AI, Turnitin, and Academic Integrity: Quick Reminders
Entering into another semester, many of us are giving extra attention to academic integrity in our courses. With the rapid rise of generative-AI tools — capable of producing polished essays, summaries, and explanations — it’s especially important to understand how AI use intersects with academic misconduct and how to use detection tools like Turnitin responsibly.
This post offers a concise overview of what instructors should know, how Turnitin’s AI-writing detection works, and how to support student integrity.
A Note on the Temptations of AI
Students may be tempted to use AI tools to complete assignments or large portions of final papers, as AI tools are readily available. However, misconduct often stems from pressure, confusion, or time constraints — not malice — which means clear expectations and strong assignment and assessment design matter. As final assessments come in, it’s essential to be prepared not only to identify potential misuse but also to respond in ways that are fair, consistent, and supportive of student learning.
How Turnitin’s AI-Writing Detection Works (and Its Limitations)
Turnitin’s AI-writing detection appears as an additional indicator within the Similarity Report. It analyzes qualifying prose and generates an estimate of how much text is likely to be AI-written or AI-paraphrased.
What Turnitin Does
- Breaks the submission into overlapping chunks for analysis.
- Scores each sentence for likelihood of being AI-generated.
- Highlights text believed to be AI-written (cyan) or AI-paraphrased (purple).
- Aggregates this into an overall AI percentage score for qualifying text.
What “Qualifying Text” Means
Turnitin only analyzes long-form prose. It does not reliably assess:
- bullet points
- outlines
- poetry
- nonstandard writing
- code
- extremely short submissions
This means the AI percentage may not represent the entire document.
Important: False Positives Are High Under 20%
Turnitin explicitly cautions that its false-positive rate is significantly higher when the AI percentage is below 20%.
In practice:
- Scores under 20% should not be interpreted as evidence of AI-generated writing. Low scores should be considered “background noise,” not indicators of misconduct.
- These low percentages can easily come from normal human writing, especially writing that is formal, concise, or produced by multilingual students.
- Turnitin reports a false-positive rate of under 1% only when a document contains at least ~20% AI-generated text. Below that threshold, reliability drops substantially.
- Even high scores require conversation and supporting evidence. Turnitin stresses that the AI indicator is not proof of cheating.
Using Turnitin Responsibly
1. Treat the AI score as a conversation starter, not a verdict. If the score is high and the writing seems inconsistent with prior work, consider:
- Asking the student about their writing process
- Requesting prior drafts or notes
- Discussing the assignment expectations
2. Look at the writing itself. Indicators that may warrant follow-up:
- Sudden change in tone/voice
- Very generic or vague content
- Factual inaccuracies that match common AI errors
- Citations that don’t lead to real sources or materials not from the course
3. Combine evidence, don’t rely on detection scores alone. AI detection is one data point, not confirmation. Chapter 14 requires:
- multiple forms of evidence
- patterns of concern
- student conversation or documentation
4. Provide transparent expectations to students. Students are less likely to misuse AI when expectations are clear, repeated, assignment-specific, and supported by examples of allowed vs. not allowed use. A reminder now, before finals, can prevent confusion later.
Final Thoughts
Generative AI isn’t going away — and neither is the need for academic integrity. Turnitin’s AI-writing detection can be a helpful tool, but it must be used thoughtfully and ethically.
The most important takeaways:
- AI detection percentages below 20% are not reliable indicators of misuse.
- AI detection should never be the sole evidence in a misconduct case.
- Faculty judgment, assignment design, process-based writing, and open communication are our strongest tools.
CATLST is here to help! Please contact us if you have questions or concerns specific to your course.