Is the QuillBot AI Detector Accurate?
QuillBot cites a 99% detection rate from a benchmark run it submitted itself. Two small independent studies tested it and disagree about false positives. Here is what each source says and how far to trust the score.
Short answer: QuillBot says its AI Detector has a 99% detection rate, based on a benchmark run it submitted itself. Independent research on it is thin: two small studies, and they disagree about false positives. It is a reasonable free check, but not one to rely on alone, which is also what QuillBot says.
What QuillBot says about its accuracy
- "A 99% detection rate", credited to the RAID benchmark (AI Detector). The RAID leaderboard shows QuillBot detecting 99.2% of AI text at a 1% false positive rate, but QuillBot submitted that run itself: vendors run their own detector on RAID's test set.
- No false positive rate is given as a number. QuillBot says its model leans toward a human verdict when results are unclear.
- Limits it states: at least 80 words (one help article says 40), more reliable above 300 words, 20+ languages, and heavily paraphrased or formulaic human writing is more likely to be misread.
- Its own rule: never make an academic or employment decision on a detector score alone.
What independent studies found
QuillBot's detector was not part of the best-known studies. Liang et al. 2023 tested "Quil.org", a different tool, and Weber-Wulff et al. 2023 used QuillBot only to paraphrase test texts.
| Study | Tested | What it found for the QuillBot AI Detector |
|---|---|---|
| Yeadon et al. 2024 (preprint) | 2024 | Caught all 30 GPT-4 physics essays, but flagged 10 of 30 human essays as AI |
| Alshammari & Rao 2025 (preprint) | March 2025 | No false positives on 49 human answers; scored humanized AI text at 58% on average, against 98% for unedited AI text |
Sources: Yeadon et al., Alshammari & Rao.
A benchmark run by competitor GPTZero reported much lower recall for QuillBot (45%). Because a rival ran it, treat it with the same caution as QuillBot's own figures.
What the evidence adds up to
- Plain AI text: both small studies found it catches most of it.
- Human writing: one study found no false positives, the other flagged a third of human essays.
- Humanized text: detection dropped noticeably in the one study that tested it.
- Scale: two small studies are not enough to confirm or dispute a 99% figure.
FAQ
Is the QuillBot AI Detector accurate? It is likely to catch unedited AI text. Its rate of flagging human writing is not established: the two independent studies disagree.
Is the QuillBot AI Detector free? Yes: up to 1,200 words per scan and 6 scans a day on the free plan (QuillBot). Premium removes the limits.
Can QuillBot detect its own humanizer? QuillBot says its Humanizer does not try to trick AI detectors or promise undetectability (help center), so humanized text can still be flagged.
Should a teacher rely on it? Not alone. QuillBot itself says never to make an academic decision on a detector score by itself.