The short answer
Scribbr’s AI detector is reasonable on clear cases and honest about its limits. Its own page says the free version detects GPT-2, GPT-3 and GPT-3.5 “with average accuracy”, that newer models such as GPT-4 need the Premium version, and that “no AI model on the market can guarantee 100% accuracy, including ours.” In Scribbr’s own research, the best free tool it tested reached 68% accuracy and the best premium tool 84%. Use a Scribbr score as a prompt to read the text closely, not as a verdict.
Get a second opinion on a draft
Scribbr’s own accuracy figures
Scribbr publishes research comparing AI detectors and quotes it on the detector page: “the highest accuracy we found was 84% in a premium tool or 68% in the best free tool.” Two points help read those numbers:
- They are averages over a test set. Your essay may be shorter, edited, mixed or in a second language, and those cases are harder.
- “Accuracy” combines two errors. Missing AI text and flagging human text are both counted. For a student, flagging human text is the costly one, and a single figure does not say how often it happens.
Scribbr also advises scanning longer passages rather than single sentences, which matches how every detector behaves: short texts give unstable scores.
What the result shows
Scribbr gives an AI score from 0% to 100% for how much of the text is likely AI-generated or AI-refined, and labels sections as AI-generated, AI-generated and AI-refined, human-written and AI-refined, or human-written. The “AI-refined” labels are worth noticing: text you wrote and then polished with a grammar or rewriting tool can land there. Whether that is allowed depends on your course rules, not on the detector.
Why a Scribbr score is not proof
Scribbr says it plainly: no AI detector can guarantee 100% accuracy, including its own. Treat a high score the same way in any tool.
Three things hold for any AI detector, this one included:
- It reads the finished text. It does not see how the text was written, so it can only estimate.
- Its figures depend on the test. An accuracy number holds for the texts it was measured on. Your text may be shorter, edited or in another language.
- A small error rate is still real people. Liang and colleagues (2023) found seven detectors, which the paper does not name, flagged 61.3% of TOEFL essays by non-native speakers on average. Detectors have changed since, but plain, predictable writing remains the risk.
If your own writing is flagged
- Don’t rewrite to chase a number. A score is an estimate about the text, not a record of how you wrote it.
- Gather your process. Version history in Google Docs or Word, notes, outlines and sources show how the work came together. See how to prove you didn’t use AI.
- Read the highlighted parts. Definitions, lists and standard academic phrasing are the usual false-positive triggers. Our guide to AI detector false positives explains why.
- Check your course policy. If some AI help was allowed, say what you used. See how to cite ChatGPT.
A second opinion
Detectors disagree, and the disagreement tells you something: if two tools give very different results, the text is in the grey zone where no score should be trusted alone. The Plagiarism Checker Plus AI detector looks for text likely written by large language models such as ChatGPT, Claude, Gemini and Llama. It is free without an account up to 1,000 words, a free account includes 3,000 words a month, and your text is never used to train AI models. For copied text rather than AI writing, use the plagiarism checker.