How to Stop ChatGPT Cheating in Online Exams

First, what does not work

The tempting fix is an AI text detector: paste an essay, get a percentage. The problem is that these tools give probabilities, not proof. They mislabel human writing, especially short answers and texts by non-native speakers, and simple rewording fools them. OpenAI withdrew its own AI text classifier in 2023 because its accuracy was too low. Accusing a student on the basis of a detector score is unfair and hard to defend.

Banning devices is not enough either. In an online exam, the chatbot can be one tab away. And a webcam, on its own, cannot see what is in another window.

1. Write questions a chatbot struggles with

Question design is the most effective and the cheapest defense. Chatbots are very good at generic textbook questions and much weaker at questions tied to a specific context.

2. Watch how answers are written

A real answer is typed, paused, edited and reworded. A copied answer often appears in one or two large pastes. Integrity tools can block pasting in an exam, log it when it happens, and flag answers whose typing pattern looks like a transcription rather than writing. Tab switches and fullscreen exits add context: leaving the exam for two minutes right before a perfect paragraph appears is worth a look.

These are signals, not proof. Some students write in a text editor and paste their work for good reasons. Treat a flag as a reason to talk, not a verdict.

3. Ask a follow-up question

The oldest integrity check is the oral exam: explain what you wrote. It works because understanding is hard to fake in real time. Teachers rarely have time to do it for every student, but an AI can ask one short spoken question about the student's own answer and grade the reply. A student who wrote the essay answers easily. A student who pasted it usually cannot.

4. Compare answers across the class

Students who use the same chatbot prompt, or share answers, often submit strikingly similar work. A similarity report that lists unusual pairs makes this visible in seconds, without reading every essay side by side.

5. Use camera proctoring when the stakes justify it

For a final exam or a certification, camera checks add a real deterrent against phones, notes and helpers. Choose a privacy-respecting option that analyzes video on the device, uploads only flagged moments and leaves the decision to a person. We explain how that works here: AI Proctoring Without Spyware: How On-Device Proctoring Works

6. Be clear about the rules

Many students are unsure what is allowed. Say it plainly in the instructions: whether AI tools are allowed for this test, which ones, and what happens if the rules are broken. When AI help is allowed, ask students to describe how they used it. Clear rules prevent more problems than any detector.

How FormRead Online helps

• Integrity levels from Light to Strict: fullscreen, tab switch and paste checks, and on-device camera checks when you need them.
• Paste-burst and typing-pattern flags on open answers.
• An AI oral follow-up question after essays, graded with your rubric.
• A collusion report of unusually similar answers.
• Audio and video answers, transcribed and graded by AI.
• Question pools, shuffling and time limits, so no two students see the same test.
• The option to print the same test on paper and grade it with a phone.

Grading open answers is the other half of the job. Here is how AI grading with a rubric works on paper and online: AI Grading for Open-Ended and Essay Questions on Paper Exams