An AI detector's percentage opens a conversation with the student, but does not prove they copied
A post on the blog “Familias con inteligencia” asks what to do when a detector flags the work of a student who says they wrote it themselves, and suggests backing the percentage up with evidence of the process.
Chema Falcó published “La máquina dice que copió. Él dice que lo redactó” (The machine says he copied. He says he wrote it) on 8 November, with the subtitle “Why a percentage is not proof”. It starts from a fictional case based on real situations: a 15-year-old student whose school accuses him of using AI in a history assignment because a detector gives it an 87% probability. He says he wrote it and still has his drafts, although he admits he used a spelling and grammar checker at the end.
The post argues that a detector provides an indication, not proof, and that it can be wrong in both directions. Before reaching a conclusion, it suggests listening to the student's explanation without either accepting or dismissing it, and then checking it: an oral explanation or redoing part of the exercise helps to verify what has been learned. As evidence of the process it cites drafts, versions, dates and notes, and it recommends that families be able to meet the form tutor to find out what the percentage means and what protocol the school applies. It also warns that an unfair accusation can cause anxiety and demotivation.
The text draws on UNESCO's guidance on generative AI in education and on the European AI Act, which agree that a person must review and interpret what automated tools say. On the same subject, the paper reported the publishing sector's warning about detectors, which spoke of a “witch hunt”.
An AI detector's percentage opens a conversation with the student, but does not prove they copied
A post on the blog “Familias con inteligencia” asks what to do when a detector flags the work of a student who says they wrote it themselves, and suggests backing the percentage up with evidence of the process.
Chema Falcó published “La máquina dice que copió. Él dice que lo redactó” (The machine says he copied. He says he wrote it) on 8 November, with the subtitle “Why a percentage is not proof”. It starts from a fictional case based on real situations: a 15-year-old student whose school accuses him of using AI in a history assignment because a detector gives it an 87% probability. He says he wrote it and still has his drafts, although he admits he used a spelling and grammar checker at the end.
The post argues that a detector provides an indication, not proof, and that it can be wrong in both directions. Before reaching a conclusion, it suggests listening to the student's explanation without either accepting or dismissing it, and then checking it: an oral explanation or redoing part of the exercise helps to verify what has been learned. As evidence of the process it cites drafts, versions, dates and notes, and it recommends that families be able to meet the form tutor to find out what the percentage means and what protocol the school applies. It also warns that an unfair accusation can cause anxiety and demotivation.
The text draws on UNESCO's guidance on generative AI in education and on the European AI Act, which agree that a person must review and interpret what automated tools say. On the same subject, the paper reported the publishing sector's warning about detectors, which spoke of a “witch hunt”.
