Can teachers detect ChatGPT? What schools use and how well it works

Can teachers detect ChatGPT? Sometimes, but less often than either they or their students tend to think. Studies show teachers usually can't pick out AI essays just by reading them, and Turnitin, the detector many schools license, is deliberately tuned to miss some AI writing rather than accuse innocent students. What catches AI misuse most reliably is a mix of evidence: a detector score, the student's drafts and version history, and a conversation about the work.

Can teachers detect ChatGPT just by reading an essay?

Usually not. In a 2024 study, Johanna Fleckenstein and colleagues asked 89 novice teachers and 200 experienced teachers to find the ChatGPT essays in a set of student essays. Neither group could reliably tell them apart. Experienced teachers made somewhat more careful judgements, but both groups were overconfident.

A real-world test at a UK university, published in PLOS ONE in 2024, went further. Researchers submitted exam answers written entirely by GPT-4 under fake student names in five modules of a psychology degree, without telling the markers. 94% of the AI submissions went undetected, and on average they earned grades about half a grade boundary higher than real students' work.

Reading does work better for some people. In a 2025 study by Jenna Russell, Marzena Karpinska and Mohit Iyyer, readers who often used AI tools for their own writing were very accurate: five of them, voting together, misjudged only one of 300 articles. Readers who didn't use AI for writing did about as well as a coin toss. A teacher who has spent time with the tools has a real advantage over one who hasn't.

What AI detection tools do schools actually use?

In many schools the main tool is Turnitin's AI writing indicator, which sits inside the Similarity Report already used for plagiarism checks. It is sold to institutions, not individuals: Turnitin says AI writing detection is available to customers who license Turnitin Originality, and only instructors and administrators can see the indicator. Students can't run their own work through it.

Teachers have come to rely heavily on these tools. In a nationally representative survey of US middle and high school teachers in late 2023, the Center for Democracy & Technology found that a majority reported using school-endorsed AI content detection tools, while getting little guidance on how to respond when they suspected AI use.

Standalone checkers such as GPTZero and Pangram also sell to educators, and there are free tools like ours; our ranking of the best AI detectors for teachers compares them on LMS support, price and student privacy. Some schools have added writing-process tools instead of, or alongside, detection. Turnitin Clarity, a paid add-on announced in 2025, gives students a composition space and lets educators replay how a submission was drafted, including pasted text and time spent. Google Docs and Microsoft Word keep version histories that teachers can ask to see.

How well do the detectors work?

Turnitin is open about its trade-off: to keep false accusations rare, it accepts missing some AI text, and it says its score "should not be used as the sole basis for action". Independent tests bear out the caution. A June 2026 study from Vrije Universiteit Brussel found Turnitin scored every one of 40 fully AI-written papers under 20%, while wrongly flagging none of the human ones. A 2023 test of 14 tools, Turnitin among them, judged them neither accurate nor reliable, and Vanderbilt University switched Turnitin's AI detector off that August. Our article does Turnitin detect AI? goes through the details.

Some standalone tools have done better. Pangram came out on top in a University of Chicago working paper, where GPTZero and Originality.ai formed a second tier, and again in the Russell and Brussels studies. None is beyond fooling: a 2026 University of Notre Dame study found one commercial humanizer got AI-written abstracts past both Pangram and GPTZero almost every time, while light, possibly permitted AI editing was flagged far more often than the disguised text. Our round-up of detector studies has the numbers.

Our own detector is a free first check: no account, no word limit, and the essay never leaves the teacher's browser, which matters for student privacy. Alongside the verdict, the report shows a sentence map and style notes, and how it is tested is set out in our open benchmark. Like any detector, it can be wrong in both directions, so treat a result as a reason to look closer, not as proof.

Why isn't a detector score enough?

Every detector trades missed AI text against false accusations, and the cost of a false accusation falls on a student. OpenAI says its own research didn't find detectors reliable enough for judgements about students with lasting consequences, and that a classifier it trained labelled Shakespeare and the Declaration of Independence as AI-generated.

Students who write in a second language, or in a formulaic style, carry more of that risk, because plain and predictable prose is what many detectors read as machine-made. Stanford researchers showed in 2023 how hard the detectors of the day were on English learners, a finding we look at in AI detectors and non-native English.

A score is useful for deciding which essays to read again. It can't tell you what the student understands, which is what most assignments are meant to measure.

What helps teachers catch AI misuse fairly?

• Clear rules. Say which uses of AI are allowed for each assignment. A rule nobody explained is a weak basis for any accusation.
• Process evidence. Ask for drafts, outlines or a shared document with version history. A document's history shows how the work developed, which a finished file can't.
• A short conversation. Ask the student to explain an argument or a source. Turnitin publishes a list of questions instructors can use to open that talk.
• Some writing in class. A baseline of the student's unassisted writing makes every later comparison fairer.
• Detectors as a pointer. Use a score to decide where to look, and record what you found beyond it.

For a step-by-step version, including what version history can and can't show, see how to check if an essay was written by ChatGPT. Our page for teachers explains how to use our report alongside these steps.

Sources

1. Fleckenstein et al., Do teachers spot AI? (Computers and Education: Artificial Intelligence, 2024), checked October 2026
2. Scarfe et al., A real-world test of artificial intelligence infiltration of a university examinations system: A "Turing Test" case study (PLOS ONE, 2024), checked October 2026
3. Russell, Karpinska and Iyyer, People who frequently use ChatGPT for writing tasks are accurate and robust detectors of AI-generated text (ACL 2025), checked October 2026
4. Center for Democracy & Technology: Up in the Air: Educators Juggling the Potential of Generative AI with Detection, Discipline, and Distrust (2024), checked October 2026
5. Turnitin Guides: Turnitin's AI writing detection capabilities FAQs, checked October 2026
6. Turnitin Guides: What questions do I ask my student if AI is detected?, checked October 2026
7. Turnitin: Turnitin Clarity, bringing transparency to the student writing process (4 March 2025), checked October 2026
8. Van Vlasselaer, Van Droogenbroeck and Spruyt, Who wrote this? Evaluating the reliability of AI detection tools in higher education (International Journal for Educational Integrity, 2026), checked October 2026
9. Weber-Wulff et al., Testing of detection tools for AI-generated text (International Journal for Educational Integrity, 2023), checked October 2026
10. Vanderbilt University: Guidance on AI Detection and Why We're Disabling Turnitin's AI Detector (16 August 2023), checked October 2026
11. Jabarian and Imas, Artificial Writing and Automated Detection, BFI Working Paper 2025-116, checked October 2026
12. Karr, Khvatskii, Hua and Chawla, Why AI Detection Fails for Academic Integrity (ACM AI Summit 2026), checked October 2026
13. OpenAI Help Center: How can educators respond to students presenting AI-generated content as their own?, checked October 2026
14. Liang et al., GPT detectors are biased against non-native English writers (Patterns, 2023), checked October 2026