Craft2026年11月29日约 8 分钟

AI Book Analysis Tools: Three Checkups, One Standard

No analysis tool replaces an editor. But three of them replace three different conversations — and a score is the least useful thing any of them produces.

Sam RiveraFounding editor2026年11月29日约 8 分钟
分享
A printed manuscript with a magnifying glass, colored tabs and a pencil on a wooden desk.
A printed manuscript with a magnifying glass, colored tabs and a pencil on a wooden desk.
本文目录

A manuscript checkup is not one service. It is at least three, and confusing them is how writers spend money on the wrong report.

Marlowe, Fictionary and AutoCrit are often mentioned in the same breath because they all promise editorial feedback. In practice they look at different layers of the same book, and each is strongest exactly where the others are weakest.

This is a capability comparison based on what the three tools document and what they explicitly refuse to do — not a scored shootout. The useful question is not “which one wins” but “which conversation am I trying to have with my manuscript today”.

Three tools, three questions

ToolThe question it answersThe layer it readsWhat it will not do
MarloweHow does this compare to published novels in my genre?Developmental: plot, pacing, character consistency, themeRewrite a word
AutoCritWhere is the prose repeating, dragging or leaning on habits?Line and story reports: repetition, passive voice, adverbs, pacingMake the structural call for you
FictionaryIs the structure sound, and will I finish?Scene and story arc, with coaching and accountabilityLine-edit the prose

Marlowe: the developmental benchmark

Marlowe is the most explicit about what it is not. Its own product page says there is no “rewrite this chapter” button, and that it does not write or rewrite a single word. It is analysis, not generation.

What it does is compare your manuscript against a corpus of published novels and report on the dimensions a developmental editor would raise:

  • plot structure and narrative arcs
  • character development and consistency
  • pacing
  • dialogue balance and effectiveness
  • theme and emotional resonance
  • genre conventions, benchmarked against bestsellers in your category
  • readability, clichés and repeated phrases

The genre-aware part is the differentiator. “Pacing is slow” is useless advice. “Pacing is slow for a thriller, where the inciting incident is expected earlier” is a note you can act on.

Practical constraints matter too: it accepts .docx or ePub, the manuscript has to be at least 20,000 words, and reports come back in minutes rather than weeks. As of this writing, single reports are listed at $29.95, with subscription options.

Use it when: you have a complete novel draft and want a developmental read before you spend on a human editor or a cover.

Do not use it when: the draft is still changing shape, or the book is a practical guide. A benchmark against published fiction will not know whether your advice is correct, current, or useful.

AutoCrit: the line-and-habit report

AutoCrit sits at the other end. Its pitch is a studio that plans, writes, analyzes and polishes, with 25+ reports and a Story Analyzer that looks at pacing, character inconsistencies, plot threads and worldbuilding errors.

The reports that writers actually talk about are the line-level ones: repeated words, passive voice, adverb density, clichés, dialogue balance, and the personal habits you did not know you had. The tool lets you add your own crutch words to the watch list, which is the closest thing to a custom editor’s margin note.

Two things make it useful in practice:

  • It is exhaustive in a way a human reader is not. Your eyes stop seeing “just” and “actually” after the third pass. A report does not get tired.
  • It produces a revision list, not a rewrite. That keeps the voice with you.

Writers who use it describe a clear split in how these tools behave. AutoCrit produces reports rather than rewrite suggestions, which is good for keeping your voice but means you do the fixing. Suggestions-first tools lean the other way: they hand you replacements, and those sometimes miss the context or the tone. Experienced users often run both — a report pass for repetition and redundancy, then a suggestion pass for fragments and mechanics — before handing the manuscript to a human.

The risk is the opposite of Marlowe’s. Line reports invite over-editing: every flagged adverb becomes a deletion, every passive construction a rewrite, and the prose ends up sanded into a house style. The report is a list of questions, not a list of orders.

Two caveats from writers who paid for it: the analysis can be uneven — one common complaint is that it fixates on minor characters while saying little about the protagonist — and the grammar layer is thin enough that users pair it with a separate grammar checker. Neither flaw is fatal, but both argue for using a single chapter as a trial before subscribing.

Use it when: the structure is settled and you are polishing chapter by chapter.

Do not use it when: you are still deciding what the book is about. Line polish on a chapter that will be cut is wasted work.

Fictionary: the structure and finish tool

Fictionary is less a report and more a workshop. It combines a structure workspace with live classes taught by certified story editors, an AI editor called Edwina that reads scenes and suggests improvements, and weekly check-ins.

Its signature is the story arc, drawn against expected positions:

  • Inciting incident: up to 15% of the novel
  • Plot point 1: 20–30%
  • Middle: 45–55%
  • Plot point 2: 70–80%
  • Climax: 85–95%

That is a structural checklist, not a quality judgment. A book can hit every marker and still be dull; a book can ignore them and work. But if your inciting incident lands at 40%, the tool has done its job by making you defend the choice.

The other half is accountability. Fictionary’s pitch is explicit: weekly check-ins mean someone expects your next chapter. For writers whose problem is finishing rather than polishing, that is worth more than any report.

Use it when: you are planning or mid-draft and the middle is sagging, or when finishing is the hard part.

Do not use it when: you want a line edit, or you are writing a practical book with its own structure. A guide earns its shape from the reader’s needs, not from a novel’s beat sheet.

The score is not the standard

Here is the part the reports will not tell you.

A score compares your book to an average — the average of published novels in a genre, the average sentence length, the average amount of passive voice. Chasing a higher score does not make the book better. It makes the book more average, which is the opposite of why anyone reads.

The same trap exists in the “AI feel” debate. One camp says AI prose is detectable and lifeless; another says the grammar and polish are the point. Both are arguing about the wrapper. A reader does not finish a chapter because the sentences are technically clean. They finish it because they learned something, recognized something, or got somewhere.

For a book built from experience, that is the whole product. The reader came for the thing you know — the drill that fixed the elbow, the clause that saved the invoice, the mistake that cost you a year. A report cannot tell you whether that survived the chapter. It can tell you the sentence is long.

So use the reports for what they measure, and hold the real standard yourself:

  • Did each chapter deliver what its title promised?
  • Could the reader act on it tomorrow?
  • Is every case and number in the chapter backed by your notes?
  • Would the person you wrote it for recognize their problem on the page?

Those questions do not produce a score. They produce a book worth reading.

What none of them do

All three stop at the same boundary: they can describe the book, they cannot decide whether it deserves to exist.

  • They cannot hear your voice. A report can tell you a sentence is long; it cannot tell you the length is the joke.
  • They cannot weigh a chapter. “This thread appears in four chapters” is a fact. “Cut it” is a judgment.
  • They cannot know your reader. Genre benchmarks describe an average reader of published books, not the specific person you are writing for.
  • They cannot check your experience. Only your notes and your knowledge base can tell a report whether the case on page ninety is the case you actually lived.

The realistic division of labor: tools find patterns, beta readers report experience, a human editor makes structural calls, and you make the final ones.

How to use a report without flattening the book

  1. Read the manuscript yourself first. Write down the three things you already suspect. If the report agrees, you have confirmation; if it disagrees, you have a real question.
  2. Sort findings into structural and cosmetic. Structural first. There is no point fixing commas in a chapter that should be cut.
  3. Verify every finding against the page. Reports misread tone, especially irony and deliberate repetition.
  4. Fix in batches by category. All the pacing notes, then all the repetition notes. Switching categories per page is how revisions stall.
  5. Keep versions. A report-driven rewrite should never overwrite the draft you might want back.
  6. Stop when the list stops being useful. The goal is a better book, not a clean report.

Where continuity fits

All three tools look at the manuscript as a finished object. The failures that hurt most in a long book happen earlier, while drafting: a number changes, a term is defined twice, a case introduced in chapter three is retold as new in chapter nine.

That is a different check, and it works best when the book’s state is stored rather than remembered. In our editor, facts live in Global knowledge and are injected into every chapter; your source notes sit in a knowledge base mounted to the book; covered topics stop later chapters from re-teaching; each chapter keeps a version history. The manuscript and the checklist are in the same place, so a contradiction is caught while drafting instead of in a $29.95 report six months later.

The bottom line

Marlowe, Fictionary and AutoCrit are not competitors so much as three different checkups. Marlowe tells you where the book sits among its peers. AutoCrit tells you where the prose is leaning on habits. Fictionary tells you whether the structure holds and whether you will finish. Pick the checkup that matches the question you actually have. Then remember the part no report covers: deciding what the book is for, and whether the reader got it.

Key takeaways

  1. 1Marlowe benchmarks a finished draft against published novels and never rewrites.
  2. 2AutoCrit is strongest on line-level habits and revision lists, with an over-editing risk.
  3. 3Fictionary is a structure and accountability tool, not a line editor.
  4. 4Scores compare you to an average; use the observations, not the number.
  5. 5The standard no report measures: did the reader get the experience the book promised?

常见问题

Which is better, Marlowe or AutoCrit?

They answer different questions. Marlowe benchmarks a manuscript against published novels in its genre and reports on structure, pacing and character consistency without rewriting anything. AutoCrit runs line-level and story reports: repetition, passive voice, adverbs, pacing, dialogue balance. Use Marlowe for a developmental read, AutoCrit for a line and revision pass.

Is Fictionary a writing tool or a course?

Both. It combines a structure workspace with live classes taught by certified story editors, scene-level notes from an AI editor called Edwina, story-arc visualization and weekly check-ins. Its strength is finishing a novel, not analyzing prose line by line.

Can an AI report replace a human editor?

No. It can surface patterns a tired reader misses: repeated phrases, pacing dips, character inconsistencies, genre mismatches. It cannot decide whether the book is worth publishing, and it cannot hear what your voice is doing. Treat reports as question lists, not verdicts.

Do the scores matter?

Not much. A score is a summary against the average of the books a tool has compared you to. A higher score is not a better book, and a lower one is not a worse one. The useful part is the specific observations underneath: the chapter where pacing drops, the term defined twice, the case that contradicts your notes.

参考资料

  1. Authors A.I. — Marlowe — 查阅于 2026-11-19
  2. AutoCrit — 查阅于 2026-11-19
  3. Fictionary — 查阅于 2026-11-19

关于作者

Sam Rivera

Founding editor

Writes about product content, publishing systems, and the unglamorous parts of shipping software. Every guide here is edited against the product it describes.

每两周,一封有用的信

关于写作、发布,以及产品上线前后那些不体面的细节。没有热点合集,也没有增长黑客。

继续阅读

全部文章
An editorial illustration of a long ribbon of uniform typed lines being cut with scissors, with looser hand-drawn strokes escaping from the cut.
Craft2026年11月13日约 4 分钟

Why AI Writing Sounds Like AI (and What to Cut First)

The tell is rarely a single word. It is the shape of the paragraph, the symmetry of the list, and the sentence that explains a noun you already understood.

#Ai Writing#Editing#Voice

Tools for the job

All tools