Is Scribbr's AI detector accurate?
Scribbr sells proofreading and gives away a detector. Proofreading raises AI scores, which makes the pairing worth thinking about before you read either result.
Scribbr is an academic services company, mainly proofreading and editing, with a free AI checker alongside. The free tool is a front door to the paid service, which is a normal arrangement and worth knowing when reading a number from it.
The accuracy question, properly asked
A detector makes two errors. It misses generated text, which costs nobody anything that afternoon, or it flags writing a person produced, which arrives attached to their work.
Accuracy averages them. The figure that describes your exposure is the wrong-flag rate on human writing.
Turnitin publishes 4%. Ours is 0.4%, which is 99.6% specificity, measured across 15,900 documents of real academic writing.
Whatever number a tool advertises, check which of the two claims it is before comparing it to anything.
The pairing worth understanding
Proofreading and editing make prose more regular. Sentences get tightened toward a consistent length, constructions get standardised, unevenness gets removed. That is the service working correctly.
Regularity is what detectors read. Our corpus puts human academic writing at a within-document sentence-length standard deviation of 8.7 words and generated text at 6.4, and each editing pass moves a document toward the lower figure.
So a professionally edited thesis chapter can score higher than the draft that went in, without a model having touched it. This applies to any editing service, and it is one of the most common routes to a wrong flag. See does Grammarly get flagged as AI and why good writing gets flagged more.
For students writing in English as a second language, who use editing services most, the two effects stack. Published evaluations keep finding elevated wrong-flag rates for that group already. See why non-native English writing gets flagged and international students and AI detection.
What a free academic checker is good for
Direction, not a verdict, and only if you use it on enough text.
Run the whole document. Turnitin's own figures show what length is worth: 4% wrong on sentences against under 1% on documents. A cap that forces you to check a chapter in fragments degrades the answer. See why detectors are unreliable on short text.
Read per passage. A single percentage over a chapter is an average that hides which section carries the signal. In most documents two or three paragraphs carry all of it.
Check what happens to the text. Worth knowing before pasting an unpublished chapter into anything. See what happens to text you paste.
A second measurement
Ours is free and unlimited with no account, takes .pdf, .docx and .tex, has no word cap, scores every paragraph separately, and never stores your text. Our wrong-flag rate is measured on exactly this register of writing.