The short answer
It looks strong on clean AI text and weaker on mixed or edited text. Originality.ai claims 99%+ for models 5.0 and 5.0 Max. A 2026 peer-reviewed study of 192 texts measured an average accuracy of 0.69 across four kinds of text, with near-zero recall on texts that were half student, half AI. A preprint on Arabic articles found 96% on clean text. That study used earlier versions, as the model launched in October 2026 is too new to have been tested in the papers we read.
What Originality.ai claims
Its accuracy page, dated 25 October 2026, gives figures by model:
| Model | Its stated accuracy | Its stated false positive rate |
|---|---|---|
| 5.0 and 5.0 Max (October 2026) | 99%+ | Not stated in the text we read |
| Turbo 3.0.2 (September 2025) | 99%+ | 1.5% |
| Lite 1.0.2 (September 2025) | 99% | 0.5% |
| Academic 0.0.5 (September 2025) | 99%+ | Under 1% |
| Multilingual 2.0.0 (May 2025) | 97.8% across 30 languages | 2.4% |
The page says several of these models were due to retire in 2026. It also makes two points that are often left out of reviews. It says no AI detection tool will ever be 100% perfect. And it says the company does not believe AI detection scores alone should be used for academic honesty purposes and disciplinary action, because even a low false positive rate is too high for that.
What published tests found
| Study | Texts | Originality.ai result |
|---|---|---|
| Hadra et al., 2026 (peer reviewed) | 192 texts: English learners, professional, AI, and half-and-half, 48 of each | 0.69 macro-average accuracy and 0.60 recall. Near-zero recall on mixed text. All professional texts correct, several learner texts wrong. |
| Almohaimeed et al., 2025 (preprint) | Human and AI articles in Arabic | 96% accuracy, 8% of human text flagged. On human text lightly polished by some AI models, the correct rate fell as low as 12%. |
The first study compared Originality.ai with Turnitin, which scored 0.61 on the same measure. Its authors conclude that neither is reliable enough to be the sole basis for a misconduct decision. Macro-average accuracy gives each of the four groups equal weight, so it is not the same thing as the company’s figure. The second paper’s point is the one to note: light AI polishing of your own writing can read as AI.
Why a Originality.ai score is not proof
Publishers often use a detector to screen freelance work. The same limits apply there: a score is a reason to ask for drafts or notes, not grounds to reject a writer on its own.
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.
What to do with a result
- If you are a writer. Keep drafts and version history, and agree in advance what your client will check. For AI text and search, see is AI content bad for SEO?
- If you are a student. See how to prove you didn’t use AI.
Our AI detector gives a separate estimate for every sentence. It can disagree with Originality.ai. We publish no accuracy figure for it.
Sources
- Originality.ai: AI detection accuracy page, read 1 November 2026.
- Studies: Hadra et al. and Almohaimeed et al.