It suggests.You decide.

Why general chatbots struggle with fiction

Paste a chapter into a general chatbot and ask for help, and you’ll get advice for an average book. It doesn’t know that your narrator lies, that the twins swapped places in chapter four, or that you never use the word “suddenly”. It forgets the conversation when the window closes. And whatever it writes comes out in the same polished, faintly familiar voice that readers have learned to spot.

An AI writing assistant for fiction needs three things a chatbot lacks: memory of the whole story, respect for your style, and a way of working that leaves the author in charge. The point isn’t to write the book for you. It’s to be the reader who has memorised your manuscript and can be asked anything.

There’s also the question of trust. A fiction writer needs to know what the AI changed, what it cost, and what readers will be told. An assistant that hides any of those is one you end up second-guessing, which defeats the point of having help.

It reads before it answers

Ask the assistant a question and it reads your chapters, searches the project and walks the story bible first. That’s why it can answer “has Mara ever been to the coast?” or “which chapters is the captain in?” with the passages that say so. Hand it a highlighted line, a scene or a whole chapter and its suggestions are grounded in what the book has already established.

Selection tools handle the small jobs: tighten, expand, fix the rhythm, or “rewrite it so that…” in your own words. Other words offers alternatives for a single word, similar, stronger, softer, plainer or more vivid. For a whole chapter, ✦ Draft a chapter works from its plan, or without one, and arrives as a proposal like everything else. The Draft brief shows exactly what the AI will be told before it starts.

Where an assistant earns its keep

Ask novelists who use AI well and few of them say it writes their prose. They use it as a reader and a sounding board. In planning, the Co-writer offers three directions for a field, a whole step or everything still open, and you mix them line by line, keep your own, or ask for more. Anchored decisions are never touched, and History can undo any card later.

In the draft, the most useful work is often checking rather than writing. Chapter readings record what happens and the setups you plant. ✦ Suggest a fix proposes an edit for any continuity issue. ✦ Has it paid off? checks whether a promise was kept, is still being carried, or was dropped. And when you’re stuck, ✦ Chapter ideas offers three takes on what happens next, with exact undo.

Reminders bridge the two: drop “bring it back here” on an arc, and ✦ Weave writes the pickup into a chapter. Each of these is a narrow job with a clear result you can judge, which is where an assistant is most trustworthy.

Your rules on every line it writes

Writing rules come in four strengths: must, should, avoid and never. Set your tense, point of view, spelling, the words you never use, the way dialogue is tagged. The rules are held on every line the AI writes and checked in chapter readings, so a draft that slips is caught rather than quietly accepted.

One click adds rules against stock AI phrasing, the tells that make generated prose feel generated. Voice samples, passages of your own writing, give drafting a voice to match, and chapters that read unlike your samples are flagged as voice drift. Anchors work the same way in planning: lock a decision and the AI never redrafts it.

• Must, should, avoid and never, held on every AI line
• Rules against AI tells in one click
• Voice samples from your own writing for drafting to match
• Voice drift flagged when a chapter reads unlike your samples

You see the cost, and you choose the model

Every ✦ action shows what it costs before it runs, so there’s no surprise at the end of the month. Choose your models by purpose: economy, balanced or high quality, a single company’s lineup, or a custom choice. A project can have its own models without changing anyone else’s.

AI work runs in the background, with progress and a Stop button in the bell, so you can keep writing while a chapter is read. Results apply automatically by default; turn off auto-apply and background changes wait in a review queue for you instead. Spending controls let you cap or pause spending at any time.

Honest about what the AI wrote

Words the AI writes into your manuscript are marked, and the mark follows them through every edit, copy, merge and branch. Large pastes are marked too, since a chapter written by a model somewhere else and pasted in would otherwise look like your own typing. Anything you type yourself carries no mark, even inside a passage the AI wrote.

If you publish on Authrot, readers see how much of each chapter was AI-written. That label is always on. It isn’t a judgement: plenty of writers use the assistant for planning and checks and write every published word themselves, and the label says so. It simply means readers and writers are working from the same facts.

Before you publish, the Publish page shows you the shares as the story stands, and why a story falls short of a stronger label if it does. You’re never surprised by what readers will see, and you can decide, chapter by chapter, how much help you want.