How can teachers tell whether a student actually wrote an assignment or handed the work over to artificial intelligence? With text-generating tools now widely available, the issue has become a major challenge for educators.
Jason Gibson, a history professor at Alcorn State University in Mississippi, decided to try an unusual approach. During an exam held in late July 2026, he inserted an instruction into the prompt that his students would normally be unable to see.
The goal was to find out which students would copy and paste the entire prompt directly into an AI assistant.
A hidden instruction in the exam prompt
Students were asked to write about the Industrial Revolution and draw parallels with modern-day transformations, including those linked to social media and artificial intelligence.
Among the instructions, the professor added a sentence written in white text on a white background. Invisible when reading the document normally, it instructed students to include the word “Madagascar” in their answer, even though it had nothing to do with the topic.
The trick relied on a simple principle: by copying and pasting the entire prompt into an AI chatbot, students would unknowingly include the hidden instruction. The software could then follow that instruction when generating its response.
By contrast, a student writing the assignment independently would have no reason to mention the secretly requested word.
32 submissions reportedly fell for the trap
The method proved remarkably effective. Of the 35 students involved, 32 reportedly submitted answers containing the word “Madagascar,” sometimes appearing in sentences that made little sense in the context of the exam.
According to Gibson, some students apparently failed to carefully reread the AI-generated text before submitting their work. A simple review would likely have revealed the strange and irrelevant passages.
The students concerned failed the exam. Only two reportedly challenged their grades afterward, according to the professor, who described the experiment in a video that was widely shared on social media.
Universities face a growing challenge from AI
Beyond this particular case, the use of generative artificial intelligence in education is becoming an increasingly significant issue for schools and universities.
In the United Kingdom, nearly 7,000 confirmed cases of AI-related cheating were reportedly recorded during the 2023-2024 academic year. The rate reached 5.1 cases per 1,000 students, compared with 1.6 per 1,000 the previous year.
Educational institutions are now looking for ways to adapt their assessment methods. Automated tools designed to detect AI-generated writing remain imperfect and can sometimes incorrectly identify human-written content as machine-generated.
Gibson’s approach partly sidesteps this problem. Rather than attempting to determine afterward whether a text sounds like it was written by artificial intelligence, the technique aims to leave evidence of AI use directly in the submitted assignment.
It is an extremely simple method, but one that highlights the growing cat-and-mouse game between students, teachers and artificial intelligence tools.
