Skip to main content
v2026.11,986 entries · CC-BY 4.0
CASRAIAsk CASRAI · AI assistant for researchAsk a research question. Answers come with their sources.Funder rules, compliance, methods and publishing. Answers link the sources they rest on.

Editorial · CASRAI · AI and ML research outputs

Stanford SCALE Review of 52 Causal Studies Finds Mixed Learning Gains From AI in Education

A Stanford SCALE review of 52 causal studies finds mixed learning gains from AI tools, warning cognitive offloading can undermine student retention.

Published 15 Nov 2026· 5 minute read

Ask CASRAI · free to try

Ask about this story

2 free questions a day, no account.

An AI assistant for research: funder rules, compliance, methods and publishing. It links the sources behind its answers and labels answers that come from a web search.

Answers draw on CASRAI's guides and dictionary plus the federal and funder documents we index: Federal Register, Grants.gov, Regulations.gov and UKRI.

Works on this site and inside Claude, Cursor and the AI tools you already use.

Everything CASRAI publishes — this page, the dictionary, the guides and the news — stays open to read, with no account and no card.

CASRAI is the reference for research administration — bookmark it for the next question.

On November 10, 2026, Stanford University’s SCALE Initiative released an expanded evidence review examining 52 causal studies on artificial intelligence in education, finding that while AI tools frequently enhance real-time task completion and student confidence, durable learning gains remain highly inconsistent. The review cautions university researchers, institutional review boards, and evaluation teams that passive cognitive offloading often undermines conceptual retention compared to structured pedagogical scaffolding.

Expansion of the Causal Evidence Base

The report, titled The Evidence Base on AI in K-12: A November 2026 Review, was published by the AI Hub for Education within the Stanford Center for Opportunity Policy in Education (SCALE Initiative). Authored by Seth Walker, Lily Fesler, JP Martinez Claeys, Chris Agnew, and Susanna Loeb, the synthesis expands a March 2026 baseline of 20 papers by adding 32 newly published causal studies released between October 2025 and June 2026, bringing the total evidence base to 52 rigorous investigations.

Unlike broader edtech literature that relies on qualitative surveys or vendor white papers, the Stanford repository enforces strict methodological criteria, including only randomized controlled trials (RCTs) and well-identified quasi-experimental designs (QEDs). The authors highlighted a notable shift in setting: while the March 2026 review included zero student-facing causal studies in operational U.S. elementary and secondary schools, the November update captured the first five domestic K-12 trials, though most research remains concentrated in adult and higher education contexts.

Cognitive Scaffolding Versus Cognitive Offloading

The core finding across the 52 evaluated studies is that educational outcomes diverge sharply based on whether AI provides cognitive scaffolding or enables cognitive offloading. When tools deliver Socratic guidance, clarify misunderstandings, and structure problem-solving steps, students achieve modest gains in conceptual mastery comparable to human tutoring.

Conversely, when students use models to offload intellectual tasks—such as generating complete essays, code, or mathematics solutions—durable learning declines. While in-session task completion accelerates, unassisted post-tests reveal marked performance deficits compared with peers using traditional learning materials.

This dynamic is exacerbated by an affective illusion of competence. In multiple randomized experiments, students using AI reported elevated confidence, decreased anxiety, and perceived mastery, yet their objective exam scores failed to demonstrate measurable academic advantage, showing that generative fluency often conceals genuine comprehension deficits.

Educator Efficiencies and Classroom Dynamics

The review also evaluated educator-facing AI applications, noting administrative efficiencies alongside subtle instructional trade-offs. Generative platforms reduced teacher preparation time for lesson plans, reading level adaptations, and grading rubrics without compromising curricular quality as assessed by subject-matter experts.

Additionally, automated diagnostic systems providing feedback to novice tutors improved pedagogical technique and pupil outcomes. However, the November update noted that in classrooms where teachers relied heavily on automated instructional prompts, spontaneous teacher-student interaction declined, occasionally dampening student motivation.

Furthermore, voluntary student engagement proved a persistent bottleneck. Across evaluated platforms, students rarely logged the dosage required for measurable academic impact unless usage was mandatorily integrated into structured class periods.

Evaluation Metric March 2026 Review November 2026 Review Trajectory
Total Causal Studies 20 papers 52 papers 160% increase in causal literature
U.S. K-12 Student Trials 0 studies 5 studies First domestic field RCTs evaluated
Primary Subject Domains Math, programming Math, coding, writing, science Expansion across core academic subjects
Educator Evaluations Preliminary pilot data Structured rubric and tutoring trials Proven administrative time savings
Unassisted Retention Largely unmeasured Negative transfer during task offloading Demonstrated risk when removing tools

Evolution From Vendor Optimism to Methodological Rigor

The Stanford review provides an empirical counterweight to commercial edtech marketing. Following the rollout of foundational models in 2023, vendors widely promoted AI as an automated one-on-one tutor capable of rapid learning acceleration. Yet commercial claims overwhelmingly rested on correlational customer surveys lacking control groups.

To establish rigorous benchmarks, the Stanford SCALE Initiative established the AI Hub for Education Research Repository in early 2025. The November 2026 synthesis highlights that while research volume has surged, true causal evidence remains limited, demonstrating that novelty effects during active tool use rarely translate into lasting cognitive transfer.

Education Researchers and Independent Outlets Weigh In

Independent coverage by Education Week underscored the tension between school district technology procurement and the finding that homework offloading undermines student retention. Analysts emphasized that districts have deployed significant funding without established pedagogical guardrails governing home use.

Similarly, reporting in Chalkbeat highlighted equity concerns, noting that under-resourced schools often lack the staffing needed to guide student AI interactions effectively. Researchers quoted across both outlets echoed Susanna Loeb’s conclusion that AI is not a uniform intervention, but a medium whose efficacy is defined by how educators structure student engagement.

Implications for University Researchers, IRBs, and Evaluation Teams

For university researchers and evaluation offices, the review establishes clear methodological requirements. Educational AI studies must incorporate delayed, unassisted post-tests to distinguish real skill retention from transient tool assistance.

Investigators should formalize outcome measures and analytical models in advance by completing a study protocol preregistration, preventing selective reporting in a field facing commercial pressure.

Institutional review boards must also adjust evaluation standards. As outlined in CASRAI’s analysis of the IRB review process, protocol assessments in classroom environments must evaluate cognitive risks and control-group fairness alongside privacy protections when students are exposed to task-offloading software.

Finally, experimental reproducibility requires detailed sharing of prompt templates, model versions, and interaction logs. Researchers should support their findings with an open data availability statement and maintain compliance via institutional responsible conduct of research training programs.

Next Steps in Empirical AI Evaluation

The Stanford AI Hub for Education will continue updating its repository monthly, planning its next major synthesis for mid-2027. Federal agencies, including the Institute of Education Sciences and the National Science Foundation, are prioritizing grant funding for multi-site causal evaluations, while multi-year cohort studies begin tracking whether sustained cognitive offloading generates cumulative deficits in foundational literacy and numeracy.

Sources

Referenced across the research world

University of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logoUniversity of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logo
  • University of Cambridge logo
  • Columbia University logo
  • Crossref logo
  • University of Edinburgh logo
  • Harvard University logo
  • University of Oxford logo
  • Princeton University logo
  • Stanford School of Medicine logo
  • University College London logo
  • ORCID logo

View CASRAI adoption →