How can teachers tell if something is AI-generated?

Before AI, there was no getting around it: writing an essay would take hours, if not days or weeks. Today, AI tools can produce comparable text in seconds, making it tempting for students to try to pass off AI-generated work as their own. But besides the ethical implications—doing so is a violation of academic integrity—it’s unlikely they’ll get away with it. There are several ways teachers can tell if something was written by AI:

• The work doesn’t match up with what the student has done before. Through in-class discussions, hand-written exercises, and other assignments, a teacher develops a sense of a student’s abilities and voice. Often, this makes it easy to tell if the work they submitted is not actually their own. For instance, if an eighth-grader with an average vocabulary turns in an essay full of more advanced words, that could be a warning sign, as could a submission conspicuously free of the spelling and grammar mistakes. Similarly, if the sophistication of the ideas in the essay goes beyond anything the student has demonstrated in the past, that could suggest AI wrote it.
• Telltale language suggests AI wrote the text. Just as many writers have a handful of words they use all the time, AI has a slew of go-to phrases that it relies on much more often than the average human does. From “faced numerous challenges” to “poignant,” teachers recognize these as signals that an essay was computer-generated.
• The essay includes hallucinations. Student writing may sometimes contain factual errors, but those made by AI are likely to be much more blatant. These mistakes, called hallucinations, could involve misstating widely-known facts—say, the name of the current vice president—or fabricating quotes. An AI-generated submission might also fail to follow simple instructions. For instance, prompting AI to write an essay quoting the third-to-last paragraph of The Great Gatsby results in output referencing a more famous line from earlier in the novel. A human writer, by contrast, would flip to the back of the book to find a quote.
• An AI detector like Pangram flags it for them. The easiest way for teachers to tell if an assignment was AI-generated is to run it through an AI detector. Pangram is among the most accurate and robust of these transparency tools, but even the software that teachers have used for years to identify plagiarism now often checks for AI as well.

These signs aren’t definitive proof that an essay was AI-generated, but if teachers see them as red flags, this may prompt a discussion about the student's writing process or even an investigation into honor code violations. Make sure to exercise caution and use a tool with low false positive rate when checking students' work.