AI writing detectors estimate whether text resembles patterns associated with generated writing. They do not establish who wrote a document, what tools a student used, or whether a course rule was broken. Use clear expectations, evidence of learning, and the established academic-integrity process.
Reviewed October 2026.
Detectors use different models, training examples, and scoring methods. Descriptions based only on “perplexity” or “burstiness” do not explain every product. A similarity report and an AI-writing report also answer different questions.
A flagged passage is an estimate, not a verified record of authorship. An unflagged passage does not prove that AI was absent. Read the specific tool’s documentation before interpreting a score.
A percentage of qualifying text flagged by a detector is not the probability that a student cheated. A tool’s reported performance on one dataset does not establish its performance on your students, discipline, language, or assignment type.
Turnitin acknowledges that its model can misidentify human-written and AI-generated writing. Its guidance says the score should not be the sole basis for adverse action against a student. See Using the AI Writing Report.
Research published in 2023 found that several detectors disproportionately flagged writing by non-native English writers. That study is an important warning about fairness; its results are not a current accuracy estimate for every detector. Read Liang and colleagues’ study.
Recheck the assignment’s stated AI expectations and what students were taught about them.
Identify specific concerns in the work, such as unverifiable sources or a mismatch with the task.
Invite a private, neutral conversation: “Walk me through how you developed this idea and checked this source.”
Consider relevant drafts, notes, or explanations. Missing version history or difficulty answering under pressure does not by itself prove misconduct.
Follow the college’s current process, including the student’s opportunity to respond. Apply standards consistently.
Before uploading work to any external detector, confirm institutional approval and applicable privacy, retention, and intellectual-property requirements. Removing a name does not necessarily remove identifying information. This page does not authorize a tool or replace college policy.
Use checkpoints, feedback, authentic application, and explanations of decisions. Make permitted assistance explicit. Provide accessible alternatives for demonstrations and discussions. No assignment format is universally “AI-proof.”
Hold students to clear standards while giving them a fair opportunity to explain their work. Ask what they can demonstrate, what support they need, and which next step follows the course expectations and college process.
These sources support interpretation and policy discussion. Consult current institutional procedures before taking action.