Manuscript Analysis Free scan · optional AI rewrite tips

AI Writing Detector for Fiction & Websites

Paste a chapter excerpt or scan a webpage URL to flag fiction AI tells, see how strongly the prose matches Claude, ChatGPT, DeepSeek, Gemini, and more, then optionally get free DeepSeek rewrite tips. Craft risk score — not a binary “% AI” verdict.

All tools Full cleanup report
Paste or fetch a URL — the tell scan runs in your browser after load. Optional AI rewrite tips use free AI (DeepSeek via Kestrel — Free). Craft risk only — not proof of authorship.
Paste a chapter excerpt or scan a webpage URL. Flag fiction AI tells, guess the likely model (Claude, ChatGPT, DeepSeek, Gemini, …), and optionally get rewrite tips. Craft risk — not a binary “% AI” score.

Pulls the main article/chapter text (SSRF-safe). Blocked/paywalled pages: paste the prose below instead.

Ctrl+Enter

Results appear here after you scan. Paste prose, Fetch & scan a webpage, or try the sample.

Related tools

Cliché & Overused Phrase Finder Filter Word Finder Show Don't Tell Hints Manuscript Cleanup Report

About this tool

Looking for an AI writing detector built for fiction — and pages you can open in a browser? Paste a chapter excerpt or fetch a public webpage URL. This free tool flags fiction-specific AI tells from ChatGPT-style stock and polished Claude literary scaffolding, then estimates how strongly the prose matches Claude, ChatGPT/GPT, DeepSeek, Gemini, Llama-family, or Grok habits (independent confidence scores — not proof). You also get a craft-risk score (0–100). Instant scan runs in your browser after text is loaded; optional rewrite tips use free AI (DeepSeek via Kestrel — Free).

What you can do

  • Paste prose or scan a public webpage / chapter URL for fiction AI tells
  • Guess likely model family (Claude, ChatGPT, DeepSeek, Gemini, Llama, Grok) with per-model confidence
  • Flags ChatGPT stock phrases and Claude-style literary scaffolding (thesis closers, So/Then chains, mirrored contrasts)
  • Craft-risk score with plain-language bands (low → high) — never a binary guilty verdict
  • Em-dash and filter-word density checks common in AI-assisted novels
  • Optional DeepSeek rewrite tips on the worst hits (Kestrel — Free pool)
  • Export flagged tells, fingerprints, and rewrite tips to CSV
  • Sample passage + local draft restore so you can test without signup

Why manuscript analysis matters for fiction

Rule-based analyzers catch patterns human eyes skim past—especially in AI drafts where filter words, adverbs, and passive voice cluster predictably. Use metrics to prioritize revision, not to flatten your voice. Run a chapter first, fix the worst offenders, then scan the full manuscript.

Five-pass revision order · AI manuscript editing · Manuscript cleanup

How to use it

  1. Paste a chapter excerpt, or enter a public webpage URL and click Fetch & scan.
  2. Review the extracted text (edit if needed), then scan for AI tells in your browser.
  3. Check the craft-risk score and flagged hits by category.
  4. Optionally click AI rewrite tips for DeepSeek human rewrite suggestions.
  5. Export CSV or revise dense clusters before publishing.

Tips for fiction authors

  • One jaw clench can be perfect; twelve in a chapter is the tell. Density matters more than any single hit.
  • Polished Claude prose often skips “something shifted” — watch for thesis closers (“Those are two different thoughts”) and procedural So/Then scaffolding instead.
  • Website fetch works best on public chapter/article pages. Cloudflare or login walls: paste the text instead.
  • Do not use this score to accuse anyone of cheating — humans share these patterns too. Use it as a craft checklist.
  • Run on emotional beats and dialogue scenes first — that is where AI tells cluster hardest.
  • After AI rewrite tips, re-scan. If the score barely moves, you may need structural rewrite, not synonym swaps.

Common questions

Is this a real AI detector that proves ChatGPT wrote my chapter?

No. It measures fiction craft tells — patterns common in AI-assisted prose. A high score means revise those patterns, not that a model is proven.

Can it tell Claude from ChatGPT?

It estimates stylistic confidence per model family (Claude, ChatGPT/GPT, DeepSeek, Gemini, Llama, Grok) from craft tells. That is a fingerprint guess — not forensic proof — and fine-tunes blur the signal.

Can I scan a website URL?

Yes. Paste a public chapter or article URL and click Fetch & scan. We extract main prose (SSRF-safe). Paywalled or Cloudflare-blocked pages need a manual paste.

Why did polished Claude prose score low before?

Older detectors chase ChatGPT stock phrases. Strong Claude fiction often avoids those and leans on thesis closers, explanatory “which” glosses, and So/Then scaffolding — this tool flags those too.

Why “for fiction” instead of a general AI writing detector?

Essay detectors chase moreover/furthermore. Fiction AI leans on body-language stacks, “something shifted,” and literary scaffolding. This tool targets those — including on story webpages.

Does the instant scan upload my manuscript?

Pasted text is scanned in your browser. Fetching a URL loads that page through our server so we can extract readable prose; optional AI rewrite tips send flagged excerpts to free AI.

What model powers rewrite tips?

DeepSeek via Entangled Text’s Kestrel — Free pool (platform/admin API key). Guests get a limited free tip quota per device.

How is this different from the cliché finder?

Cliché finder is one phrase list. This detector scores multiple fiction AI tell categories together, can fetch a webpage, and can suggest rewrites.

Related guides

Why AI Writes Every Character Wanting the Same Thing How to Use AI to Diagnose and Fix Sentence-Level Monotony How to Use AI to Find Where Your Scenes Overstay Their Welcome How to Use AI to Write a Character's Interior Monologue Style

Comments (0)

Share a tip, question, or experience — keep it useful for other writers.

Prefer a public profile and edit access? Sign in or create a free account.

5,000 left

No comments yet. Be the first to share a thought.