Generative AI works by predicting text based on a trained model and some context provided through input (e.g. prompts). The tools are designed to draw from the information on which they are trained, therefore it is virtually impossible to reliably distinguish the AI-generated text from that which might be written by a human author. Patterns and phrases that typically appear in AI-generated text may raise suspicion; however, these alone are not reliable evidence that the tools were actually used.
AI detectors use statistical pattern recognition and machine learning classifiers to estimate the mathematical probability that a piece of writing was generated by a language model (Bassett et al., 2026). Many GenAI detection tools produce reports that note how much of a submission could be “AI-Generated” and may pinpoint sections of concern. However, it’s the instructor’s responsibility to discern what constitutes a violation of academic integrity within the academic misconduct policy.
Considerations before using GenAI detection tools
There are many tools available that attempt to detect GenAI use in writing (e.g. iThenticate, GPTZero, StrikePlagiarism). Companies like GPTZero cite the limitations of their applications and encourage educators to use the system to “flag situations in which a conversation can be started”.
These tools are not recommended for use in identifying academic misconduct due to the fundamental lack of reliability of such detection and other concerns, including:
- Evidence of Cultural Bias: research suggests that students whose native language is not English are often falsely accused of using Generative AI by such tools.
- Infringement on student intellectual property (IP) rights: detection tools are commercial products that generally use submitted work to train their models; submitting student work without permission would be a violation of the student’s IP rights.
- Violation of privacy policies: sharing of student personal information with unauthorized outside agencies is not permitted under the university’s Privacy Policy.
- Unclear Sources: because the content is generated, the tool cannot identify the originating source of the suspected content as hard evidence.
Reference:
- Bassett, M. A., Bradshaw, W., Bornsztejn, H., Hogg, A., Murdoch, K., Pearce, B., & Webber, C. (2026). Heads we win, tails you lose: AI detectors in education. Journal of Higher Education Policy and Management, 1–16. https://doi.org/10.1080/1360080X.2026.2622146
Originally Published: August 17, 2026
