An AI detector for long documents, up to whole books

This AI detector for long documents takes a novel, a thesis or a full manuscript in a single check: paste the text or open a .docx file below. It finds the chapters, shows where the higher-scoring passages sit and, for fiction, adds a narrative fingerprint. It is free, needs no account, and the manuscript never leaves your computer.

Whole books in one check

There is no limit on words, characters or chapters. A 208,000-word novel has been analysed in one go. Every word is scored, never a sample. About 30,000 words take two to three minutes on a typical laptop and a full novel 15 minutes or more. The scan streams partial results as it works, and can be cancelled at any point. Above 30,000 words the text is shown in pages of 5,000 words so your browser stays responsive.

Long files are fine as long as your device has the memory for them. You can paste the text or open a .txt or .docx file.

How the AI detector for long documents shows chapters

Chapters are found from their headings, so a manuscript laid out with “Chapter 1”, “Chapter 2” and so on gets a result for each one. This is where checking a long document properly pays off. One overall verdict on a 90,000-word book says very little. A chapter that scores well above its neighbours tells an author or editor exactly where to start reading.

Read chapter results as a guide: a single high chapter is a prompt to look at that chapter's drafts and history. It doesn't settle anything by itself.

A narrative fingerprint for fiction

For fiction, the report adds a narrative fingerprint: simple counts of storytelling features in your manuscript. They include how much of the text is dialogue, how often smell turns up in descriptions, how often the narrator spells out what a scene means, direct address to the reader and markers of jumps in time. Each count sits beside the human and AI base rates published in StoryScope, a study of AI fiction by Russell and colleagues (COLM 2026). That research found that AI and human stories differ in their narrative choices as well as their wording. AI stories, for instance, tended to explain their themes outright.

The fingerprint is descriptive. Our counts are heuristics inspired by the paper, not a reimplementation of it, and they never feed into the verdict. It can still be useful editorial feedback when nobody suspects AI at all. There is more on this in how AI fiction differs.

Theses, reports and other non-fiction

A thesis, a long report or a book-length article goes in the same way as a novel: one check, with chapters found from their headings. The detector reads English only, and chapters in other languages come back Inconclusive.

How to check a manuscript or thesis

1. Save the document as .docx or .txt, or select all of it and copy.
2. Put each chapter heading on its own line, for example “Chapter 12”, so chapters can be found.
3. Open the file or paste the text in the box above, and press Check for AI.
4. Read the overall outcome, then use the chapter menu to jump to any chapter.
5. For fiction, open the narrative fingerprint and compare each feature with the published base rates.

What a long-document result means

Like any detector, it can be wrong in both directions, so treat a result as a reason to look closer, not as proof, and never as the only evidence against an author. Our methodology page explains how it is tested.

Sources

1. StoryScope: Investigating idiosyncrasies in AI fiction (arXiv 2604.03136), checked October 2026