Free AI detector
Paste your text, click Check for AI, and get an AI probability for the whole document and for every paragraph. The paragraphs that read machine-made are highlighted, so you know exactly which ones to fix — and the complete report downloads as a file you can keep.
Free up to 800 words per scan, no account needed. Scans from 800 to 3,000 words need an account and cost 1 credit per 50 words.
Overall AI score
0%
—
Score breakdown
Paragraph by paragraph
Under 40% reads human · 40–60% mixed signal · over 60% flagged as AI
How it works
Paste your text
Copy text from ChatGPT, Claude, Gemini or any other tool, or upload a Word, text or Markdown file. A few paragraphs is the sweet spot.
Click Check for AI
The detector scores the document and every sentence in it, then groups those scores by paragraph. Most scans finish in a few seconds.
Read the highlighted result
Red paragraphs are the ones carrying the signal. Download the report, or send those paragraphs to the editor and rewrite them in your own words.
How to read the score
Under 40% — reads human
The patterns detectors look for are not concentrated here. Treat this as an all-clear for that paragraph, not a proof of authorship.
40–60% — mixed signal
Something in the paragraph reads templated: even sentence lengths, generic phrasing, or an editing pass that smoothed the rhythm. Read it and decide.
Over 60% — flagged as AI
Rewrite these first. The full report lists the sentences that pushed the score up, so you are not editing blind.
Where every detector, including this one, falls short
Short text carries little signal. Below roughly 150 words there is not enough rhythm for the statistics to settle, and a two-sentence paragraph can score high or low almost by accident. Scan a full draft when the answer matters.
Formulaic human writing gets flagged. Academic abstracts, legal boilerplate, technical documentation and fluent second-language writing all look statistically "safe" — which is exactly what machine text looks like. A high score is a reason to look closer, never proof.
Paraphrased and humanised text slips through. Once an AI draft has been rewritten line by line, most of the signal is gone. That is why the useful move is to fix the flagged paragraphs rather than chase a number.
How we test explains what we measure and what we refuse to claim.
Detected something? Rewrite just those paragraphs.
Paste the flagged paragraphs into the editor, keep your meaning and your Word formatting, and check the result again. 200 free words once you confirm your email.
Get started freeQuestions
Is the AI detector free?
Yes, up to 800 words per scan and no account required. Scans between 800 and 3,000 words need an account and cost 1 credit per 50 words, because the detection model is billed per word.
How accurate is it?
It returns a probability, not a verdict. Text over about 150 words is where the score carries real signal; short snippets, heavily paraphrased drafts and edited AI text all read closer to human, and a fluent human writer in a formulaic genre can read as machine-made. That is true of every detector on the market, including this one.
What do the percentages mean?
Under 40% reads human, 40 to 60% is a mixed signal, and over 60% is flagged as AI. The paragraph view shows which parts of your document drove the score, so you can rewrite those and leave the rest alone.
What happens to my text?
It is sent to our detection provider to be scored and is not stored on your account or used to train anything. Nothing is published anywhere. See the privacy policy for how submissions are handled.
Which languages does it check?
Text in any of the 40+ languages we support, with the interface in ten. Non-English text is translated for scoring, so the scores are comparable across languages while the sentence-level split is approximate.