Tevfik Darıyemez
Atatürk University, Turkey

Published : 1 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 1 Documents
Search

An Exploratory Analysis of ChatGPT and Gemini Responses on False Positives in AI-Generated Text Detection Ahmet Erdost Yastıbaş; Ali Dincer; Tevfik Darıyemez
Online Learning In Educational Research (OLER) Vol. 6 No. 3 (2026): Online Learning in Educational Research
Publisher : CV FOUNDAE

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.58524/oler.v6i3.1284

Abstract

The increasing use of AI tools has raised concerns about the mislabeling of human-written texts as AI-generated (i.e., false positives), potentially leading to unfair academic decisions in education settings. This study aimed to explore AI-based explanations for this misclassification and develop strategies that L2 teachers can use to reduce them. Adopting a qualitative framework, a textual corpus was created from responses by ChatGPT and Gemini to two prompts about mislabeling texts as AI-generated. The corpus comprised 4 responses totaling 2024 words. It was thematically analyzed. The findings showed that, according to ChatGPT and Gemini, there are five overarching themes regarding possible reasons: linguistic monotony, conventional writing patterns, bias related to non-native English writers, overly polished writing, and limitations of AI detection systems. According to the findings, ChatGPT and Gemini recommended three strategies for L2 teachers to minimize false positives: trace evidence of authors’ writing processes, customize their writing assessments, and use verification techniques in decision-making. Overall, this study contributes to the literature by providing a structured synthesis of AI-generated explanations of false positives and identifying pedagogically relevant strategies for L2 writing assessment, while emphasizing the importance of critically interpreting AI detection results.