AI Detection Tools & Academic Integrity

In April 2023, Turnitin released an AI-writing detection feature built to flag text generated by tools like ChatGPT. Campus ran a pilot to test it. Based on those results, along with a growing body of outside research, UC Berkeley opted out of using Turnitin's AI detection tool. More detail on the pilot and that decision is in Availability of Turnitin Artificial Intelligence Detection for UC Berkeley.

What the research says about detection tools

A widely-cited 2023 study, GPT Detectors Are Biased Against Non-Native English Writers, tested seven commonly used GPT detectors against real TOEFL essays and found that more than half of non-native English writing samples were misclassified as AI-generated, while native-English samples were identified accurately almost every time. A more recent 2025 study, The Accuracy-Bias Trade-Offs in AI Text Detection Tools, tested newer detectors against more current AI models and found the same pattern holds. It also found detectors struggle even more with text that's been AI-assisted rather than fully AI-generated, meaning a student who used AI to polish their own writing is especially hard for these tools to classify accurately.

If AI use comes up in your course

OpenAI's guidance for educators, How Can Educators Respond to Students Presenting AI-Generated Content as Their Own?, describes a few approaches instructors have used when this question comes up, including:

• Asking a student to walk through their process or share their chat history, as a starting point for a conversation rather than a conclusion
• Treating the moment as a chance to build AI literacy, since many students are genuinely unsure where the line is
• Weighing that detectors can misfire in both directions, which some instructors factor in alongside whatever else they know about the situation

Additional resources

• Academic Misconduct — Center for Student Conduct. The official definitions instructors and students are bound by, covering cheating, plagiarism, and fabrication under the Code of Student Conduct.
• Graduate Student Academic Integrity — Graduate Division. Addresses GenAI concerns in graduate coursework directly, and outlines a different process than undergraduate cases: faculty are asked to loop in their chair or dean and the Director of Graduate Academic Conduct and Climate before attempting to resolve a case.
• Academic Honesty — College Writing Programs. Offers a clear, concrete definition worth knowing: using a tool like ChatGPT or Grammarly to write part of an assignment without disclosing it counts as plagiarism under their policy.
• Berkeley Law's Generative AI Policy — An example of a stricter departmental approach to GenAI, for instructors curious what that can look like in practice at Berkeley.