Editing Software That Reads Your Whole Manuscript (Not Chapter by Chapter)

Most novel editing software will happily take your whole manuscript. Far fewer will read it whole. Upload a 90,000-word draft and many tools quietly split it into chapters, scenes, or fixed-size chunks, then comment on each piece as if the others did not exist. For a line edit that is fine. For the questions a novelist actually lies awake over, it is the wrong shape.

This post sets out why whole-book reading matters for fiction, the three ways editing software handles a long manuscript, and a table of eight tools with the limits each vendor states on its own site. Where a vendor does not state a limit, the table says so rather than guessing. If you are not sure yet which kind of editing your draft needs, start with what kind of editing your book needs, because the answer changes which tools are even relevant.

Why a novel has to be read whole

A novel is a system. The problems that sink one are rarely inside a single chapter; they live in the relationship between chapters. Four kinds in particular cannot be seen from one piece at a time:

• Setups and payoffs. The locked drawer in chapter three needs to be opened by chapter thirty, and the clue in chapter twelve has to be fair when the twist lands. A reader of chapter thirty alone cannot know what was promised.
• Character arcs. Whether a change in your protagonist feels earned depends on everything she did before it. Judged scene by scene, every scene can look fine while the arc as a whole does not move.
• Pacing across acts. A sagging middle is a proportion problem: too much of the book spent in one gear. You can only see proportion from far enough back to take in the whole shape.
• Continuity. Eye colours, ages, timelines, who knows what and when. These slip precisely because each chapter was written, and often reviewed, on its own. We wrote a practical guide to keeping characters consistent across chapters.

These are the questions of a developmental edit, and they come first for a reason: there is no point polishing sentences in a subplot you are about to cut. What order to edit your novel in goes through the sequence.

Three ways software reads a long manuscript

Chapter-limited or chunked. The tool analyses one piece at a time, either because it has a hard cap per run or because it asks you to split the book. This is how most prose and style checkers work, and it suits them: a sentence-level suggestion does not need chapter thirty to be in view.

Whole-manuscript report. You upload the full draft and get one report back covering the book. Marlowe and inkshift work this way, as do ProWritingAid's separately bought advanced reports. The report is written once; revising, and deciding what the notes mean, happens on your side.

A general chatbot. ChatGPT, Claude and similar tools read whatever fits inside the conversation. The newest models have very large windows, so a whole novel can fit. What changes is what happens as the conversation grows, which we come back to below.

Manuscript limits, tool by tool

To be fair to every tool here: each is built for a particular job and does it on purpose. ProWritingAid's own help centre recommends splitting a long document into chapters so it runs faster and does not bury you in suggestions, which is sound advice for line editing; our ProWritingAid review covers where that fits. AutoCrit gives you the choice of whole-book or per-chapter analysis, and our AutoCrit review looks at the rest of the platform. Fictionary's scene-by-scene approach is a deliberate teaching method: it walks you through evaluating each scene yourself, which is why we describe it in Edith vs Fictionary as a framework rather than a report. Marlowe and inkshift both genuinely take the whole book, and we set each against Edith in Edith vs Marlowe and Edith vs inkshift; our Marlowe AI review goes deeper on that one. For prices and a wider field, see our comparison of the best book editing software.

One caution in reading any table like this: a word limit tells you what a tool will accept, not what it does with the book once it has it. That is a question to test, not to take on trust, and there is a quick way to test it further down.

General chatbots: the limit is the conversation

The honest news about chatbots is that the raw size problem has mostly gone. On paid plans, Claude's newest models hold up to a million tokens, which is more than a long novel. By OpenAI's rule of thumb that 100 tokens is about 75 words, even the 128k-token input figure from its February release note works out at roughly 96,000 words.

What remains is what happens over a long working session. An edit is not one question; it is dozens, and every question and answer takes up room in the same window as your book. Anthropic's help centre says that when a Claude conversation nears its limit, earlier messages are summarised to make space, which keeps the chat going but means the fine detail from early on is no longer there word for word. And fitting is not the same as attending evenly: research on long inputs has found that models use the start and end of a long text better than the middle, which is the subject of why AI forgets the middle of your book. None of this makes a chatbot useless; for a question about one scene it is quick and free. It is the reason we compare the two approaches directly in Edith vs ChatGPT.

A ten-minute test: did it really read the whole book?

Whatever tool you try, the free tier or a sample is enough to check. Ask it, or look in its report for, things that can only be answered from the whole book:

1. Name something you set up early and pay off late, and ask whether the payoff lands. A tool reading in pieces may only speak to one end or the other.
2. Ask where the middle loses momentum, and in which chapters. A whole-book reader can point to a stretch; a chunked one may answer in generalities.
3. Plant a small continuity slip (a changed name, a wrong age) far from where the character was introduced, and see whether it is caught.
4. Ask how your protagonist has changed between the first and last chapters, and whether the change is earned.

If the answers are specific to your story, it read the book. If they could apply to anyone's, that tells you something too. How to get honest feedback on your novel has more questions in the same spirit.

Where Edith fits

Edith is built on the whole-book read. She reads your entire manuscript at once, up to 150,000 words, before she writes a word, and her first answer is an editorial letter on the book as a whole: what is working, what is holding it back, and what to fix first. You can see what that looks like in the sample editorial letter. Handing over a whole book raises a fair privacy question, which we answer in is it safe to upload your manuscript to an AI tool?

The letter is where the edit starts, not where it ends. Every note can be talked through in a chat: ask why, push back, explain what you meant, and settle it your way. Edith remembers what you settled and carries it through all six stages, from the developmental read to proofreading, and after you revise she re-checks the book for consistency. She suggests; she never rewrites. Every word stays yours, and the finished edit downloads as a Word file, with margin comments in the story stages and tracked changes in copyediting and proofreading.

It pointed out things I didn't consider and made me re-evaluate and change things in a way that made the book so much more cohesive.

Your first 10,000 words are free, no credit card needed. The whole-book Editorial Review is $15, and the complete six-stage edit is $249 once per book, with the $15 credited toward it and no subscription.

Sources:AutoCrit support: each chapter must be 200 to 10,000 words to run the Story Analyzers; free Fiction Analyzer limit·AutoCrit: Story Analyzer analyses "as one cohesive whole or chapter by chapter"·ProWritingAid Help Center: Chapter Critique up to 4,000 words; advanced reports up to 300,000; split documents into chapters·ProWritingAid Help Center: Manuscript Analysis not included in Premium or Premium Pro·Authors A.I. FAQ: Marlowe minimum 10,000 words, no maximum·Authors A.I. pricing: one credit covers any book length·Fictionary StoryTeller: AI feedback on every scene·Fictionary: importing a manuscript (Word file, chapters and scenes)·Fictionary FAQs: 25,000 words or more is a great time to start·inkshift FAQ: $25 critique for manuscripts up to 250,000 words·inkshift pricing: free critique of your first 10,000 words·OpenAI ChatGPT release notes, 20 February 2026: Thinking context window 256k tokens·OpenAI Help Center: understanding and counting tokens·Claude Help Center: context window on paid Claude plans·Liu et al., Lost in the Middle: How Language Models Use Long Contexts (2023)