Detecting use of AI

It is understandable that staff want a simple, reliable way to identify whether student work has been created by AI. Unfortunately, no such tool currently exists. AI detection tools present results that may look authoritative but are not — they cannot determine with any certainty whether content was written by a human or an AI.
AI detection tools do not provide conclusive outcomes. 

Issues with AI detection:

• False positives. Tools regularly flag genuinely student-written work as AI-generated, which can lead to unfair and distressing accusations of misconduct.
• Bias against non-native speakers. Research consistently shows that students who do not have English as their first language are disproportionately and incorrectly flagged by AI detection tools.
• Rapidly changing AI. Generative AI is evolving quickly. Tools trained to detect older AI patterns become obsolete and generate even less reliable results over time.
• No meaningful threshold. A “50% AI probability” score provides no actionable information. Detection tools give the impression of certainty where none exists.

Equity concern: Using AI detection tools against students with English as an additional language, neurodivergent students, or students from widening participation backgrounds risks compounding existing inequalities.

Never submit student work to unapproved third-party tools

Even setting aside accuracy concerns, submitting student work to any external tool that has not been formally approved by UWE is not permitted. This applies whether the tool is free, widely used, or appears to be secure.

• It threatens trust. Students submit their work to UWE, not to commercial platforms. Sharing it elsewhere without consent undermines the staff–student relationship.
• It may constitute a data breach. Student work contains personal data. Passing it to unapproved external services may violate GDPR.
• It threatens intellectual property. Student work belongs to the student. Uploading it to third-party services may expose it to commercial use, training datasets, or public indexing.
• It models poor academic integrity. Asking students to behave with integrity while staff disregard data governance and consent requirements is inconsistent and undermines the broader culture we are trying to build.

What to do if you have concerns about a specific submission: Forward your concerns to your Academic Offence Officer. They are trained to handle these situations appropriately and within institutional policy.

What can I do?

Across our teaching teams there are things we can do to limit inappropriate or accidental use of AI in assignments:

1. Frame appropriate AI use clearly. Be explicit with students about what constitutes appropriate and inappropriate AI use in your module. Clear expectations reduce ambiguity and prevent unintentional misuse. See the Guiding Students at UWE resource for practical guidance.
2. Design AI-resistant assessments where appropriate. Personal reflection, situated tasks, oral assessments, and ongoing portfolios are harder to generate meaningfully using AI. Assessment design is difficult and resource-heavy, but it remains one of the most effective long-term responses to AI misuse.
3. Educate students in academic integrity. We can help students understand not just the rules, but the personal value of doing their own work – what it means to be a good student and the necessary development of skills they will utilise in the workplace.
4. Use follow-up conversations. Where you have concerns about a submission, a one-to-one conversation is often the most effective response. Ask students to talk through their work, explain their reasoning, or revisit a section in discussion.