Research shows our AI scanner shows no statistically significant bias against English Language Learners

1. In response to customer feedback and papers claiming that AI detector tools are biased against writers whose first language is not English, we expanded our false positive evaluation to include writing samples of English Language Learners (ELL). We tested another nearly 2,000 writing samples of ELL writers.
2. We found that in documents meeting the 300-word count requirement, ELL writers received a 0.014 false positive rate, and native English writers received a 0.013.
3. This means that there is no statistically significant bias against non-native English speakers.

Independent research shows the “Turnitin [AI writing detector] achieved very high accuracy”

1. Two of the 16 detectors, Turnitin and Copyleaks, correctly identified the AI- or human-generated status of all 126 documents, with no incorrect or uncertain responses.
2. Three AI text detectors — Turnitin, Originality, and Copyleaks — have very high accuracy with all three sets of documents examined for this study: GPT-3.5 papers, GPT-4 papers, and human-generated papers.
3. Of the top three detectors identified in this investigation, Turnitin “achieved very high accuracy” in all five previous evaluations. Copyleaks, included in four earlier analyses, performed well in three of them.