafting column, the most contested position on the spectrum. For some writers that is a legitimate use: someone drafting fan fiction for fun, a screenwriter blocking out a spec, a novelist who wants a burnable first draft to react against. For others it is a bad fit. If your identity as a writer is anchored in the sentence, our drafting output will feel like a threat rather than a tool. That is a fair reaction, and we would rather you use us for outlining or revision than force a fit.
We fall short in a specific way: we do not currently offer a built-in disclosure log that a writer could hand to an editor showing which passages were AI-drafted versus human-drafted. That would be useful. It is on the roadmap. The honest version of this post has to include what we haven't built yet.
The competitive picture
Sudowrite has staked out the "co-writer for novelists" position and does it well, particularly for genre fiction workflows. NovelCrafter emphasizes writer control and structured worldbuilding, which appeals to plotters. Jasper and Copy.ai are optimized for marketing prose and feel wrong for narrative work. ChatGPT and Claude, used directly, are the most flexible and the cheapest, but they demand that the writer bring the workflow. None of these tools, ours included, has solved the disclosure problem. That is the industry's shared gap.
The debate we should be having
Prohibition will not work. 13% is a floor, not a ceiling, and the number climbs every quarter the Authors Guild runs the survey. Unlimited use without disclosure will not work either, because readers care and will keep caring. What remains is the boring, useful path: writers articulating personal thresholds, publishers requiring disclosure at contract stage, awards bodies deciding category by category what they will accept.
None of that requires consensus on where the line sits. It only requires that each writer know where their own line is, and be willing to say so out loud.