Quick Answer: Yes, Turnitin can flag ChatGPT text. Its AI writing indicator gives every submission a score from 0% to 100% - anything above 20% tends to get a second look from instructors. It's not foolproof: the false positive rate is around 1%, and heavily rewritten AI text can slip through. AI Busted runs the same check on your text before you submit, so you know your score before anyone else does.
What Is Turnitin AI Detection?
Turnitin added an AI writing indicator in April 2023. It runs alongside the plagiarism check and gives instructors a second number: a 0-100% estimate of how much of a submission looks AI-written.
The tool was built to catch the patterns LLMs like ChatGPT and GPT-4 tend to leave behind. It runs as a separate pass from the plagiarism check.
High scores don't mean anything automatically. The instructor sees the indicator and decides what to do with it. Turnitin won't call anyone a cheater - that's the institution's job.
How Does Turnitin Flag ChatGPT Text?
Turnitin runs your text through a statistical model. LLMs like ChatGPT write with very uniform structures - low perplexity, consistent phrasing, the same kinds of word choices appearing in the same spots. People don't write like that.
According to Turnitin's own published documentation, the system cuts submissions into chunks of roughly five to ten sentences and scores each one separately. The full document goes through a model built on a large mix of human-written and AI-written text.
It's not looking for specific ChatGPT phrases or watermarks. It's comparing the pattern of your writing against a baseline.
A few things that affect the score:
- Short texts under 300 words often return unreliable scores
- Lightly edited AI text may still score above 20%
- Non-native English speakers may see higher false positive rates
- Text with unusual formatting or code may confuse the scorer
If you want to know where your text stands before anyone sees it, AI Busted runs the same type of AI-scoring pass.
What Does a Turnitin AI Detection Report Show?
When Turnitin's AI writing indicator fires, instructors see two things: an AI score percentage and a color-coded view of the submission showing which passages scored highest for AI authorship.
The percentage is the headline number. It represents the share of the submission Turnitin believes was AI-written.
The color-coded view goes deeper:
- Cyan marks text flagged as AI-written.
- Purple marks text that appears to have been paraphrased from an AI source.
- Gray marks passages the system could not score with confidence.
| AI Score Range | Signal Strength | Typical Instructor Response |
| 0-9% | Little to no AI signal | Usually not flagged |
| 10-19% | Low AI signal | Rarely escalated; at instructor's discretion |
| 20-49% | Moderate AI signal | Often triggers a closer look |
| 50-74% | High AI signal | Likely referred for review |
| 75-100% | Very high AI signal | Almost always escalated |
One thing Turnitin states openly on its site: the score is not proof of cheating. It is a signal for instructors to investigate further. A student who writes with a polished, consistent voice may score higher than someone who writes choppily.
For a broader look at how AI scoring works across tools, see our guide on how reliable AI scoring tools actually are.
How Reliable Is Turnitin at Catching ChatGPT Text?
According to Turnitin's published data, the system correctly identifies AI-written text above 98% of the time, with a false positive rate under 1% - when tested on its validation set.
That sounds strong. But validation sets are controlled conditions. Real-world classrooms are messier.
A 2023 breakdown of how AI content scoring systems work found that AI-scoring systems regularly struggled with text that was lightly edited after AI output, writing from non-native English speakers (where false positive risk is higher), and mixed human-AI documents where students used AI for research but wrote sections by hand. Independent academic research, including a peer-reviewed study on identifying LLM-produced text, confirms that identification rates drop when writers vary sentence structure, pacing, and phrasing beyond the model's learned patterns.
Turnitin itself warns instructors that its score should not be the sole basis for an academic integrity finding. It is a flag, not a verdict.
What Factors Affect Turnitin's ChatGPT Detection Score?
Several things change how Turnitin scores a piece.
Text length. The AI writing indicator is less reliable on short submissions. Turnitin recommends at least 300 words for a valid score. Below that threshold, the output is not worth reading into.
Editing depth. Light edits - swapping synonyms or rearranging sentences - often do not move the score much. Deeper rewrites that change sentence structure and phrasing reduce the AI signal more. See our guide on rewriting AI text for approaches that hold up under scrutiny.
Writing voice. A polished, consistent style may score higher than choppy, casual writing. This is one reason non-native English speakers sometimes see false positive flags on work they wrote themselves.
Prompt style. AI text from a conversational ChatGPT prompt often scores differently from text produced by a structured, constrained prompt. More constrained prompts tend to produce output with lower variation.
Text type. Fiction writing, code, and lists behave differently from academic prose. The model was built for prose, so other formats may produce less stable scores.
What Happens If Turnitin Flags Your Submission?
A high AI score does not automatically mean anything happens. The process varies by institution, but it usually goes like this:
- The instructor sees the AI writing indicator alongside the similarity report.
- If the score is high - often above 20% - the instructor looks at the color-coded passages and compares them against other samples of your writing.
- The instructor decides whether to escalate to an academic integrity review or talk to you first. Many start with a conversation.
- If escalated, you get a chance to explain your process or show evidence of your own work.
Turnitin is a flag, not a judge. What happens next belongs to the institution.
For students using AI tools in their work: knowing your score before submission is worth doing. You can check it at AI Busted before anyone else sees it. For a direct comparison of how Turnitin stacks up against other tools, see our GPTZero vs Turnitin breakdown.
FAQ
Does Turnitin flag ChatGPT text or just AI writing in general?
Turnitin's AI writing indicator does not name a specific tool. It scores the statistical properties of your text against a model built from AI output across many sources, not just ChatGPT. A high score means Turnitin thinks the text was AI-written. It does not say which tool produced it.
What percentage on a Turnitin AI score means flagged?
There is no universal threshold. Scores above 20% often trigger a yellow indicator in the interface, but each institution sets its own policy. Some instructors act on scores as low as 10%. Others wait for 50% or higher.
Can a professor see which AI tool you used on Turnitin?
No. Turnitin only gives a score and marks suspicious passages. It cannot tell a professor whether the text came from ChatGPT, Claude, Gemini, or any other tool.
Does Turnitin flag lightly edited AI text?
Often yes. Light edits - changing a few words or rearranging sentences - often leave enough of the original pattern for the score to stay high. Deeper rewrites that change sentence structure, voice, and phrasing are more likely to bring it down.
Can Turnitin catch Claude or Gemini?
Yes. Turnitin's AI writing indicator targets the statistical patterns common to all large language models, not one specific tool. Text from Claude, Gemini, and other models may score high if they share those patterns.