Key takeaways
- Turnitin analyses text in segments and flags each with an AI probability score.
- The overall AI percentage reflects flagged segments relative to total document length.
- Turnitin cannot detect AI in images, handwritten work, or non-text submissions.
Turnitin's AI Writing Indicator is the most widely deployed AI detection tool in higher education. It appears in the same report as your similarity score, but measures something entirely different. Here is a clear explanation of how it works, what it catches, and where it falls short.
Segment-level analysis
Turnitin divides your document into overlapping text segments—typically a few sentences each—and runs each segment through its AI classifier. Segments scoring above the detection threshold are highlighted in the report. The overall AI writing percentage is the proportion of your document composed of flagged segments.
What the AI percentage means
A 40% AI writing score means roughly 40% of your document's sentences were classified as likely AI-generated—not that 40% was copied from ChatGPT verbatim. Mixed documents with some AI-assisted and some human-written sections produce partial scores.
Features in the 2026 Turnitin AI report
- Sentence-level highlighting showing exactly which passages were flagged.
- Separate AI score displayed alongside the similarity score.
- Ability for instructors to exclude flagged quotes or bibliography sections.
- Model updates covering GPT-4, Claude, Gemini, and other major LLMs.
- Integration with Turnitin's draft submission workflow for pre-checking.
What Turnitin AI detection cannot do
- Detect AI used only for research, brainstorming, or outlining without text in the submission.
- Analyse content in PDF images, scanned documents, or non-selectable text.
- Identify which specific AI model generated flagged text.
- Provide legally admissible proof of AI authorship on its own.
- Reliably detect heavily edited AI text or human-AI collaborative writing.
Turnitin's guidance to institutions
Turnitin explicitly recommends that instructors use the AI indicator as one data point in a broader assessment of student work. The company warns against automatic penalties based solely on the AI percentage and acknowledges false positives on human-written academic text.