What researchers studied

Instructor-level randomized trial conducted during one semester across matched course blocks at a large U.S. public university. 2,379 undergraduate students, 30 instructors, and 34 course sections spanning multiple disciplines, with an exact-match analysis of 1,353 students in sections of the same course

What they found

  • Within sections of the same course, access to the course-integrated AI study assistant reduced final grades by 0.37 standard deviations.
  • Access reduced recorded participation in instructor-designed online course activities by 0.90 standard deviations, with additional declines in course page views and active days.
  • Only about 15 percent of treated students used the study assistant at least once, so the experiment estimates the effect of introducing access and encouragement rather than the effect of sustained individual use.
  • Estimated grade losses were more than twice as large for first-generation students as for continuing-generation students, although subgroup evidence should be interpreted cautiously.

What the study does not prove

  • The trial took place at one university during one semester and evaluated one GPT-4o-based tool configured by participating instructors.
  • Most instructors left the tool in direct-instruction mode rather than a more scaffolded tutoring mode, so the findings do not apply to every AI tutor design.
  • Students in both conditions could use other university-supported AI tools, and introducing the study assistant may have changed broader AI use beyond the specific tool.
  • Course grades were instructor-assigned rather than independently administered learning assessments, and platform activity captured only part of student engagement.
  • Randomization occurred at the instructor level, and matched sections could still differ in grading and participation opportunities.

Evidence strength: Large multi-course randomized university trial with an especially credible exact-course comparison, tempered by low tool uptake, one institution, one semester, and outcome-measure limitations.

Why this matters for families

If an AI tool makes college work feel faster, check what it may be replacing. A student still needs to read the course material, attempt the problem, participate, and explain the answer independently.

Noor interpretation

How Noor translates the evidence into practice

A tool can be integrated into a course and still displace the effort, participation, and feedback that learning requires. The useful question is not whether AI is present, but whether it keeps the student thinking and connected to the instructional sequence.

Noor Lyra should position AI, when used, as a supervised support inside a human learning plan. The educator should require the student's own attempt, verify understanding, protect engagement with course materials, and decide what happens next. Noor should not claim that AI tutoring always lowers grades or that its own approach produces the opposite effect.

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/y3f8-vh05

Research notes

AI tutoringhigher educationstudent engagementcollege supportrandomized 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.