Monitoring student AI use without surveillance

[ MONITORING ]

Watch the work, not the person

Schools can see more of students' AI use than ever. The question is what to do with that view. The research points to one principle: watch the work, not the person. Tell students what is visible, keep personal chats out, and use what you see to help, not to label.

Where monitoring has gone wrong

In Lawrence, Kansas, more than 1,200 Gaggle alerts in 10 months led school officials to judge almost two-thirds of them non-issues, including more than 200 flags on homework. In Vancouver, Washington, a records request exposed nearly 3,500 unredacted alert documents, and at least six students may have been outed. RAND found "scant evidence" of either benefits or risks from such tools.

Students change what they say

About half of monitored high schoolers agreed they hold back their true thoughts or ideas online because they may be monitored, and 13% at monitored schools said they, or an LGBTQ student they know, had been outed by the software. 78% of teachers at monitored schools said the software helped identify students for discipline (CDT, via Education Week, 2022).

Families often do not know

88% of teachers say their school uses student activity monitoring, while only 45% of parents know (CDT, Jan 2025). Whatever your school monitors, say so plainly in the parent notice, including device monitoring that works outside any AI tool.

Labels mislead

Early-warning labels show the risk of turning data into judgments about students. Wisconsin's dropout early-warning system was wrong about 74% of the time when it predicted a student would not graduate on time, its false-alarm rate was 42 points higher for Black students and 18 points higher for Hispanic students than for White students, and the label had no significant effect on graduation. Learning dashboards have not done much better: across 38 studies, effects on achievement were mostly negligible or small.

How Nexus applies these rules

Students are told when a class's teachers can review a chat, personal chats are not shown to teachers, and Nexus records who opened each class conversation. Nexus does not run AI-writing detection, does not infer emotion or ability, and does not label students.

The teacher view "Where students need help" shows concepts with counts and dates only: no student is named or ranked, and groups under 3 students are hidden.