A Comparative Analysis of AI Detection Systems

The increasing sophistication of generative artificial intelligence (AI) and large language models (LLMs) for example, GPT-5 and Gemini is presenting similar challenges in academic integrity and discrimination between AI versus human writing. This research involved a comparative study of five AI content identification systems QuillBot, Writer.com, GPTZero, CrossPlag, and Copyleaks by evaluating the systems with 45 writing samples composed of verified human writing and AI-generated writing. The evaluation of the results was undertaken with generalized statistical measures of sensitivity, specificity, and predictive values. The results suggested, while GPTZero identified with the most balanced accuracy (93%) and CrossPlag identified slightly lower (88%), they did identify students significantly different in specificity levels and additional systems were highly biased toward human writing over AI-generated writing. The findings suggest that as LLMs continue to evolve, modifications to current system may not serve reliably as data in our academic evaluation due to linguistically and semantically based changes. Overall, the research points to the need for hybrid based models, retraining AI content identification system on new data, and the importance of human review, consensus, and engagement in the educational context to ensure fairness and transparency.