How Do Professors Check for AI in College Courses

Most professors who check for AI start with Turnitin's AI writing indicator inside their learning management system (LMS), when their campus licenses it. They then confirm with process evidence: version history, earlier drafts, and a conversation with the student. In the guides we publish here at Chalkbox, we describe what each checking tool shows an instructor before anyone uses it to grade. An automated score alone does not serve as conclusive proof of academic misconduct.

What Professors Use to Check for AI

Some campuses license AI detection software and some do not, so the table separates what comes with a campus license from what any instructor can use.

The standard instructional workflow moves in three clear stages: score notification, process verification, and an in-person conversation. Instructors examine the automated alert first, cross-reference the submission with intermediate drafts, and then ask the student to explain specific stylistic choices. You can explore how campus tools evaluate these assignments in our breakdown of the Turnitin AI score.

How Turnitin Evaluates Student Submissions

Turnitin sells AI writing detection as Originality, an add-on for Turnitin Feedback Studio customers, and Turnitin says the indicator works only once the campus licenses and enables it.

According to Turnitin's guide on the AI Writing Report, the model helps educators identify text that might be prepared by a generative AI tool, including word spinners and bypass utilities. In the newer grading interface, instructors select the AI Writing tab in the top menu to view the report. In the classic view, an AI Writing indicator sits in the right-side toolbar, as Turnitin's guide on accessing the AI Writing Report describes. A blue icon means the file was processed, gray means it was not, and an error icon means processing failed.

Turnitin explicitly notes that its metric is different from and independent of the standard similarity score. AI highlights do not appear in the Similarity Report. They appear in a separate AI Writing Report. Furthermore, Turnitin documents concrete constraints on what the system can process:

• The submitted document must contain at least 300 words of prose in long-form writing and must not exceed 30,000 words.
• Supported file types are .docx, .pdf, .txt and .rtf, under 100 MB.
• Supported languages are English, Spanish, Japanese, and Arabic, though only the English detector identifies paraphrasing and bypasser tools.
• Turnitin's detector does not reliably detect AI in non-prose writing, such as computer code, poetry, scripts, bullet lists, tables, or annotated bibliographies.
• For reports generated after July 2024, a score from 1% to 19% shows only an asterisk (*%), with no number or highlights. Turnitin's guide says false positives are more common in that range.

Turnitin states that its detector has a false positive rate of less than 1% for documents containing more than 20% AI-generated content. Turnitin also says it does not make a determination of misconduct, and its guide says the score should not be the sole basis for adverse actions against a student. Some campuses have raised serious concerns over these metrics. For example, Vanderbilt University disabled Turnitin's AI detector after calculating that a 1% false positive rate across about 75,000 papers a year could mislabel around 750 student papers. MIT Sloan Teaching & Learning Technologies says AI detection software has high error rates. MIT Sloan cites a finding that 61% of TOEFL essays by non-native English speakers were flagged as AI, and it recommends clear policies, process statements and assignment design instead.

Turnitin also offers Turnitin Clarity, a paid add-on composition environment where instructors can view writing process playback. However, administrators must note that Clarity is licensed separately, and AI detection features appear within it only if the campus also purchases Turnitin Originality. For a closer look at detector limits across platforms, see our guide on whether AI detectors are accurate.

Checking Code Submissions for Artificial Intelligence

Turnitin's guide says its AI detector does not reliably detect AI in code, so prose detection does little for a programming assignment.

Instructors frequently turn to similarity tools to review code, but these tools do not identify AI origin. For instance, Gradescope's guide on Code Similarity clarifies that its system shows how similar two programs are within a single assignment, rather than detecting plagiarism automatically. Gradescope supports languages like Python, Java, C, C++, and Go, allowing faculty to upload starter code so standard templates are ignored. Gradescope's own pages describe AI-Assisted Answer Groups, which group similar answers for grading. That feature does not detect AI-generated code. You can review similar testing workflows in our guide to Gradescope alternatives.

Similarly, Stanford's MOSS system provides a free similarity service for non-commercial educational use. MOSS compares programs written in languages like C++, Java, Python, and C# to identify structural overlap between student repositories. Stanford's MOSS page says MOSS is not a system for completely automatic plagiarism detection. Someone still has to look at the code to judge why two programs match.

Because similarity tools only show how closely two submissions match, instructors look for alternative evidence when investigating potential AI usage in code:

• Advanced logic structures or libraries that exceed the scope of the course syllabus.
• An inability by the student to explain specific functions during an oral code walkthrough.
• Version control repositories that show massive code drops in a single commit without incremental builds.
• Inconsistent variable naming conventions alternating between different architectural patterns.

Discussion Posts and Timed Quizzes

Routine discussion boards and multiple-choice quizzes operate under different mechanical constraints than term papers. Many discussion posts are shorter than the 300 words of prose Turnitin's AI detector requires, so Turnitin cannot score them.

Inside Canvas, instructors cannot rely on any native scanner. An Instructure Community Coach wrote on the Canvas Community forum in February 2025 that Canvas does not give instructors an AI-generated paper detection feature. An Instructure-hosted case study describes a school adding GPTZero detection to Canvas through K16 Solutions, using an LTI integration. Instead, instructors evaluate discussion posts through qualitative markers, such as generic responses that fail to reference specific lecture comments, excessively formal introductory transitions, or citations to fabricated sources.

For quizzes and exams, multiple-choice questions offer no continuous student writing to inspect. As a result, faculty rely on exam security systems rather than text detection. Respondus LockDown Browser locks the testing environment inside an LMS such as Canvas, preventing students from opening external windows, taking screen captures, or copying and pasting prompt text. With Respondus Monitor, students record themselves on webcam during an online exam, and Respondus says flagged events and proctoring results are then available for the instructor to review. Learn practical strategies for managing classroom verification without software in our guide to how to catch AI without a detector.

Managing Suspected AI Submissions in College Courses

When an automated tool flags a paper or an instructor spots anomalous phrasing, the next steps come from your institution's academic integrity policy. Your department chair or integrity office can give you a copy.

The typical campus review process follows these operational stages:

1. Evidence Collection: The instructor gathers the automated report, notes specific passages containing unverified claims, and retrieves document history if collaborative drafting tools were used. For a Google Docs file, the instructor needs edit access, because Google's help page says you need permission to edit a file to browse its earlier versions. The file owner can permanently delete version history, so a missing history is not proof on its own.
2. Student Consultation: The professor invites the student to an informal meeting or office hours to discuss the paper's thesis. The instructor asks the student to explain their drafting process, define specialized vocabulary used in the text, and walk through their primary sources.
3. Departmental or Committee Review: If questions remain after the conversation, the instructor forwards the documentation to the campus academic integrity board or department chair. The committee reviews the assignment prompt, the student's explanation, and the drafting artifacts before determining whether a violation occurred.

Detection tools can mislabel human writing, so know how the process looks from the student's side. For an overview of the student resolution process, see our guide on being falsely accused of using AI. Instructors seeking broader evaluation methods can also review our comparison of an AI detector for teachers alongside our guide to how teachers detect AI.

Setting Expectations and Designing Resilient Assignments

Decide before the semester starts what you will accept as evidence of unauthorized AI use, and put that standard in your syllabus. If you teach writing or programming, outline exactly which generative tools are permitted for brainstorming, outlining, or proofreading.

Our assessment of automated detectors would change if developers eliminated false positives and created systems capable of independently verifying source authenticity across diverse author backgrounds. Until that happens, an automated score serves only as an initial prompt for instructor inquiry. To need less detection this semester, use our guide to designing AI-resistant assignments around staged drafts and in-person discussion.