A detector can achieve high aggregate accuracy while producing materially different over-detection rates across author groups¶

Claim · All claims
Evidence · 1 study · review r3 · q3
Currency · newest study 2026 · checked against its sources 2026-10-20 (automated, 1 of 1 entry) · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q3 i? GPTZero achieved 97.22% aggregate accuracy on scholarly abstracts while its over-detection rate in the AI-assisted readability condition was 25% for non-native-authored abstracts versus 11% for native-authored abstracts. → Victor Angelier 2026

Evidence¶

Victor Angelier 2026¶

Victor Angelier. (2026). The pitfalls of AI detection in academic writing: bias, false positives, and the need for inclusive assessment. AI and Ethics. https://doi.org/10.1007/s43681-026-01376-w

q3 · i? · review · r3

The review reports Pratama's study of scholarly abstracts, in which "97.22% aggregate accuracy" coexisted with group-level over-detection rates of 25% versus 11% and mean AI scores of 44.61% and 30.68%. No standardised effect size is printed.

"In one scenario GPTZero achieved 97.22% aggregate accuracy, yet in the AI-assisted readability condition its over-detection rate was 25% for non-native-authored abstracts compared with 11% for native-authored abstracts, with mean AI scores of 44.61% and 30.68% respectively."

Discussion¶

Related Claims¶

• AI-detection tools produce inequitable outcomes, disadvantaging non-native English speakers through false positives — a broader claim this one bears on
• AI-text detectors misclassify authentic TOEFL essays by non-native English speakers at a mean false-positive rate of 61.3% across seven detectors, far more often than native-speaker essays — a narrower finding that bears on this claim
• AI language polishing alone can flip human-authored scholarly texts from human-classified to AI-classified — related
• Independent evaluation found commercial AI-text detectors neither sufficiently accurate nor sufficiently reliable for high-stakes use — related
• Universities disabling AI detection cite false-positive risk at scale, and detector influence persists in individual grading, journal screening, and institutions without policy revision — related
• Originality showed a borderline non-significant trend toward higher accuracy on professional than EFL student writing, while Turnitin showed no significant difference — related
• Light AI-assisted edits to human-written abstracts are flagged by commercial detectors at 64–80% (Pangram) and 38–49% (GPTZero) rates — related
• Detector flag rates on 2023–2025 original abstracts vary sharply by domain, with non-STEM rates far above STEM — related
• AI-content detection tools achieve only marginally higher success than human reviewers, around 80% versus 78.4% — related
• AI-generated-text detectors are reported as ineffective against current sophisticated language models, and students can alter text to evade detection — related
• Blinded review identified 68% of ChatGPT-generated abstracts as chatbot-produced while 14% of authentic abstracts were misclassified as chatbot-generated — related
• AI detectors using perplexity disproportionately misclassify non-native English writers, and the mechanism plausibly extends to other low-perplexity writers — related
• Educators run more proctored AI-free assessments conditionally on compromised outcomes, and reject AI detectors based on classroom experience of false positives — related
• Both detectors achieved substantially lower accuracy on scientific writing than on humanities texts — related