AI news
AI model reviews
AI tools
How-to guides
Our staffAboutContact

Can AI write a book? What it can do, what it can't, and the law in 2026

AI can produce a book's worth of text, but a good novel still needs a human. Here's where AI helps, where it stalls, and what copyright and Amazon say.

Vintage typewriter on a dark wooden desk
Photo: Patrick Fore on Unsplash

Yes, AI can write a book, in the narrow sense that ChatGPT, Claude or a fiction app like Sudowrite will keep producing chapters for as long as you keep asking. Whether it can write a good book on its own is a different matter, and I don't think it can yet. The AI-assisted books that work are ones where a person plans the story, makes the choices and rewrites the prose, with the software drafting and doing grunt work in between.

There's a legal catch as well. In the US, text a machine produces from your prompts isn't protected by copyright unless a human shaped its expressive parts, so a novel written entirely by AI belongs, in a real sense, to nobody.

Can AI write a book you'd want to read?

Picture a first-time writer on a Sunday afternoon typing "write me a 300-page mystery set in Lisbon" into a chatbot. Chapter one arrives in seconds, and it's fine. By chapter nine, though, the detective's sister has a new name, a clue planted early on has vanished, and every character seems to sigh at the same moments. Nothing's broken there. That's what usually happens when a tool that predicts text a chunk at a time is asked to carry a whole plot.

So the useful question is narrower. Producing book-length text is easy now. A plot that pays off, with characters who stay themselves for 300 pages, still needs a human steering, as far as I can tell, and the steering turns out to be most of the work.

The machinery behind a machine-written chapter

Large language models write by predicting the next token (a word or a piece of one) from everything in front of them. They learned those patterns from enormous amounts of text, and that included a lot of books. Anthropic's $1.5 billion class action settlement with authors over pirated books used in training got final court approval on July 20, 2026, at roughly $3,000 per work. So when a model writes a passable chapter, it's drawing on the work of people who mostly weren't asked.

Two limits fall out of that design. The first is memory: a model only "sees" what fits in its context window, and even when a whole manuscript fits, details from 200 pages back tend to slip. The second is averageness. A system trained to pick likely words drifts towards likely choices; a novelist spends years learning to avoid exactly those.

There's research on that second point. In a 2024 study in Science Advances, Anil Doshi and Oliver Hauser had people write short stories with and without AI-generated ideas. Readers rated the AI-helped stories as more creative and better written, especially for writers who started out less creative. But the AI-helped stories were also more similar to each other, which is good news for one writer and less good for the bookshelf.

The fiction apps add structure on top of the same models. Sudowrite says it uses dozens of models, including Claude, several OpenAI models and an in-house fiction model called Muse, wrapped in its own processing. Novelcrafter takes a different route; you plug in your own AI key and it keeps a "Codex" of your characters and places alongside the manuscript. It's a bit like handing a forgetful co-writer a notebook before every session. Useful, but someone still has to write the notebook.

Books that already used AI

The best-known case is probably Rie Kudan's Tokyo Sympathy Tower, which won Japan's Akutagawa Prize in January 2024. In her acceptance speech Kudan said: "Probably about 5 percent of the whole text is written directly from the generative A.I." The novel is set in a society that leans on AI, and she used the chatbot's smooth, slippery phrasing where the story needed exactly that kind of language.

I find that ratio telling. The prize-winning book was roughly 95 percent hers, and the 5 percent was used on purpose.

On the legal side, Stephen Thaler spent years trying to register an image his AI system made with no human input, listing the AI as sole author. The DC Circuit ruled against him in 2025, and on March 2, 2026, the Supreme Court declined to hear the case. It was a picture rather than a novel, but the human-authorship rule applies to text in the same way.

The US Copyright Office spelled out where the line sits in Part 2 of its AI report, published on January 29, 2025. AI output can be protected only where a human determined enough of the expressive elements, through creative arrangement or changes, and "not the mere provision of prompts". Using AI to assist, or including some AI text inside a larger human work, doesn't block copyright.

Publishers and platforms have drawn similar lines. Amazon's KDP content guidelines require you to disclose AI-generated text, images or translations when you publish, and that label applies "even if you applied substantial edits afterwards". AI-assisted work (editing, error-checking, brainstorming ideas you then write up yourself) doesn't need disclosing. Going the other way, the Authors Guild runs a Human Authored certification that's free for members and $10 per title for everyone else.

AI's strengths and weak spots for novelists

Here's how I'd sum up the split, based on what the tools are built to do and where long-form writing tends to break.

TaskWhere AI does wellWhere it struggles
BrainstormingDozens of premises, names and twists in a minuteIdeas cluster around familiar tropes
OutliningTurning a premise into chapter beatsSetups and payoffs often don't connect
Drafting scenesFast first drafts of a single sceneFlat, samey voice across a whole book
ContinuitySummarising chapters you've already writtenRemembering details from far back in the draft
Editing and feedbackLine edits, pacing notes, rephrasing dialogueKnowing which of its notes actually matter
Blurbs and synopsesQuick drafts of back-cover copySounding like every other blurb in the genre

The upside is real. A stuck scene gets five options instead of a blank page, and the help is cheap. Looking at Sudowrite's plans, Hobby & Student costs $10 a month paid annually or $19 monthly for 225,000 credits, Professional is $22 or $29 for a million credits, and Max is $44 or $59 for two million rollover credits. Its beta-reader style Feedback tool only comes with Professional and Max. On Novelcrafter's price list, AI starts with the $8 Hobbyist plan, where you bring your own key; the $4 Scribe plan has no AI at all.

The downsides are just as real, and they don't go away with a bigger plan. Long-range continuity still needs a human checking it. Prose tends towards the same handful of moves. And readers in some genres have started asking whether a book was written by a person, which is exactly why certifications like Human Authored exist.

Practical steps for writing a book with AI

If you want to write a book using AI without ending up with something generic (or something you can't protect), this order of work makes the most sense to me:

  1. Write the premise, the ending and the main characters yourself before you open any tool.
  2. Keep a story bible with names, ages, places and rules, and paste the relevant parts in for every session, or use an app that does it for you.
  3. Ask for options at the scene level, never whole chapters, and pick or combine what you get.
  4. Rewrite every AI sentence you keep in your own voice; that's where your copyright, and your style, actually come from.
  5. Log which parts started as AI output, so a KDP disclosure or a copyright application is easy to answer honestly.
  6. Use AI as a reader once you have a draft: pacing notes, continuity checks and a blunt summary of what each chapter does.

For tools, general chatbots are fine for brainstorming and feedback, and our Claude vs ChatGPT comparison covers which one suits longer writing. Dedicated fiction apps make more sense once you're drafting seriously, and the AI writers roundup compares more options. When the manuscript's finished, the same tools can help with the packaging; there's a separate list of AI book cover generators, though cover art made by AI counts as AI-generated on KDP too.

Personally, I'd use AI most heavily at the two ends of the process (early brainstorming and late editing) and least in the middle, where the actual sentences get written.

Some of this is settled. Human authorship is required for US copyright, the Supreme Court left that rule alone, and Amazon wants AI-generated content disclosed. The Copyright Office has also released a pre-publication version of Part 3 of its report, on training AI with copyrighted works, with a final version still to come.

The rest is guesswork, so here's mine. Models will probably get better at long-range consistency, since that's mostly an engineering problem, and the "chapter nine" failure will likely fade. The sameness problem seems harder to fix, because it comes from the way these systems choose words in the first place. I'd also guess disclosure labels spread beyond Amazon, and that "written by a human" becomes something readers in a few genres actively look for.

Money is the open question. The Authors Guild said in July that no date for settlement payouts had been announced yet, and other lawsuits over training data are still running. For anyone writing a book now, though, my view's fairly clear: AI makes a capable assistant and a poor author, and the law, as it stands, agrees.

If your question is the gentler one, whether AI can help rather than replace you, our explainer on using AI to help write a book goes further into that.

FAQ

What is the best AI to write a book?

It depends on the stage you're at. ChatGPT and Claude are good for brainstorming and feedback. Sudowrite is built for fiction drafting, with plans from $10 a month paid annually. Novelcrafter suits planners who want character and world notes organised, and it costs $8 a month if you bring your own AI key.

How to write a book with AI?

Plan the premise, ending and main characters yourself. Then use AI for scene options and line edits, feeding it a story bible so details stay consistent. Rewrite what you keep in your own voice, and note which parts began as AI output for Amazon KDP's disclosure.

How to use AI to write a story?

Start with a premise and a clear ending, then ask the AI for a few different openings or plot beats and pick the one you like. Keep prompts specific about tone, length and point of view. For short stories, AI is most useful for brainstorming and editing; the actual sentences read better when you write them.

How to use AI to write a novel?

Treat AI as a co-writer for small pieces, not the whole book. Keep a running list of characters and plot facts and paste it into each session, because models lose track over long manuscripts. Work scene by scene, revise heavily, and use AI again at the end for continuity checks and pacing notes before an editor sees it.

Keep reading