Artificial intelligence, with tools like Yiaho or ChatGPT, has revolutionized the way students approach their homework. According to Médiamétrie, in January 2024, over 20% of French students had already used artificial intelligence for academic or personal needs.
But faced with AI detection software, like our online ChatGPT detector, Turnitin or GPTZero, a new generation of strategies is emerging.
Students are competing in ingenuity to make their texts undetectable, turning writing into a cat-and-mouse game. Here’s a look at these techniques, enriched with recent trends, and their implications for education.
AI-written text: Students’ new tricks
Mimicking an “imperfect” style:
A popular trick, spotted on platforms like TikTok and reported by Futurism, involves asking the AI to write a text in a deliberately clumsy style. For example, one student explained that she instructed ChatGPT to write “like a somewhat dumb freshman“.
Result: a text riddled with approximations and a less polished style, which more easily passes as human in the eyes of detectors and teachers.
AI cascade to cover tracks:
Some students don’t settle for just one AI. They run their text through multiple tools, like Yiaho, ChatGPT, then Claude or Grok. Each pass slightly modifies the style and structure, making the trace of the initial AI almost impossible to spot, according to observations reported by tech media.
Advanced personalization:
By incorporating personal anecdotes or specific references to their course, students give their texts a touch of authenticity. For example, mentioning a local event or a class discussion can convince a professor that the work is original, even if the framework comes from an AI.
Custom prompts:
Students refine their instructions to get texts that match their academic level. A prompt like “Write an essay in French like a 17-year-old high school student with occasional mistakes” produces a credible result, far from the overly smooth style of an unguided AI.
We also notice that students deliberately add spelling mistakes to “humanize” their text even more!
Hybrid mix:
A common technique involves merging AI-generated text with manually written parts. This mix dilutes the typical markers of AI models, such as overly uniform syntax or overly sophisticated vocabulary, making detection more difficult.
Also read on this topic: Do anti-plagiarism software detect texts generated by ChatGPT?
A rapidly expanding global phenomenon
The use of generative AI is not just a gadget for Anglo-Saxon students. In France, the phenomenon is gaining momentum. According to a Médiamétrie study from January 2024, approximately 20.3% of French students have already used an AI like ChatGPT.
This figure reflects rapid adoption, especially among young people, who see these tools as a way to optimize their time.
Why do these tricks work?
AI detectors rely on clues like syntactic regularity, the absence of human errors, or linguistic patterns specific to AI models. Students’ techniques, by introducing imperfections or diversifying sources, exploit these limitations.
For example, a text passed through multiple AIs or deliberately “degraded” can escape current algorithms, which struggle to distinguish human work from a well-designed hybrid work.
But this advantage is temporary. Detection software publishers, aware of these strategies, are refining their tools. Some are exploring approaches based on analyzing a student’s writing style over time to spot sudden inconsistencies, like an unexplained jump in the quality of an assignment.
See also: Is AI plagiarism? Here’s what the law says
The risks of such a practice
These techniques, while creative, are not without consequences:
- Academic sanctions: A text identified as AI-generated can result in severe penalties, ranging from a failing grade to more serious disciplinary measures.
- Loss of skills: By delegating writing to an AI, students risk stagnating in their analytical and writing skills, essential for their future.
- Ethical dilemma: Presenting AI work as one’s own raises a question of integrity. Is it really “winning” to bypass the system at the expense of one’s own learning?
Teachers’ response
Faced with this wave, professors are fighting back with adapted strategies:
- Controlled assessments: In-class assignments, where digital tools are inaccessible, are regaining popularity.
- Ultra-specific questions: By asking for analyses based on specific courses or required readings, teachers complicate the use of AI.
- Oral exchanges: A face-to-face discussion allows verification of whether the student truly masters the content of their work.
An opportunity to rethink education?
Rather than demonizing AI, some institutions are exploring its integration. Some professors allow the use of ChatGPT for tasks like brainstorming ideas or structuring an outline, provided students mention its use. This approach, already tested in some universities, aims to transform AI into a pedagogical tool rather than a threat.
Students, armed with creativity and tools like ChatGPT, are pushing the limits of AI detectors with increasingly sophisticated techniques. But this technological tug-of-war raises a deeper question: how can education evolve to integrate AI while preserving authentic learning? In France, where one in five students already uses these technologies, the stakes are high. The solution may lie in dialogue between teachers, students, and AI developers, to make these tools allies rather than adversaries.
Source: Futurism
Source: Mediamétrie
