Self-diagnosis
Readers rarely say which patterns triggered them. They just write “generic”, “repetitive” or “feels like ChatGPT” in a two-star review. These are the patterns.
A book sounds like AI when it leans on a small set of recognizable patterns: stock vocabulary, uniform sentence rhythm, chapters built on one repeating template, and prose that could sit in any book on the topic. No single sign condemns a manuscript. Several of them together are what readers react to, even when they cannot name a single one. Below are the 15 signs the ManuScore rubric checks, grouped into four families: words, rhythm, structure, and voice.
A cluster of words appears in AI drafts at rates no human writer matches: delve, tapestry, realm, unlock, unleash, elevate, robust, seamless, game-changer, plus landscape, navigate and leverage used metaphorically. One occurrence means nothing. In the ManuScore rubric, more than 10 occurrences per 10,000 words costs 4 of the 5 available points on this criterion, and more than 25 scores zero. A recent calibration manuscript we audited carried 14 per 10,000 words, including “delve” four times and “seamless” six times.
Whole phrases lifted from the same mold: “in today’s fast-paced world”, “it’s important to note”, “in conclusion”. They carry zero information and readers skim straight past them.
“In today’s fast-paced world, consistency is key.”
one possible rewrite“Most print-on-demand stores die in month two. The ones that survive post every single week.”
Paragraphs that open with furthermore, moreover, additionally or however in strict rotation. Human writers connect ideas through content. AI drafts connect them through connector words, and once you see the rotation you cannot unsee it.
“This may potentially help you”, “results might vary”, “it could be argued”. AI models hedge by default because hedged claims are never wrong. A how-to book that never commits to a claim gives the reader nothing to act on.
Open any spread of your book and squint. If the paragraphs form even, regular blocks of nearly identical size, that visual monotony is a tell on its own. One audited manuscript had nine consecutive three-sentence paragraphs in a single chapter. Human writing is lopsided: a long winding paragraph, then a three-word one.
“You should start with a niche. The niche determines your audience. Your audience defines the design style.” Subject, verb, object, repeat. Grammatical, correct, and hypnotically flat over 200 pages.
AI drafts love neat triads: “faster, cheaper, and easier”, “plan, execute, and review”. One triad per chapter is style. One per paragraph is a pattern generator at work.
The signature AI construction: “This isn’t just a business model, it’s a mindset shift.” It fakes profundity by relabeling the thing it just said. If your manuscript uses it more than once or twice, it was almost certainly not you.
The em dash is a legitimate punctuation mark that AI models use at several times the human rate, often two or three per paragraph. Heavy em dash density has become one of the most widely recognized public signals of AI text, fairly or not. Search your manuscript and count.
“So how do you actually find a profitable niche?” Once per book: fine. Opening every second section: a template. Readers experience it as being talked at by a slide deck.
Short intro, three bulleted sections, closing summary. Then again. Then again. In one audit, all 11 chapters followed the identical internal template, and that single finding cost the book its entire structure score. Human authors vary structure because the material demands it: a case study opens one chapter, a checklist another.
“In this chapter, we covered…” Summaries that re-chew what the reader finished seconds ago are a default AI closing move, and readers report skipping them entirely.
The five-random-pages testOpen five random pages. Each page must contain at least one detail that only this book could have: a personal story, a stated opinion, a specific number from lived experience. A page that could sit unchanged in any other book on the topic fails. This test carries 15 of the 100 points in the ManuScore rubric, the single largest criterion, because it measures the exact thing readers punish in reviews.
AI drafts present every option evenly: “both approaches have their merits.” A human expert has scars and favorites. If your book never says “most guides tell you X, and I think that’s wrong”, it reads as compiled rather than written.
No failed attempts, no boring-but-necessary steps, no “this part took me three months longer than it should have”. AI text is frictionless because the model has never been stuck. Friction is the most human texture a book can have, and its total absence is the deepest tell of all, because it cannot be fixed with a find-and-replace.
Signs 1 through 12 are mechanical: you can find and fix them with focused editing passes. Signs 13 through 15 are structural: fixing them means adding material only you possess, which typically means rewriting 30 to 50 percent of the text by hand. That is the difference between a book that needs a cleanup and a book that needs humanization, and it is exactly what separates a “minor revision” verdict from a “major revision” verdict in an audit. If you want the step-by-step method, read How to humanize AI-written text next.
The free ManuScore mini-audit runs a sample of your book (up to 5,000 words) against the same rubric used in the full audit: estimated verdict band, your three weakest dimensions, an AI-tell phrase count, and one quoted example from your own text.
Run the free mini-auditno signup · instant · English manuscripts