Leveraging Explainable AI for LLM Text Attribution: Differentiating Human-Written and Multiple LLMs-Generated Text

AI-generated text can be reliably detected: The study achieved 98.5% accuracy in distinguishing between human-written and AI-generated text using machine learning models, significantly outperforming existing commercial tools like GPTZero (78.3% accuracy), suggesting that detection technology is feasible and improving.

Different AI systems leave distinct fingerprints: The research successfully differentiated text generated by five different LLMs (ChatGPT, LLaMA, Google Bard, Claude, and Perplexity), indicating that each AI tool produces text with identifiable characteristics—useful information for academic integrity and content attribution purposes.

Explainable AI reveals why detection works: By using Explainable AI (XAI) techniques, researchers identified specific linguistic features that distinguish AI from human writing, moving beyond "black box" detection to help stakeholders understand what patterns the system is actually recognizing.

Current commercial tools have blind spots: GPTZero failed to classify approximately 4.2% of test samples, while the study's model successfully analyzed the complete dataset, highlighting that existing detection solutions may miss certain AI-generated content—a critical gap for institutions relying on these tools for academic integrity.

Practical applications extend beyond plagiarism: The research addresses multiple responsible AI concerns including skill development (students over-relying on AI writing tools), plagiarism detection, and content attribution—suggesting detection technology could support multiple stakeholder needs in education and content verification.