The Short Version
- Set clear expectations. You decide how AI may be used in your course. State it in your syllabus and in each assignment.
- Don't rely on detectors alone. AI detection tools like GPTZero and Turnitin's AI indicator can produce false positives. They are not, on their own, evidence of misconduct.
- If you suspect misconduct, gather supporting artifacts, talk with the student directly, and focus on next steps for their learning.
Want to share guidance with your students? Link them to Academic Integrity for Students. It explains the policy options from their side, how to disclose AI use, and what to do when they're unsure.
Setting Clear Expectations
The most important thing you can do is tell students clearly how AI may be used in your course. How much AI to allow is your teaching decision. Limiting AI and embracing it are both valid choices, depending on what your students need to learn.
Syllabus Statements You Can Adapt
The District AI Task Force developed these statements. Start with the opening statement, then choose the option that fits your course. Adapt the wording as needed.
In this class, all work submitted must be your own. The use of generative AI tools will be considered academic misconduct (see Administrative Regulation 2.3.11 1.B(b)) and will be treated as such. If you are unsure if the tool or website you are using is a generative AI tool, please contact the instructor for further clarification before using the tool or website.
There are situations and contexts within this course where you may be permitted to use generative AI tools. In these cases, specific guidelines will be provided in the assignment details. If you are unsure if the tool or website you are using is a generative AI tool or if it is permitted on a specific assignment, please contact the instructor for further clarification before submitting your work.
In this class, you are permitted to use generative AI tools to complete your assignments. When submitting work that incorporates generative AI content, please clearly indicate what content was generated by AI tools. In such cases, no more than [set your percentage]% of your work should be generated by a generative AI tool. If any part of this is confusing or uncertain, please reach out to me for a conversation before submitting your work.
If you use Option 3, set your own percentage limit, or replace it with another limit that fits your course.
Assignment-Level Guidance
A syllabus statement sets the default. Assignment instructions are where students look when they decide what to do. This matters most under Option 2. For each assignment, consider saying:
- Whether AI is allowed. Say so even when it isn't, so students don't have to guess.
- For which parts. For example, brainstorming yes, drafting no; or feedback on a draft yes, rewriting no.
- How to disclose it. A short note at the end, a shared chat link, or a process reflection.
- Why. One sentence connecting the rule to the skill the assignment builds helps students follow it.
Talking with Students about AI
Many integrity problems start with confusion, not bad intent. Students take several classes with different AI rules, and many AI features are now built into tools they already use. A short conversation early in the term can prevent problems later.
- Go over your policy in class or in a Canvas/RioLearn announcement, not only in the syllabus.
- Explain your reasoning. Students are more likely to follow a rule when they understand the skill it protects.
- Name the gray areas. Say whether grammar checkers, translation tools, and AI features in apps like Word, Google Docs, or Canvas count as AI in your course.
- Make asking easy. Tell students it's always okay to ask before using a tool, and that asking won't count against them.
- Share the student guide. Academic Integrity for Students is written for students and is designed for you to link from your syllabus or Canvas course.
Understanding AI Detection Tools
Tools that claim to detect AI-generated writing include GPTZero, Turnitin's AI writing indicator, ZeroGPT, Originality.ai, and Winston AI. Most measure at least two qualities:
- Perplexity: How predictable the word choices are. AI writes by choosing likely next words, so its text tends to be predictable.
- Burstiness: How much sentence length and structure vary. AI tends to produce more uniform sentences than people do.
- Repetition and formatting: Overuse of transition words, rigid paragraph structure, repeated phrasing patterns.
- Semantic uniformity: The flow of logic and ideas, emotional nuance, and whether a personal voice or opinion is present.
These are indirect signals, not proof. Plagiarism checkers match student writing to existing text. AI detectors can only estimate how likely it is that text came from AI.
Why Detectors Aren't Enough
- The makers say so. Turnitin states that its AI writing indicator “is best used to inform the educator’s judgment, not to be the sole measure of academic integrity.” GPTZero states its results “should not be used to punish students.”
- False positives happen. Human writing can be flagged as AI, especially writing that is formal, formulaic, or heavily edited.
- Some students are flagged more often. Research has found much higher false positive rates for non-native English writers. In one study, over half of non-native English writing samples were misclassified as AI-generated, while native samples were classified almost perfectly.
Relying on a detector alone risks falsely accusing a student, with serious consequences for them. Clear expectations do more to protect integrity than detection does.
AI Watermarking
You may hear that AI content is now "watermarked," so misuse will be easy to spot. Watermarking helps with transparency, but it has many of the same limits as detection tools.
What's changing. The European Union's AI Act (Article 50) requires providers of generative AI to mark AI-generated text, images, audio, and video in a machine-readable way that can be detected. The rules apply from August 2, 2026, with a grace period to December 2, 2026 for existing systems. A Code of Practice published in June 2026 recommends two layers:
- Content credentials: signed information attached to a file, such as C2PA metadata on an image, recording that AI made or changed it.
- Invisible watermarks: patterns hidden in the content itself, such as in an image's pixels or an AI's word choices. Google's SynthID is one example.
The law applies to AI companies offering tools in the EU. Many major AI tools are used worldwide, though, so faculty and students here are likely to run into these markings.
Why a watermark isn't proof of misconduct:
- It can't tell editing from writing. A student may write their own draft and then use an AI tool to polish it. The EU law exempts tools that only do standard editing, like grammar correction, without substantially changing the text. But where "editing" ends and "substantially changing" begins isn't clearly defined, and each tool decides for itself. A student's own work could come back with a watermark signal, and there's no reliable way to tell whether AI helped edit it or wrote it.
- Mixed work gives mixed results. Most student work that involves AI is part human, part AI. Detectors like SynthID can return "uncertain" instead of a yes or no.
- Text watermarks are fragile. Google says SynthID's text detection "can be greatly reduced when an AI-generated text is thoroughly rewritten, or translated to another language." It is also "less effective on factual responses" (Google).
- Image markings are easy to lose. Metadata disappears with a screenshot or re-save. A student's own photo edited with a built-in AI feature, such as object removal, may be marked as AI-altered.
- No watermark doesn't mean no AI. Not every tool adds one. Open-source models and tools outside the EU may not, and rewriting can remove the signal.
Treat a watermark or content credential like a detector result: one piece of information that can start a conversation. It is not evidence on its own.
Additional resource: The EU AI Act's Transparency Rules: A Practical Guide to Article 50
If You Suspect Academic Misconduct
A detector result can be one piece of information, but not the only one. Other signs include:
- Sources that don't exist, or major factual errors (hallucinations)
- Outdated information presented as current
- Writing style that doesn't match the student's earlier work
- A sudden change in voice, tone, or word choice
- The AI identifying itself in the text (for example, “As a large language model…”)
- Mixed fonts or formatting from copying and pasting
Share your concerns and the artifacts you gathered. Give the student the benefit of the doubt and time to explain. Ask about their process, the sources they used, and the tools they used. Students can walk you through their thinking, drafts, or version history.
Keep the conversation focused on solutions. Could the student redo the assignment for full or partial credit? Would an additional assignment help them reflect on their choices? Even if the outcome is a zero, help the student understand how it affects their course grade and academic progress.
Student Due Process
Students have a right to due process. These processes are described in:
- Residential Faculty Agreement 20.4.6 (Final Course Grade Appeal)
- Administrative Regulation Appendix S-6 (Instructional Grievance Process)
- Administrative Regulation 2.3.11 (Academic Misconduct)
- Administrative Regulation 2.5.2 (Student Code of Conduct)
Attribution: Adapted from Guidance for Maricopa Faculty Regarding Generative AI Tools and Academic Misconduct and Syllabus Statements Regarding Generative AI, Maricopa Community College District AI Task Force, licensed under CC BY 4.0.