AI feedback helped teachers build on what their students said

A student explains an idea, and the teacher has a choice: move on, ask for more detail or use that contribution to advance the discussion. AI feedback tools are making those small teaching decisions easier to revisit. M-Powering Teachers analyses lesson transcripts and highlights moments that a teacher or coach can discuss afterwards.[5]

Listening becomes something a teacher can inspect

The research calls building on someone else’s contribution uptake. It can include reformulating a pupil’s explanation or asking a follow-up question that depends on what the pupil actually said. The original measurement work trained a language model to distinguish the real next reply from randomly selected replies. This estimates how closely the teacher’s response depends on the student’s contribution, reaching beyond repeated vocabulary.[1]

Consider an illustrative exchange about a pupil’s method for comparing fractions. A generic acknowledgement supplies little information about whether the teacher followed the method. Asking the pupil to explain the denominator they chose connects the next step to their reasoning. The model’s task is to recognise patterns of that kind across a transcript, providing a starting point for reflection rather than a complete judgement of a lesson.[1]

Three experiments, three different interventions

The Code in Place paper also estimated a 13.2% increase among instructors whose feedback viewing was changed by the reminders. That is a separate estimate derived from the encouragement design. It does not mean that randomly giving every teacher access produced a 13.2% improvement. The study involved a five-week introductory programming course; evidence on student satisfaction and assignment completion was suggestive, so the clearest result concerns instructors’ responses.[2]

On Polygence, mentors supported high-school students through research projects. Their private feedback pages included transcript examples and a space to reflect on what to try next. The study analysed 5,037 sessions. Students also reported better experiences and greater optimism about their academic future, but those survey results do not establish gains on an academic achievement test. Here, the practical change was a more responsive conversation in an ongoing mentoring relationship.[3]

The Utah experiment moved into physical maths and science classrooms. Its intervention included AI identification of focusing questions, plus weekly emails for which expert maths coaches manually selected examples. Other measured dialogue practices did not improve significantly, and the study did not establish better pupil attainment. The result therefore belongs to a specific package of automated analysis and human-selected feedback.[4]

The transcript helps a coach find the useful moment

Harvard’s description of M-Powering Teachers shows how this can fit into coaching. The platform charts who spoke, identifies mathematical vocabulary and flags questions and student reasoning. A coach can return to a particular exchange and discuss why it worked, or how a follow-up might open the discussion. This makes the recorded lesson a shared reference for a professional conversation.[5]

As of October 2026, Harvard describes M-Powering Teachers as an active research project and pilot combining AI feedback with traditional mathematics coaching. In that setting, a highlighted exchange can become a specific goal for the next lesson, with a coach helping interpret why it matters. The useful test is whether that process helps teachers build habits that persist over time.[6]

Report an error or suggest a correction