Is this an AI detector?

No, and on purpose. AI detectors claim to prove authorship and are routinely wrong in both directions — flagging human writing and missing edited machine output. This tool checks five measurable style habits that make text read machine-flat, regardless of who typed it. A human can fail all five; a careful AI-assisted writer can pass all five.

What are the five tells it checks?

Em-dash density above 4 per 1,000 words; 26 stock phrases ("delve into", "moreover", "in today's fast-paced world"); rule-of-three padding ("fast, clean, and simple" repeated as a habit); uniform sentence rhythm, measured as length variation below 35%; and zero first-person markers across 200+ words.

Why does sentence rhythm matter so much?

Burstiness — the variation between short and long sentences — is the strongest single signal separating drafted-by-human text from raw model output. Models regress toward medium-length sentences; people interrupt themselves. The check computes the coefficient of variation across your sentence lengths and flags anything under 35%.

My text was flagged but I wrote it myself. Is the tool broken?

No — that is the point of checking style instead of authorship. Corporate and academic writing trained generations of humans into the same flat habits models learned. A flag means the passage reads machine-flat to an editor, which costs trust with readers either way. Fix the rhythm, keep your byline.

Does my text get uploaded anywhere?

No. All five rules run as plain JavaScript in your browser tab — the text never reaches a server, ours or anyone's. Paste a confidential draft, check it, close the tab; nothing persists.

Can it rewrite the flagged passages for me?

Deliberately not — a free rewriter would just trade one set of machine habits for another. Each flag ships with the editorial fix instead. For the full rewriting system with prompts that strip these patterns at the drafting stage, that is what Humanizer Pro is for.

// free tool · runs in your browser · not a detector

AI Writing Checker

Every page on this site passes five style rules before publish — em-dash budget, no stock phrases, no list padding, uneven rhythm, at least one sentence only a person would write. This tool is those exact rules, public, with the position of every violation and the fix next to it.

Quick answer

The five measurable AI writing tells are: em-dash density above 4 per 1,000 words, stock phrases like "delve into", repeated "fast, clean, and simple" triads, sentence-length variation under 35%, and zero first-person markers in 200+ words. This checker flags each with its exact position, locally in your browser.

Paste your draft — analyzed locally, nothing leaves this tab

Style tells, not an AI detector — it cannot prove who wrote a text, only where it reads machine-flat. The same 5 rules gate this site's own copy before publish.

// why tells, not detection

Detection is a losing game. Editing is not.

The detector industry sells certainty it cannot deliver: the same paragraph scores "98% AI" on one service and "human" on another, students get falsely accused, and every model release resets the arms race. I refuse to play that game on principle — and on math. What an editor can actually act on is narrower and more honest: specific habits that make prose read flat, each measurable, each fixable in one pass.

These five rules exist because I enforce them on this site's own copy. The em-dash budget came first — my drafts averaged eleven per thousand words before I started counting in 2025. The stock-phrase list grew from real edits, not a blog post's listicle. None of this proves authorship, and the report says so on every run. It proves something more useful: whether a reader's trust survives paragraph three.

The checker grades the symptom. Fixing the cause — drafting with AI without inheriting its voice — is a prompting-and-process problem, and it is exactly what Humanizer Pro packages: the rewrite prompts, the pattern library, and the workflow that produces drafts this page passes on the first run.

// common questions

AI writing tells — common questions

Is this an AI detector?: No, and on purpose. AI detectors claim to prove authorship and are routinely wrong in both directions — flagging human writing and missing edited machine output. This tool checks five measurable style habits that make text read machine-flat, regardless of who typed it. A human can fail all five; a careful AI-assisted writer can pass all five.
What are the five tells it checks?: Em-dash density above 4 per 1,000 words; 26 stock phrases ("delve into", "moreover", "in today's fast-paced world"); rule-of-three padding ("fast, clean, and simple" repeated as a habit); uniform sentence rhythm, measured as length variation below 35%; and zero first-person markers across 200+ words.
Why does sentence rhythm matter so much?: Burstiness — the variation between short and long sentences — is the strongest single signal separating drafted-by-human text from raw model output. Models regress toward medium-length sentences; people interrupt themselves. The check computes the coefficient of variation across your sentence lengths and flags anything under 35%.
My text was flagged but I wrote it myself. Is the tool broken?: No — that is the point of checking style instead of authorship. Corporate and academic writing trained generations of humans into the same flat habits models learned. A flag means the passage reads machine-flat to an editor, which costs trust with readers either way. Fix the rhythm, keep your byline.
Does my text get uploaded anywhere?: No. All five rules run as plain JavaScript in your browser tab — the text never reaches a server, ours or anyone's. Paste a confidential draft, check it, close the tab; nothing persists.
Can it rewrite the flagged passages for me?: Deliberately not — a free rewriter would just trade one set of machine habits for another. Each flag ships with the editorial fix instead. For the full rewriting system with prompts that strip these patterns at the drafting stage, that is what Humanizer Pro is for.

// the live version of Chapter 4 · Can I Quote You Cleanly?

This tool is a chapter of Found by AI

Whether an AI can quote you cleanly is Chapter 4 of Found by AI. This checker flags the writing tells that get in the way; the book turns that into a citability method, with a free workbook of all seven tools.

Get the free workbook →Read Found by AI →

// Also on this topic

postThe 7 AI writing tells editors catch first
Em-dash overload, "delve into", autopilot triads, same-length sentences. The tells editors catch first, why detectors miss the point, and our own checklist.
postClaude Code skills for marketing: the three that survive contact with real work
Marketing skill packs promise forty workflows. Three categories survive real use, one fails in a way that costs you customers, and the reason is structural.
postOne podcast episode to 12 content pieces: the actual repurposing workflow
Turn one 45-minute podcast episode into 12 pieces across six platforms over 14 days, with a 4-hour production budget after recording.
guideHow to get your website quoted in ChatGPT, Perplexity and Google AI answers
AI tools quote the cleanest sentence, not the prettiest paragraph. Here are four rules for writing answers ChatGPT and Perplexity can lift.
guideThe client deliverable is due in an hour and AI wrote most of it. How do you check it fast?
Re-reading an AI draft top to bottom catches little. Check only the facts that would hurt if wrong, against a real source. Steps for the next hour.

// the checker grades, the system fixes

Write with AI without sounding like it.

Humanizer Pro — the 3-skill rewrite system whose output passes this page's five rules. $19, one-time.

See Humanizer Pro →