AI Content Detector
Check text for the patterns that language models tend to produce — and see exactly which measure drove the score. It reports signals, not verdicts about who wrote something.
Result
Paste a passage to see the signal breakdown.
This measures writing patterns, not authorship. No detector can prove who wrote something, and formal human writing often scores high. Never use a score like this to accuse anyone.
What this tool measures
Six measurable properties of the text, weighted into one score. Sentence-length variety, or burstiness, carries the most weight: human writing tends to swing between short punchy sentences and long winding ones, while generated prose settles into a steady rhythm.
The rest cover stock AI phrasing (roughly sixty phrases models lean on), repeated sentence openers, how evenly paragraphs are sized, vocabulary range across the passage, and em-dash density — models punctuate with them far more often than most writers do.
Each signal is shown separately with its own score and a plain-English explanation, so you can see what drove the result rather than taking a single number on trust. Forty words is the minimum; below that, none of these measures mean anything.
What it cannot do
It cannot tell you who wrote something. No detector can. These are statistical patterns, and plenty of human writing produces them — technical documentation, legal drafting, academic abstracts and translated text all tend to score high because they are deliberately uniform.
The reverse is also true. A generated draft that a person has edited will usually score low, because editing is exactly what breaks the patterns being measured.
Published research and repeated real-world testing have found commercial AI detectors misclassify text often, and that writing by non-native English speakers is flagged disproportionately. Treat any tool claiming a confident verdict — including the ones charging for it — with suspicion.
Please do not use this score to accuse a student, a writer or a job applicant of anything. It is a writing-quality signal. If the prose is uniform, repetitive and full of stock phrases, that is worth fixing regardless of who or what produced it.
How to read the score
Below 40: few AI signals. The passage varies its sentence lengths and avoids stock phrasing.
40 to 64: mixed signals. Common in AI drafts that a person has edited, and in formal human writing that is naturally uniform.
65 and above: strong AI signals. Several patterns line up with generated text — but so does a well-written policy document.
The breakdown matters more than the total. A high score driven entirely by em-dashes means something quite different from one driven by flat sentence rhythm and forty stock phrases.
Is my text sent anywhere?
No. Every calculation runs in JavaScript on your own device. There is no API call, no model, no server, and nothing is uploaded, stored or logged.
That is also why it works offline once the page has loaded, and why it returns a result instantly rather than after a queue.
Keep going
Related tools
More utilities that pair well with this one.
FAQ
Questions, answered
Everything you need to know about the AI Content Detector.
It accurately measures the patterns it describes, and those patterns correlate with generated text. It cannot accurately determine authorship, and neither can any other detector. Uniform, repetitive prose scores high whether a person or a model wrote it.