How AI detectors work and how to navigate them: detection signals, false positive risks, and what the scores mean for academic writers.
39 articles · Page 3 of 7
GPTZero and Turnitin get treated like two versions of the same test. They are not. One is licensed to your university and attached to your actual submission. The other is a consumer product you can run yourself at midnight. Here is what separates them.
Anthropic began marking Claude's text output on 11 August 2026. Almost every explanation of it since has described the wrong mechanism. The mark is not a character hidden in the page, it is a statistical bias in which words the model picked, and that distinction decides everything: what carries it, what clears it, and how little a positive result actually settles about who wrote a document.
An AI detection appeal letter only works if it reads as an argument rather than a protest. Here is a full template you can adapt, the evidence checklist that goes with it, and the single most common reason appeals fail.
ZeroGPT's own FAQ claims a rate pushing toward 98 percent on internal testing, and separately admits that formulaic writing can read as AI regardless of who wrote it. Here is what the company states about its method, what independent testing on this exact product has found, and what a free score is actually useful for.
A practical walk through what AI use is almost always fine, what is almost never fine, the grey area in between, and how to keep a simple record of what you did, starting from the one policy that actually governs an assignment: the course's own.
Turnitin publishes two false positive figures, not one, and neither describes the paper sitting in front of one particular professor. Here is the document-level versus sentence-level split, the sub-20-percent asterisk, and the arithmetic that turns a fraction of a percent into a real count of students.