Recognizing & Responding to Suspected AI Use
AI Committee · Session Introduction
AI Committee Framing Question
That’s still the question underneath everything today. Detecting AI use is step one — it doesn’t tell you whether it was a problem.
The day opened with a conversation led by Ann, Jeff, and Greg on what is working for teachers, followed by a two-part introduction and three 50-minute breakout rotations.
Part 1
AI use is rarely provable outright; the protocol establishes likelihood, not proof. Many students don’t yet know where the line is, so use the moment to show them, not just to catch them.
Approach with Curiosity
Observe & Document
Next Steps
Not: “Did you cheat?” Instead: “Walk me through your process.”
Part 2
Why detection alone doesn’t hold up, and what to design instead. Full tell-tale signs and detector research are on the Detecting AI Writing page.
74%
of lightly-paraphrased AI text goes undetected by AI checkers
61%
of non-native speakers’ essays wrongly flagged as AI-generated
15%+
false-positive rates found in independent studies, against the <1% vendors target
Bottom line: treat any flag as one data point. Never proof. If we can’t out-detect it, we design so it doesn’t matter.
Design it differently
Watch the process, not just the product
Detection isn’t proof.
Process evidence beats detectors.
Specificity is the lock.
Prevention beats policing.
Five sessions, pick three. 50 minutes each, hands-on with the tools.
Designing a New Unit or Course with AI
Data, Decisions, and Time Back: AI for the Other Half of Teaching
Meeting Students Where They Are: Differentiation & Tutoring with AI