K-12Issue 653 records
K–12
A Nov. 10 Stanford SCALE update of 52 causal studies finds K-12 generative AI helps more when students use it to learn than to produce work, but neither use is consistently beneficial. It cites a large NBER trial of more than 6,000 Tennessee middle-school students in which an AI math tutor improved post-mistake correctness but slowed practice and showed only modest, marginal delayed-test gains. A Nov. 10 Texas 2036 report adds that AI use outpaces district policy and that districts, not vendors, carry student-privacy duties. Evidence is mostly preprints and non-U.S. or mixed-age.
- Evidence records
- 3
- Cross-source patterns
- 1
- Evidence classes
- 2 academic research1 independent research
- Outcomes
- 2 mixed1 cautionary
- Source freshness
- 2 new this fortnight1 undated
- Research completed
- 2026-11-11
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Learning effects depend on how AI is structured, not whether students have it
The Stanford review concludes effects depend on design and use, with learning-support use more promising than work production, and the NBER trial found added value only when a guard-railed tutor was embedded in a structured workflow with modest, marginal delayed gains. The Texas report likewise says learning improved where tools had safeguards, educator oversight and active-engagement methods.
Operating questionWhich specific student uses will the district permit, and what delayed unaided measure will show whether a tool is helping learning?
Supporting evidenceStanford University SCALE Initiative, AI Hub for EducationNational Bureau of Economic Research (Oreopoulos, Liut, Sungu and Low)Texas 2036
Evidence ledger
Full record · every source keeps its link and limitations
Stanford SCALE AI Hub for Education: The Evidence Base on AI in K-12, a November 2026 review of 52 causal studies
An update to the March 2026 review adds 32 strong causal papers published October 2025 to June 2026, for 52 in total. Reviewers conclude AI is more likely to support learning when students use it to understand material than to produce work, but neither use consistently improves learning, and offloading tasks can reduce learning. AI can improve learning experience and confidence without improving outcomes, and teacher-facing AI can cut planning time or raise feedback quality but one new study found it reduced student motivation and confidence.
Why it matters, evidence and limitations
- Why it matters
- Gives K-12 districts a current, independent synthesis for deciding where to allow student-facing AI and where to prefer educator-facing, structured uses. It supports requiring learning-support design rather than open-ended answer generation, and measuring outcomes instead of satisfaction.
- Evidence and measured results
- Review of randomized controlled trials and quasi-experimental studies drawn mainly from arXiv preprints in the AI Hub Research Repository. Of 868 papers added in the review period, 201 were RCTs or QEDs, 97 advanced to human review and 32 met the strong-causal standard (28 student-facing, 4 educator-facing). Five new student-facing studies were in U.S. K-12 settings. A language model did preliminary screening before Stanford researchers reviewed prioritized papers against a codebook based on What Works Clearinghouse standards.
- Limitations and uncertainty
- The authors say findings should not be treated as broadly generalizable across U.S. K-12 settings; the U.S. K-12 share of studies is limited. Many sources are non-peer-reviewed preprints, the repository is arXiv-weighted so relevant journal studies may be missed, and LLM-assisted screening may have excluded relevant studies. Some included studies involve undergraduates or adults rather than school-age students. This is a secondary synthesis; effect sizes are not pooled.
NBER working paper: randomized trial of an AI math tutor with more than 6,000 Hamilton County, Tennessee middle-school students
Students were randomized to AI versus computer-assisted-learning-only support, mastery versus non-mastery progression and math topic. AI students attempted fewer questions and moved more slowly but were more accurate per attempt, and AI improved next-attempt correctness after mistakes. On a delayed test one week later, gains were modest: about three percentage points on the practiced Exercise 1 item for mastery students, only marginally significant, with no overall delayed-learning gain. Mastery alone did not improve delayed learning.
Why it matters, evidence and limitations
- Why it matters
- Newly relevant because the Nov. 10 Stanford SCALE review cites it as one of its U.S. K-12 studies; it gives districts a large, in-school, U.S. trial showing that a guard-railed AI tutor adds limited value and only within a structured workflow.
- Evidence and measured results
- Randomized field experiment in March-April 2026 across 20 schools and just under 100 teachers (grades 6-8), one 50-minute practice session and a delayed assessment one week later, on the NUMI computer-assisted learning platform; the full paper reports about 7,000 students. The headline delayed-test result is described by the authors as only marginally significant.
- Limitations and uncertainty
- Working paper, not peer reviewed. One short session and a one-week delay; the authors say it does not estimate effects of repeated use. The delayed-test effect is marginally significant and concentrated on one item. Platform-specific (NUMI), single-region study, and the tutor was a guard-railed design, so results may not carry over to general chatbots. Sample size is stated as more than 6,000 in the abstract.
Texas 2036: In Ctrl of AI in Schools, a Texas policy report on AI governance, student safety, learning evidence and AI literacy
The report finds AI is already in Texas classrooms but largely unguided: in fall 2025, 86% of school and district leaders reported teacher AI use and 61% student use, yet only 29% reported a district policy on student AI use and 20% one on teacher and staff use. It argues FERPA obligations fall on schools rather than vendors, that student PII includes indirect identifiers, and that learning evidence is genuinely mixed. It recommends a state baseline for responsible use, pre-classroom vetting of tool safety and efficacy, educator AI literacy and student AI literacy.
Why it matters, evidence and limitations
- Why it matters
- Directly useful for district governance and procurement: it turns privacy law, tool design risks and mixed learning evidence into vetting requirements, and shows a state-level standard-setting gap that districts must otherwise fill individually.
- Evidence and measured results
- Policy report synthesizing the March 2026 Stanford evidence review, two meta-analyses (generative AI feedback; higher-order thinking), individual studies, and national and Texas surveys. Literature search ended August 2026; survey and policy data current to October 2026. The adoption and policy figures come from a Texas AI in Education Task Force report cited by the authors.
- Limitations and uncertainty
- Advocacy-oriented policy report from a Texas organization, not a primary evaluation. It states that most underlying evidence comes from outside the United States, reviews draw more on higher education than K-12, causal studies are few and some not peer reviewed, and findings are early signals. Adoption and policy percentages are self-reported leader survey figures. Legal statements are the authors' interpretation, not legal advice.
How to read this edition
Source findings, measured results and limitations come from the cited publications. Patterns, operating questions, role takeaways and implementation considerations are Lighthouse Advisory interpretation, stated as questions to validate locally rather than guaranteed outcomes. Vendor and operator claims are labeled as claims. Full research method.
- Academic research
- Research produced through an academic institution or peer-reviewed venue.
- Independent research
- Research conducted outside the implementing organization.