AI and Schoolwork: What Counts as Cheating, and How Detection Tools Actually Work
AI-detection tools are far less reliable than schools using them often imply — worth understanding before either a student or a parent takes a flagged result at face value.
Whether AI use on an assignment counts as cheating depends entirely on the specific policy for that class and that assignment, not on a general rule — the same tool that's a banned shortcut on one assignment might be an explicitly encouraged aid on another, like using a calculator on a math test versus using one during a lesson on mental arithmetic. The policy to check is the syllabus or the assignment instructions, not an assumption based on other classes.
AI-detection tools work by analyzing statistical patterns in writing — measuring how predictable each word choice is given the words before it, on the theory that AI-generated text tends to be more statistically uniform than human writing, which varies more in word choice and sentence rhythm. That's a real, measurable pattern, but it's a probability estimate, not a verified fact about how a specific piece of text was produced.
The reliability problem is well-documented: these tools produce both false positives, flagging genuinely human-written text as AI-generated, and false negatives, missing AI text that's been lightly edited by a human afterward. Non-native English speakers and very formulaic writing styles are flagged as "likely AI" at a meaningfully higher rate than fluent, varied writing, independent of whether AI was actually used at all.
This matters practically because a detection tool's score, on its own, isn't strong enough evidence to treat as proof — a school relying solely on a percentage from a detection tool, without other supporting evidence like an earlier draft, a revision history, or a conversation with the student, is relying on a tool with a real, acknowledged error rate as if it were a verified test result.
The practical advice for a student: read the actual policy for the specific assignment rather than assuming, and if a workflow like using AI to brainstorm but writing the final draft yourself is genuinely what happened, keep the artifacts that show it — outlines, earlier drafts, revision history — since those are more convincing evidence than arguing against a detection score after the fact.