What researchers studied

Randomized controlled trial of a real-time generative-AI support system used by human tutors. More than 700 tutors and 1,000 students from underserved communities, with analysis of more than 350,000 tutor-student messages

What they found

  • Students whose tutors used Tutor CoPilot were 4 percentage points more likely to master math topics.
  • The estimated gain was 9 percentage points among students assigned to lower-rated tutors.
  • Message analysis found greater use of probing questions and less generic praise among tutors using the tool.
  • The study evaluates AI as real-time support for human tutors rather than as an autonomous tutor replacing them.

What the study does not prove

  • This is an EdWorkingPaper and should not be described as a peer-reviewed journal article unless a later published version is verified.
  • The primary outcome is math-topic mastery in a specific tutoring context, not proof of long-term standardized achievement gains.
  • The effects should not be generalized to every generative-AI tool, tutor population, subject, or instructional setting.
  • The study does not establish that AI recommendations should override educator judgment.

Evidence strength: Strong experimental evidence; randomized controlled trial; EdWorkingPaper.

Why this matters for families

This study supports a human-led model of AI assistance. The promising result came from helping tutors ask stronger questions and support student thinking, not from removing the tutor from the learning process.

Noor interpretation

How Noor translates the evidence into practice

AI is most compelling when it helps a human educator make stronger instructional moves while the educator remains responsible for evaluating the suggestion and deciding what happens next.

Use AI to surface possible questions or interventions, but require the educator to accept, modify, or reject the recommendation. Track whether the suggestion produced better reasoning, mastery, retention, or transfer instead of treating AI use itself as a quality metric.

Read the original source

Noor links to the original or authoritative source so families can distinguish the evidence itself from our interpretation.

Open original source →

DOI: 10.26300/81nh-8262

Research notes

Tutor CoPilotAI tutoringhuman-AItutor qualitymath masteryrandomized trial

No single study determines a student's plan. Noor uses research as one input alongside the learner's goals, observed performance, academic context, and response to instruction.