How is the score calculated?

Five signals are combined: sentence-length variety (30%), word predictability from a word-frequency list (20%), density of 25 stock AI phrases (20%), contractions and conversational punctuation (15%) and repeated sentence openers (15%). Two rules can cap the result: when sentences keep opening the same way and first-person text has no contractions, the human-like score is capped at 28, and when two or more stock phrases appear at a high density (roughly 1.5 per 100 words), it is capped at 35.

Does it send my text to ChatGPT or another AI service?

No. The checks run on our own server with rule-based code and a word-frequency list. No generative model and no external AI API is called, and the text is not stored after the result is returned.

Why does a short text give a strange score?

Most of the signals need several sentences. Sentence-length variety cannot be measured in one sentence, and a single stock phrase in 20 words weighs far more than in 500. Paste a few paragraphs for a steadier reading.

Is the AI detector free?

Yes. It is free, needs no account and has no daily credit limit beyond a fair-use rate limit. Each check accepts up to 5,000 characters.

AI Text Detector

Score a draft for the habits AI-generated text tends to share, and see exactly which signals drove the number. It is an estimate of writing patterns, not proof of who wrote the text.

Original Text0 / 5,000

0 words0 chars<1 min

Five Signals, Each One Shown

The detector combines five measurements into a human-like score from 5 to 99. Sentence-length variety carries 30% of the weight: AI drafts tend to keep sentences close to one length, while people mix short and long ones and end some with a question.

Word predictability carries 20%. Each word is looked up in a word-frequency list, and a narrow spread of mid-frequency words reads as more machine-like than a mix of everyday and rare ones.

Stock phrases carry 20%. The text is searched for 25 phrases AI drafts overuse, such as 'delve into', 'it is important to note that' and 'plays a crucial role'. Contractions and conversational punctuation carry 15%, and repeated sentence openers carry the last 15%.

Why the Result Is an Estimate, Not a Verdict

These are rules, not a trained classifier, and they have not been validated against a labelled set of human and AI text. Careful, formal human writing can score as AI-like, and an AI draft that has been edited can score as human.

Use the result to find passages worth revising, never as evidence against a student, employee or writer. No detector on the market can establish authorship reliably, and this one does not claim to.

Reading the Score Bands

A human-like score of 85 or more is labelled 'Strongly human-like patterns', 65 to 84 'Mostly human-like patterns', 45 to 64 'Mixed patterns', and below 45 'Many AI-like patterns'. The list under the score names the signals behind it, such as low sentence-length variation or the stock phrases found, so you know what to change.

From Detection to Revision

The same editor can rewrite the text. Humanize runs it through the rule-based rewriting engine, then scores the original and the rewrite side by side, so you can see which signals the rewrite changed and which ones still need a human edit.

Common Use Cases

Checking your own draft for stiff, formulaic phrasing before you publish it
Finding the stock AI phrases left in a ChatGPT-assisted draft
Seeing whether an edit made a passage read more naturally
Comparing an original and a humanized version on the same measures
Teaching what makes writing sound mechanical, with concrete examples

How It Works

Paste your text

Up to 5,000 characters. A few paragraphs give a steadier reading than one sentence.

Check writing patterns

The five signals are measured locally, with no external AI API.

Read the signals

Each factor behind the score is listed, from sentence variety to stock phrases.

Revise or humanize

Edit the flagged passages yourself, or rewrite and compare both scores.

Frequently Asked Questions

No. It measures writing patterns that AI drafts tend to share, and people produce the same patterns, especially in formal writing. The result tells you which passages read as formulaic, not who wrote them. Do not use it, or any other detector, as the basis for an accusation.