25 June 2026 · 5 min read
How to Stop Students Cheating with ChatGPT in Online Exams
AI assistants made it trivial to answer an online test from a second tab. The honest starting point: no tool stops every form of cheating. What you can do is raise the cost of cheating and design assessments that AI cannot quietly complete. Here is what actually works.
Can you fully prevent ChatGPT cheating in an online exam?
No, and anyone selling you a 100% guarantee is overpromising. A determined student with a second device can consult an AI no matter what software runs on the exam screen. The realistic goal is deterrence and detection: make casual cheating risky, make the test hard to outsource, and keep a human reviewing the edge cases. Treating it as an arms race you can "win" with surveillance leads to invasive tools students resent and false accusations you then have to defend.
What assessment design beats AI cheating?
Design is your strongest lever, before any software. Ask for the reasoning, not just the answer: require students to show working, justify a choice, or critique a flawed example. Use course-specific context — a dataset from class, a local case, last week's lecture — that a generic model cannot reproduce. Lower the stakes by testing more often with smaller weight, so a single outsourced answer matters less. These changes shrink the surface ChatGPT can quietly fill, regardless of the platform.
How does behaviour-based proctoring help without a webcam?
Browser-based proctoring watches the exam session, not the student's face. It scores risky behaviour — tab-switching, copy and paste, right-click, screenshot key presses, leaving fullscreen — into a per-student integrity timeline you can review afterward. This is evidence for a conversation, not an automated verdict. Paired with an identity watermark tiled across every screen, any leaked or photographed question traces back to its source. It deters the easy routes without recording anyone in their bedroom.
Is no-webcam proctoring enough on its own?
It is enough for most coursework when combined with good assessment design and a manual review of flags. Webcam proctoring adds recording, privacy complaints and legal risk for a marginal gain, and students increasingly refuse it. A lighter, honest layer — behaviour signals plus watermarking plus human review — is defensible, accepted by students, and keeps your institution out of the surveillance headlines.
A practical setup
ExamGuard Online runs exams in any laptop browser with no webcam and nothing to install. Objective questions auto-grade; written answers go to a manual-review queue so phrasing never costs a student marks unfairly; and every flag is reviewable evidence rather than an accusation. It is free to start.