From the K–12 edition of September 24, 2026
National principal survey: AI use outruns policy/training, with disadvantage and sector integration gaps
Christopher Campos (University of Chicago) and John Singleton (University of Rochester) · Academic economics / education research · United States (national sample of about 1,200 traditional public, charter, and private K-12 schools)
- Publisher
- Author PDF (cqcampos.com) / NBER w35347
- Original publication
- June 2026 (working paper; Booth Review coverage 11 Sep 2026)
- Source retrieved
- 2026-09-25
What happened
Campos and Singleton’s June 2026 working paper reports an original national survey of K-12 principals (fielded Dec 2025 to Feb 2026) measuring institutional AI integration. Teacher AI use spread rapidly while scaffolding lagged: 58% of principals report no written generative-AI policy; 76% report no AI-tutoring license/contract; 67% did not use ESSER funds for AI. An integration index combining policy, training, student-use guidance, leadership, and tools shows schools serving more disadvantaged students score 0.07 to 0.11 SD lower per SD of disadvantage, and charter/private schools score 0.23 to 0.44 SD lower than traditional public schools. District size accounts for roughly one-third of the public-school disadvantage gap; principal enthusiasm and budgets raise integration levels but do little to close gaps. Why now: after 2026-09-23 coverage of NYC/LA pauses reverberating nationwide, this evidence reframes pause-versus-enablement as a system-capacity equity problem, not merely a values debate.
Why it matters
Gives SLED buyers an equity lens for GenAI: buying tools without district-scale implementation capacity can widen integration gaps for disadvantaged and non-district schools. Supports engagements that build shared policy, PD, and vendor infrastructure rather than school-by-school experimentation.
Evidence and measured results
Original survey with school-tied invitations merged to federal demographic data and reweighted; randomized participation incentives used to probe selection; correlational gaps with controls; not an achievement RCT.
Limitations and uncertainty
Principal self-report; integration is not learning outcomes; June 2026 vintage requires why-now framing; charter/private lag may partly reflect different governance models.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-09-25; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- AI tools spread school-by-school while disadvantaged and non-district schools lag on policy, training, and contracted supports.
- Stakeholders
- State chiefs
- District superintendents
- Charter/private leaders
- ESA/co-op directors
- Equity officers
- Discovery questions
- Do you have a written GenAI policy today?
- What share of schools have AI tutoring under contract?
- How does integration differ by FRPL or neighborhood income?
- What shared services exist for small/charter schools?
- Value hypothesis
- A system-level integration program can raise scaffolding where school-level enthusiasm cannot close gaps.
- Potential engagement
- 6 to 8 week AI integration equity diagnostic: index baseline, gap analysis, shared-service options, and board-ready roadmap.
- Unsupported claims
- That closing the integration index raises achievement
- Guaranteed equity gains from any single vendor
- That small districts can match large-district capacity without shared services
Pre-sales engineering
Role takeaway
- Fit
- District/consortium platforms with shared identity, logging, and contracted AI services; poor fit for unmanaged consumer tools.
- Architecture and integration
- Central IdP, approved-model gateway, optional tutoring SIS integrations, and school-site configuration baselines.
- Prerequisites
- Written GenAI policy
- Procurement vehicle for DPAs
- PD delivery channel
- Equity-targeted rollout list
- Constraints
- Survey is correlational
- Sector governance differences
- Rapid product change
- Security
- Prefer contracted processing with no-training clauses; inventory and disable unmanaged consumer AI on managed devices where policy requires.
- Proof of value
- Policy coverage by school type
- Training completion in high-disadvantage schools
- Share of AI use under contract
Delivery
Role takeaway
- Work
- Draft/adopt policy, stand up PD, negotiate contracted tools, and publish student-use guidance with equity monitoring.
- Dependencies
- Legal/privacy
- Finance/procurement
- ESA or co-op partners for small systems
- School leader communications
- Ownership
- Superintendent/cabinet owns integration program; equity office monitors gap metrics.
- Skills and adoption
- Train principals and teacher champions together so adoption is not left to individual tinkering.
- Governance checkpoints
- Policy adoption
- Contract execution
- PD wave completion
- Quarterly gap dashboard
- Acceptance criteria
- Written policy in force
- Documented training reach in target schools
- AI tutoring only via approved contracts where offered
- Risks
- Shallow adoption labeled as success
- Charter/private exclusion
- Scale bottlenecks without shared services
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Favor district- or consortium-level approved platforms, shared DPAs, and common PD pipelines over fragmented school-site consumer accounts—especially where district scale is the scarce resource.
Governance
Who approves, reviews and stays accountable for outcomes?
Written policies, student-use guidance, and training must be treated as core integration investments; interventions that only nudge principals or free discretionary spend are unlikely to close disadvantage gaps.
Security and privacy
What data, permissions and controls need testing?
Paper measures institutional scaffolding, not vendor security posture; shallow integration implies many schools may be using tools without contracts—raising privacy/procurement risk.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Lower integration in higher-disadvantage schools risks unequal access to guided AI supports and unequal exposure to unguided risks; workforce PD capacity is part of the gap.
Procurement
What should contracts, pricing and exit terms secure?
Prefer cooperative purchasing and shared DPAs that extend large-district capacity to smaller and charter systems; budget for training/policy as complements to licenses.
Operating model
Which teams own the service once it runs?
Shift from bottom-up teacher experimentation to system operating models where policy, PD, and contracted tools co-evolve.
Publication history
- 2026-09-24K–12 · Issue 195 resources
Stable resource ID: campos-singleton-ai-diffusion-gaps-2026