The first chatbot draft landed in a teacher’s inbox roughly two years ago, and the situation has only accelerated since. A student can now ask ChatGPT to produce five paragraphs on the causes of the Great Depression or the symbolism in The Great Gatsby and receive a coherent response in less time than it takes to sharpen a pencil. For educators, this has changed the nature of academic integrity. They are no longer hunting only for traditional plagiarism, where one human copies another human’s work. They are also looking for synthetic prose—text that sounds human but was assembled by a model with no understanding, no memories, and no stake in the argument.
When the Voice Suddenly Shifts
Teachers read hundreds of student essays each semester. Over time, they develop an ear for individual cadence. They notice the habitual comma splices, the pet phrases, the way a particular student structures a conclusion. When a paper arrives that sounds nothing like the voice they have come to expect, it triggers a closer review.
This is not about punishing improvement. Good teachers live to see growth. But there is a difference between a student who has steadily improved their syntax and one who submits an essay that reads like a polished legal brief after eight months of clunky, simple sentences. The prose may be immaculate, yet it feels borrowed because it lacks the archaeology of practice. That sudden, unsupported leap in sophistication is often the first clue.
Professional Surface, Hollow Center
AI-generated writing tends to occupy an uncanny valley of correctness. The grammar is sound. The transitions flow. Yet the text floats above the material rather than digging into it. Teachers often describe this quality as “professionally generic.”
Consider an essay on Romeo and Juliet. A student writer might become fixated on Mercutio’s minor role, misread a scene in an interesting way, or compare the family feud to a conflict in their own community. The thesis might be messy, but it is unmistakably the product of a single mind wrestling with the text. An AI draft, by contrast, often summarizes the plot accurately, lists three universal themes, and concludes that love requires sacrifice. It is correct. It is also empty. That absence of risk—no quirky opinion, no flawed but genuine insight—is one of the most reliable indicators that the writer was not the student.
References That Lead Nowhere
Another red flag is the hallucinated citation. ChatGPT and similar tools occasionally invent books, journal articles, and even authors. The entries look plausible, complete with academic-sounding titles, publisher names, and years of publication, but they have no connection to reality.
Veteran teachers verify sources. They plug titles into Google Scholar, search the campus library database, or simply run an author’s name through a search engine. When every citation leads to a dead end, the paper’s credibility collapses. Checking bibliographies has always been part of serious research instruction, but it is now becoming standard practice even for short analytical essays. The works-cited page is no longer just a formatting exercise; it is an integrity checkpoint.
The Limits of Detection Software
Many districts have responded to the surge in synthetic text by licensing AI-detection tools. These programs scan for statistical patterns—low perplexity, predictable sentence structures, repetitive rhythm—that are common in machine-generated prose. They can be useful. A high probability score gives an instructor reason to look more carefully.
But teachers know these tools are imperfect. They sometimes return false positives, particularly against prose written by non-native English speakers or by students who naturally favor a formal, measured style. A detection percentage should never be treated as proof of cheating on its own. Most educators use the software as one clue among many, not as a smoking gun. The tool supports human judgment; it cannot replace it.
Analysis Without a Thinker
Large language models summarize facts well. What they struggle to deliver is sustained critical thinking or authentic personal reflection.
Dans un devoir d'histoire, par exemple, une IA pourrait énumérer les causes économiques, politiques et sociales d'une révolution avec une précision de manuel scolaire. Ce qu'elle ne peut pas facilement faire, c'est mettre ces causes en balance, remettre en question le récit conventionnel ou relier l'événement à un enjeu contemporain qui intéresse réellement l'élève. Les rédacteurs humains se trahissent par leur spécificité. Ils font référence à un documentaire qu'ils ont regardé, à une conversation avec un parent ou à un schéma qu'ils ont remarqué à travers trois unités différentes. Les textes générés par l'IA contiennent rarement ces points d'ancrage, car la machine n'a aucune expérience sur laquelle s'appuyer et aucune opinion à défendre.
La conversation de cinq minutes
L'une des méthodes les plus anciennes reste la plus efficace : demander à l'élève de parler de son travail.
Un enseignant pourrait prendre un élève à part pendant