{"resourceId":"stanford-ai-tutoring-spectrum-2026","versions":[{"version":"external-6a699404547feef3c4d29c3c0b5064bf83a07b824e424933ab1c1b7e3bf9557d","resource":{"id":"stanford-ai-tutoring-spectrum-2026","title":"Stanford tutoring brief distinguishes educator assistance from unsupervised student AI","organization":"SCALE Initiative and National Student Support Accelerator, Stanford University","sector":"K–12 primary and secondary education","geography":"United States guidance drawing on US and international research","publishedAt":"August 20, 2026","publicationDate":"2026-08-20","eventDate":null,"sourceName":"AI Tutoring is Not a Monolith: What We Actually Know","sourceLabel":"Research synthesis and practice guidance","sourceUrl":"https://scale.stanford.edu/research-in-action/ai-tutoring-not-monolith","evidenceClass":"standards-guidance","outcomeClass":"cautionary","topics":["knowledge-work","data-security","governance-procurement","accessibility-workforce","operating-model"],"finding":"The brief distinguishes human-led AI-assisted tutoring from AI-led and AI-only models, arguing that evidence becomes thinner as sustained human involvement falls.","sledRelevance":"Interpretation: relevant to fall district purchasing and tutoring program design. August publication is new to the checked archive, not a September event.","evidence":"Synthesis of tutoring research plus interviews with providers, developers and researchers; no new pooled effect estimate or representative interview sampling frame is reported.","architectureImplications":"Interpretation: classify who sees and approves AI output, then configure permissions and intervention routes for that workflow.","governanceImplications":"Interpretation: require buyers to name which parts of human tutoring a proposed service preserves or changes.","securityPrivacyImplications":"Interpretation: execute and verify data-use terms before student access; test transcript export and deletion and define who can inspect conversations.","caveats":"Practice guidance, not product certification or a new causal evaluation. Provider interview contributions are disclosed. One link attached to a two-trial claim opened a different UK study; this edition does not repeat that numerical claim.","streamIds":["k12"],"roles":{"sales":"Interpretation: District tutoring leaders, intervention teams, finance and family representatives need comparable descriptions of proposed services. Ask who maintains the student relationship, who notices disengagement, and what evidence matches the exact delivery model. Offer an options assessment distinguishing tutor assistance from direct student automation. The value hypothesis is avoiding a mismatch between a purchased service and the district's intended intervention. Do not transfer human-tutoring effect sizes to software or claim Stanford endorses a vendor. A short evidence review can identify a bounded pilot, but this guidance alone cannot establish educational return or local staffing capacity.","engineering":"Interpretation: Translate each tutoring model into a data-flow diagram with the human decision points visible. Reuse existing learning and identity systems where appropriate, with limited transcript access and a tested help route. Prerequisites include defined student groups, available adult support and usable curriculum materials. Validate the whole interaction: login, approved practice, disengagement detection, human handoff and deletion. A proof of value should measure participation and independent performance separately, including learners using accommodations. The brief does not establish a preferred model provider, hosting topology or infrastructure capacity; those require product documentation and local testing.","delivery":"Interpretation: The intervention lead should own a pilot plan with teachers, tutors, privacy and accessibility staff. Identify who monitors attendance, contacts students and handles distress or inappropriate output. Train staff on those responsibilities before student access, and communicate the limits of AI assistance to families in accessible language. Proposed acceptance: sampled students reach an empowered adult through the published route, transcript controls pass agreed tests, and independent outcome reporting accompanies usage data. Review staffing and equity findings before expanding. Risks include treating supervision as nominal, underfunding follow-up and allowing attractive engagement metrics to displace the learning objective."},"retrievedAt":"2026-09-07T02:48:23Z","enrichedAt":"2026-09-07T02:52:08Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Interpretation: preserve accessible human help and test whether a student can summon it without relying on sophisticated self-advocacy.","procurementImplications":"Interpretation: compare services by evidence for their actual delivery model, not the generic label AI tutoring.","operatingModelImplications":"Interpretation: assign attendance, engagement and escalation owners and include supervision costs in the service plan.","updateExplanation":"Not an archive repeat. Newly discovered source adds implementation or evidence-quality context for the 2026–27 school-year review; publication timing is stated separately.","sourceVerification":{"openedUrl":"https://scale.stanford.edu/research-in-action/ai-tutoring-not-monolith","referenceExcerpt":"High-impact tutoring remains defined by live human-led instruction.","promptVersion":"sled-research-v3.1","model":null,"basis":"agent-reported inspection"}}}]}