Skip to Main Content

AI and Teaching: Detecting AI

Detection Tool Limitations

Before using any AI detection tool, carefully consider their limitations in order to make an informed decision.

  1. False positives and false negatives: AI detectors may produce inaccurate results. They might incorrectly flag legitimate student work as being written by an AI or fail to detect work that was actually generated by an AI. Relying solely on AI detectors without human judgment can lead to unfair assessments.

  2. Bias: AI detectors are notoriously unreliable when evaluating text written by non-native English speakers. 

  3. Limited training data: AI detectors are trained on specific datasets, and their effectiveness depends on the quality and diversity of the data they were trained on. If the AI detector hasn't been trained extensively on a variety of samples, it may not be reliable or produce accurate results.

  4. Adaptability of AI models: AI language models are constantly evolving and improving. New versions and variations are released frequently and AI detectors cannot keep up. An AI detector trained on one version of an AI model might not be effective in detecting newer or modified versions.

Instead of relying solely on AI detectors, we recommend faculty adopt a holistic approach to evaluating student work. This includes considering factors like writing style, coherence, depth of understanding, and engagement in class discussions. Open communication and establishing clear guidelines on the acceptable use of AI tools can also help create a fair and supportive learning environment.

Detection Tools

This is a small selection of openly available AI detection tools. Before using these tools, read the "Limitations" section on this tab, and read the terms of service for the tools you decide to use. Most detection tools require you to own the copyright of, or have permission to use, text that you input. Our recommendation is that you do not use these tools with student work without obtaining their permission. 

  • AI Text Classifier: Freely provided by OpenAI (the company behind ChatGPT and DALL-E), this tool predicts how likely it is that text was created by AI. 

  • GPTZero: The free subscription limits you to 5,000 characters, paid subscriptions start at $9.99/month. Requires you to state that you "are the creator and owner of or have the necessary licenses" of the content that you enter. 

  • GPT DetectorProvided by Writefull, this free detector provides a probability score for the text you enter. There is a daily quota of 30 documents per day.. 

Sources Consulted

D'Agostino, Susan. (January 2023). "AI Writing Detection: A Losing Battle Worth Fighting" in Instide Higher Ed.

Myers, Andrew. (May 2023). "AI-Detectors Biased Against Non-Native English Writers" from Stanford University's Institute for Human-Centered AI.

Warkins, Marc. (January 2023). "It’s Time to Step off the AI Panic Carousel Before We Harm our Students: We should not upload student work to AI text detectors without their consent." in Rhetorica.