AI Detection Tools

AI is increasingly being used to help detect plagiarism, check for AI-generated writing, and monitor student work. Here is what you should know about how these tools work — and their significant limitations.

How AI detection tools work

AI text detectors typically analyse statistical patterns in writing — things like word choice predictability, sentence length variation, and perplexity (how surprising each word choice is). AI-generated text tends to be more predictable than human writing, though this gap is narrowing as models improve.

The false positive problem

Current AI detectors have significant false positive rates — they flag human-written text as AI-generated at a concerning frequency. Studies have found that non-native English speakers are particularly likely to be wrongly flagged, because their writing patterns may resemble the more constrained, predictable output of AI. This creates real fairness concerns.

Why detection is fundamentally hard

There is no definitive "AI signature" in text. AI-generated text can be edited to avoid detection; human-written text can coincidentally resemble AI outputs. As models improve, the statistical differences between human and AI writing narrow further. Detection is an arms race that the detectors are increasingly losing.

The better question

Rather than trying to detect AI use after the fact, educators are increasingly focusing on assessment design — oral presentations, iterative drafts, in-class writing, and assignments that depend on personal knowledge or local context that AI cannot fake. These approaches are more robust than any detection tool.

AI detection tools are neither reliable enough to use as evidence of wrongdoing nor pointless enough to ignore. Understanding their limitations — and the context in which you are using AI — is the most honest approach.