Using Claude to Draft a Novel: What Anthropic's Pages Say

NarratedIndie drafts its own pages using Claude; read our AI use policy with that in mind. Claude is a general-purpose assistant, so it suits novelists who want help with planning, feedback and drafting in a chat. The Anthropic pages we read describe no story-bible or manuscript view, so plan on keeping the book itself in your own files.

Plans

Claude lists Free, Pro, Max, Ultra, Team and Enterprise plans on its pricing page, as read on 2026-11-08. The page says prices and plans are "subject to change at Anthropic's discretion".

A solo writer would likely look at Free, Pro and Max (our reading). Team and Enterprise are listed for organizations, and the Team row is described as being for 2 to 150 users.

What a year costs

Here is our arithmetic on the listed prices, an estimate that assumes the prices stay as listed and you pay for twelve months. Pro billed annually is $200; Pro billed monthly is $20 x 12 = $240, so annual billing saves $40. Max is $100 x 12 = $1,200 and Ultra is $200 x 12 = $2,400. Free is $0. The page does not say what a "usage window" amounts to in words or chapters, so these sums say what you pay, not how much drafting you get. The annual option is paid up front, which matters if you want to stop after one book.

What a usage limit depends on

Claude's paid plans have a five-hour session limit and a weekly limit, and Anthropic's help page, read 2026-11-08, gives no word or chapter figure for either. Verified Ultra accounts have no five-hour limit. Anthropic says it may adjust limits at its discretion, and that Claude's web, desktop and Claude Code use all count against the same allowance.

The factors it lists map closely onto novel work:

• Long messages and attachments use more. So does a long running conversation, and the model and effort level you pick.
• Project files are cached. Reused project content counts less against your limits, but the cache expires after inactivity, so your first message after a long break counts that content in full.
• Settings > Usage shows where you stand on paid plans, with progress bars for the session and the week.

The practical test is to do one normal working session on the plan you are considering and read that page afterwards. We have not measured how many chapters a limit covers.

Which plan a novelist's needs point to

These differences come from Anthropic's help pages and pricing page as read on 2026-11-08. This table compares features, and the price table above has the costs.

Projects as a place for story notes

Projects are described by Anthropic as "self-contained workspaces with their own chat histories and knowledge bases". You can upload documents and set custom instructions, so a project can hold your outline, character notes and style guide as reference for each chat.

The help page, which displays "Updated over a week ago" rather than a date, adds these points as read on 2026-11-08:

• Free users can create a maximum of five projects.
• Enhanced knowledge handling (retrieval) needs a paid plan: Pro, Max, Ultra, Team or Enterprise. When a project's content nears the context limit, the page says this mode can expand capacity "up to 10x".
• Sharing a project with other people is Team and Enterprise only, with view-only and edit permission levels. A solo Pro subscriber cannot share a project with an editor or co-author through this feature.
• The page describes a beta new version of projects, rolling out in stages starting with Claude Code. In it, a project is organized around one conversation, and access is limited to select Pro, Max and Ultra subscribers using Claude Code, plus some users whose existing projects were moved. The page says moved projects keep their instructions, files and chat history.

That last point matters if you build a workflow around Projects: the feature is changing, so re-read the page before you commit a long book to it. We have not measured how reliably a project applies your notes across a long manuscript. See AI for worldbuilding and series bibles for a method that keeps canon in your own files.

Anthropic's help pages suggest a simple way to set one up:

1. Upload working documents to the project at the start. The upload page, dated July 23, 2026, accepts DOCX, TXT, RTF, ODT, EPUB and PDF, at up to 30MB per project file, with text extraction only for most files.
2. Give files clear, descriptive names. Anthropic's retrieval page says this helps Claude find the right material.
3. Keep project instructions short, and put task-specific instructions in the chat, as its limits page advises.
4. Remove files you no longer use, which that page also advises.

Memory works per project: each project has its own, and Claude uses it alongside your memory from chats outside projects. Turning off "Use account memory" on the project's page keeps the project's memory separate from your account memory, which is worth doing if you keep separate pen names or books apart.

Will a whole manuscript fit?

A 90,000-word manuscript probably fits inside the context window on paid plans, on our rough estimate, but fitting is not the same as being read in full on every reply. Anthropic's limits page says that on paid plans the newest models support up to a 1M-token window and others 500K or 200K.

Here is our arithmetic, an estimate. Novelcrafter's cost page uses 1 token = 0.75 words as a guide for GPT models, so 90,000 ÷ 0.75 = 120,000 tokens. Anthropic's price page says Claude 4.7 and later models produce about 30% more tokens for the same text, giving about 156,000. That is 78% of a 200K window before instructions, chat history and the reply, and 16% of a 1M window. These are two vendors' rules of thumb combined, so count your own text before relying on it.

Once project knowledge nears the limit, retrieval lets Claude search for relevant excerpts instead of loading everything. Anthropic says that can expand capacity up to 10x. It also means a question about your whole book may be answered from the passages retrieved. For a full-manuscript read, its best-practices page advises sending entire texts for editing in one message, and separately lists message length and attachment size as factors that affect your limit. Chat uploads allow up to 20 files, and we have not measured how reliably either route tracks continuity.

Who owns the output, and what happens to your text

Two Anthropic documents matter, and they should be read, not summarized from here.

• Ownership: the Consumer Terms, effective October 8, 2025, say Anthropic assigns "all of our right, title, and interest—if any—in Outputs" to you, subject to compliance. The words "if any" matter, and copyright in AI-generated material is a separate question; see AI and copyright for authors.
• Training: the same terms say your materials may be used to improve services, including training, "unless you opt out of training through your account settings". Feedback and safety-flagged content may still be used. The privacy center, last updated March 16, 2026, says training use can happen through the Model Improvement setting, safety flags or explicit opt-in, that Incognito chats are not used to improve Claude, and that conversations with thumbs feedback are stored for up to five years.

The two pages describe the default differently, so open your account settings and read the setting itself before pasting a manuscript. These are the consumer terms. The pricing page's Team row lists "no model training on your content by default", but the Team and Enterprise terms themselves were not read, and neither were API terms.

A Claude plan or an API key?

A Claude subscription and the Claude API are separate products billed separately, and the difference matters if you are considering a writing app that connects to AI models. The Pro plan does not include API usage, and Novelcrafter's FAQ says a Claude Pro subscription cannot be used inside it.

API pricing is per token. Anthropic's price list, read 2026-11-08, shows Claude Sonnet 5.5 at $2 per million input tokens and $10 per million output tokens. On Novelcrafter's hypothetical prompt of 4,176 tokens in and 500 out, that is $0.0134 a prompt, or about $2.67 for 200 prompts of 400 words each (our arithmetic, an estimate that keeps every word and ignores rewrites, chat and fees). The terms differ as well: Anthropic's Commercial Terms, effective June 17, 2025, govern API keys and say it "may not train models on Customer Content". The three-way comparison works through that bill with assumptions shown.

Switching costs and lock-in

Claude holds your chats and project files, and individual users can export their data from Settings > Privacy on the web or desktop app. The export includes chats, projects, memory and account data, but not the original files you uploaded. The download link arrives by email, expires after 24 hours and may work only once, and that page does not state the file format. Exported data cannot be imported into another personal Claude account. The practical advice is to treat the chat as a workspace and keep every decision that matters, such as the outline, character facts and style rules, in documents you own. If you later move to a manuscript tool, see Sudowrite vs Novelcrafter vs Claude for how the three store notes and what each documents for export.

KDP and AI-generated text

Passages Claude writes and you keep are AI-generated under KDP's rule, which holds "even if you applied substantial edits afterwards". Using Claude to critique or brainstorm while you write the prose falls under AI-assisted, which KDP does not require you to disclose. See our KDP disclosure guide and using AI without losing your voice.

What to confirm before paying

1. Open your account settings and read the training setting, then decide whether you are comfortable with it for unpublished work.
2. On the live pricing page, check the billing option you are selecting, since annual is charged up front.
3. Find the usage limits for your plan; the pricing page gives multiples of Pro usage, not word counts.
4. Check whether your publishing contract allows sending manuscript text to a third party.
5. Re-read the Projects help page if you plan to rely on it, given the beta new version.
6. Run one normal working session and check Settings > Usage before committing to annual billing.

Honest downsides

• No manuscript, scene or story-bible interface appears in the pages we read; you manage text in chats and Projects.
• Training depends on a setting you must check, and feedback or flagged content may still be used.
• Usage is metered in windows whose size in words we cannot state.
• Data export leaves out your original uploaded files, so keep your own copies.
• Free plans have five projects, and sharing needs Team or Enterprise.

Who should and should not use it

Use Claude for planning, feedback and brainstorming if you already want a general assistant. Look elsewhere if you want a dedicated manuscript workspace; compare Sudowrite, Novelcrafter and Claude or see best AI for fiction writers.

More tools and guides are in the AI drafting tools hub.

Last verified 2026-11-08. Rules and prices change: check the linked primary page before you rely on this.