# ExaGPT: Example-Based Machine-Generated Text Detection for Human Interpretability Source: [ExaGPT: Example-Based Machine-Generated Text Detection for Human Interpretability]() Collection index: [llms.txt]() Data updated: 2026-05-07T05:00:12.324Z Authors: Ryuto Koike, Masahiro Kaneko, Ayana Niwa, Preslav Nakov, Naoaki Okazaki Published: 2025-02-17T01:15:07.000Z ## Abstract Detecting texts generated by Large Language Models (LLMs) could cause grave mistakes due to incorrect decisions, such as undermining students' academic dignity. LLM text detection thus needs to ensure the interpretability of the decision, which can help users judge how reliably correct its prediction is. When humans verify whether a text is human-written or LLM-generated, they intuitively investigate which of them it shares more similar spans with. However, existing interpretable detectors are not aligned with the human decision-making process and fail to offer evidence that users easily understand. To bridge this gap, we introduce ExaGPT, an interpretable detection approach grounded in the human decision-making process for verifying the origin of a text. ExaGPT identifies a text by checking whether it shares more similar spans with human-written vs. with LLM-generated texts from a datastore. This approach can provide similar span examples that contribute to the decision for each span in the text as evidence. Our human evaluation demonstrates that providing similar span examples contributes more effectively to judging the correctness of the decision than existing interpretable methods. Moreover, extensive experiments in four domains and three generators show that ExaGPT massively outperforms prior interpretable detectors by up to +37.0 points of accuracy at a false positive rate of 1%. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [AI-generated text detection on M4GT-Bench arXiv (test)](): AI-generated text detection; M4GT-Bench arXiv (test); Accuracy: 98.7. - [LLM Detection Interpretability on M4 (test)](): LLM Detection Interpretability; M4 (test); Accuracy (Human Judgements): 61.5. - [AI-generated text detection on M4GT-Bench Reddit (test)](): AI-generated text detection; M4GT-Bench Reddit (test); Accuracy: 91.1. - [Machine-generated text detection on Wikipedia M4 benchmark (test)](): Machine-generated text detection; Wikipedia M4 benchmark (test); AUROC: 98.6. - [Machine-generated text detection on WikiHow M4 benchmark (test)](): Machine-generated text detection; WikiHow M4 benchmark (test); AUROC: 99.5. - [Machine-Generated Text Detection on Average M4 benchmark (test)](): Machine-generated text detection; Average M4 benchmark (test); AUROC: 99.2. ## Original sources and implementation [Original paper]() [PDF]()