What researchers studied

Observational yearlong district implementation using machine learning weighted propensity scores to compare achievement growth among users and observationally similar lower-use or non-user students. 6,065 middle school students in Deer Valley Unified School District, Arizona, during the 2025-26 school year

What they found

  • Students averaging at least two solved problems per week scored 0.109 standard deviations higher on within-district standardized spring MAP math growth than the weighted comparison group.
  • Students solving 10 to 19 problems per week showed an adjusted difference of 0.18 standard deviations, while students solving at least 20 problems per week showed 0.31 standard deviations.
  • About half of the analytic sample reached the two-problem weekly threshold, and 72.6 percent of eligible teachers activated the platform.

What the study does not prove

  • The study was observational. Propensity weighting balanced measured characteristics but could not remove bias from unmeasured differences in students, teachers, motivation, or implementation.
  • The platform included activities with and without its AI tutor, so the estimates do not isolate the effect of AI tutoring itself.
  • Thirteen percent of platform users could not be linked to district records, and indirect teacher use may have exposed comparison students to platform-generated materials.
  • The evidence comes from one Arizona district during one school year, and the largest dosage estimates may partly reflect who chose to use the platform most.

Evidence strength: Large district implementation with baseline adjustment, propensity weighting, dosage analysis, and robustness checks; observational selection, incomplete linkage, treatment contamination, and one-district context prevent causal claims.

Why this matters for families

Ask not only whether an AI tool is available, but who is guiding its use, how consistently the student engages, and what learning evidence informs the next lesson.

Noor interpretation

How Noor translates the evidence into practice

Technology becomes instruction only when an educator connects it to the learner's actual misconceptions, assigns meaningful practice, and uses the resulting evidence to decide what happens next.

Noor Lyra combines carefully selected tools with educator judgment, direct explanation, work review, and a learning sequence built around the individual student.

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/vcdh-yc23

Research notes

AI in educationmathematicsmiddle schoolteacher guidancelearning technology

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.