Turnitin is designed to encourage ethical writing, and honesty by detecting traditional and AI-assisted plagiarism. While Similarity Report checks for text matches from traditional sources, the AI checker detects AI generated text.

Turnitin's Similarity Report and AI Checker

With the growing presence of AI in our classrooms, AI checker has become a critical tool for instructors to assess and determine the authenticity of students' work. Let's take a more in-depth look into how the AI checker works.

What do the different AI Checker states indicate?

For a submission to generate an AI percentage, the submission should meet the following requirements:

- File size less than 100 MB
- Between 300 to 30,000 words of prose text in a long-form writing format
- In a supported format: .docx, .pdf, .txt, .rtf
- In a supported language: English, Spanish and Japanese

If the text does not meet these criteria, the AI checker will not display a number.

AI checker with no number

If the percentage is lower than 20%, the AI checker will display an asterisk with no percentage.

AI number shows an asterisk and percentage,

For submissions between 20% to 100%, AI checker will display a number.

AI number is 75%

How does AI Checker work?

A submission is broken down into small segments that are run against Turnitin's AI detection model to determine whether it is written by a human or by AI. Using the average score of all the segments within the document, the model then generates an overall prediction of how much text in the submission is generated by AI. The percentage detected by AI checker indicates the text that might be generated by AI, or it could be generated by AI and modified further by an AI paraphraser or an AI bypasser tool.

What is the reliability of AI Detector?

While Turnitin AI model predicts the numbers with 99% accuracy, there can be room for errors and false positives, where Turnitin may incorrectly identify fully human-written text as AI-generated.

Reasons for 1% false positive:

- Repetitive writing - Content without a lot of structural variation with repetitive text that has been paraphrased without developing new ideas can be flagged as AI generated.
- Shorter documents - For a document of only a few hundred words, the prediction is mostly "all or nothing" because the detector is predicting on single segment without the overlap of human-generated and AI-generated text.
- Lists, outlines, short questions, poems or poetry may get flagged as AI generated.

How does Turnitin ensure that the false positives rate remains less than 1%?

Turnitin regularly updates their AI detection model by performing periodic tests on large data sets of academic papers that were written before the release of ChatGPT.

What are UH guidelines regarding AI use by students?

Turnitin alerting users to review the AI number

A high AI number does not automatically imply plagiarism. Rather, it provides data to make informed decisions. Instructors should review the nature of the assignment, students' past performance, and their writing style before making a decision. Instructors should also state the AI policy clearly in the syllabus. Here are Sample AI Syllabus Statements for instructor use.

What actions can I take if unauthorized use of AI is suspected?

Any unauthorized use of AI is considered a violation of UH Academic Honesty Policy. Instructors are encouraged to engage in a dialogue with the student and understand their writing process, before making a decision.

What could I do if I have questions about the AI checker?

If you find AI written documents that Turnitin has missed or notice that student work has been flagged as AI generated, contact us at fdis@uh.edu. You can also provide feedback directly to Turnitin.