K-12Issue 304 records
K–12
Tonight's K-12 signal is that AI's value and its risks both depend on how districts run it, not just on access. A new English trial found a curriculum-aligned AI planning tool cut primary teachers' lesson preparation by about 49 minutes a week without lowering quality, but few teachers kept using it. A Washington survey links clear policy and useful training with teacher confidence. Meanwhile, Fairfax built its own browser extension because Google's admin console cannot switch off AI in search results. Orlando's board is asking whether Florida's opt-in and 30-day record rules are worth the cost and classroom strain.
- Evidence records
- 4
- Cross-source patterns
- 2
- Evidence classes
- 2 independent research2 independent reporting
- Outcomes
- 1 effective1 emerging1 cautionary1 mixed
- Source freshness
- 4 new this fortnight
- Research completed
- 2026-10-07
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
AI built into vendor platforms is outrunning the controls districts need to enforce their rules
Fairfax had to write its own Chrome extension because Google's admin console cannot switch off AI Overviews in search. In Orlando, the CIO said direct student use can include voice, image generation and embedded feedback, and the district is still asking vendors how to retain and return 30 days of student AI interactions. In both districts, policy depends on vendor features the districts do not control.
Operating questionFor each AI feature our students can reach, which control enforces our policy, and which vendor feature are we relying on?
Supporting evidenceWTOP News; source statement from Fairfax County Public Schools (Department of Information Technology)Central Florida Public Media; sources include Orange County Public Schools board work session and CIO Maurice Draggon
Teacher AI value depends on fit and support, not access alone
The English Aila trial found real time savings, yet use was low and fell over ten weeks, and gains concentrated among newer or less confident teachers. The Washington survey found that teachers with clear policies or useful training reported more confidence, and that attitudes split by subject. Neither source shows that simply providing AI tools changes practice.
Operating questionWhich teachers and subjects are we targeting, and what policy and training will sustain use beyond the first weeks?
Supporting evidenceEducation Endowment Foundation (funder and publisher); National Foundation for Educational Research (evaluator); Oak National Academy (tool developer)Central Washington University College in the High School faculty (Josh Aubol, Steve Stein, Peter Klosterman); published via Washington State Standard and Seattle's Child
Evidence ledger
Full record · every source keeps its link and limitations
EEF/NFER RCT: Oak National Academy's Aila cut primary teachers' lesson-preparation time by 24% without lower resource quality, but use was low and declining
A randomised controlled trial in 108 English primary schools (464 teachers randomised; the project page reports data from 416) allocated Key Stage 2 teachers to use or avoid Aila, a free AI lesson-planning tool, for ten weeks in autumn 2025. Aila teachers spent about 2.5 hours a week on lesson and resource preparation versus 3 hours 19 minutes, 76% of the comparison time, saving about 49 minutes a week, and their resources were no worse in quality. The trial received a high security rating. Benefits were greater for newer or less confident teachers. Use was low: 81% tried it, it was used for about 7% of lessons in an average week, use declined, and fewer than one in five felt it meaningfully changed their planning or teaching.
Why it matters, evidence and limitations
- Why it matters
- New, rigorous evidence on teacher-facing AI published today. It supports bounded time-saving claims for U.S. districts weighing teacher AI tools, while showing that free access alone did not sustain use. Transferability is limited: English curriculum, a purpose-built publicly funded tool, and a comparison group that could still use other AI.
- Evidence and measured results
- Independent RCT evaluated by NFER, with weekly teacher diaries, surveys and independent teacher assessments of resource quality; EEF high security rating. No pupil outcome measure; time is self-reported in diaries.
- Limitations and uncertainty
- England only; KS2 primary teachers; self-reported time; comparison group free to use other AI; low and declining use; developer is publicly funded and quoted in the release. Report PDF not opened; methods details from the linked project page.
WTOP: Fairfax County Public Schools builds its own Chrome extension to hide Google AI Overviews because Google's admin controls cannot turn them off
Fairfax County Public Schools, Virginia's largest district, says its filters already block student access to ChatGPT, Claude and Gemini, but Google's administrator controls give no way to disable AI Overviews in Google Search. Its IT department therefore built the FCPS Customizer, a Chrome extension that hides generative AI overviews, search summaries and image-search features on school-issued Chromebooks. It is rolling out at 11 elementary and middle schools first, with full elementary and middle school coverage expected by November. The district reports positive teacher feedback. The board banned generative AI for elementary students last summer, while teachers have a ChatGPT for Teachers pilot.
Why it matters, evidence and limitations
- Why it matters
- Shows a concrete control gap: a district policy that bans student generative AI cannot be enforced through the vendor console when AI is built into search. A large district is filling the gap with in-house engineering, which most districts cannot do.
- Evidence and measured results
- Short news report based on a district statement; first reported by the Wall Street Journal. No usage data, effectiveness measures or teacher-feedback numbers are given.
- Limitations and uncertainty
- Based on a district statement; extension effectiveness, bypass rates and maintenance cost are not reported. Google's admin options may change. Single district.
Central Florida Public Media: Orange County Public Schools maps Florida's AI rule and board members question opt-in workload, equity and 30-day record costs
At an Oct. 6 work session, OCPS staff said existing policy already covers human review, approved tools, protected data, academic integrity and professional learning, but Florida's rule (18 requirements, effective July 1, 2027) adds parental opt-in, extra review of AI tools for pre-K through grade 5, record-keeping, reporting and vendor requirements. Families who do not opt in must get a comparable non-AI alternative. The CIO said direct use can include voice interactions, image generation and embedded AI feedback; district-reviewed generative tools include Adobe Firefly, Conmigo and Google Gemini Notebook. Board members raised teacher workload in split classrooms, unequal learning experiences, storage costs for 30 days of student AI interactions during budget cuts, and paying for unused tools.
Why it matters, evidence and limitations
- Why it matters
- The archive holds Florida's rule and legal analysis; this is the first account of a large Florida district working through it in public, with concrete tool names, cost worries and equity objections from board members.
- Evidence and measured results
- Public-media reporting of a board work session with named officials' quotes. No cost figures; the CIO said costs depend on scope. Corroborated by WKMG News 6 coverage of the same session (not used as a separate source).
- Limitations and uncertainty
- News coverage of a work session, not a policy decision; costs not quantified; district awaits further state guidance (superintendent meeting the department this week).
CWU College in the High School survey: 222 Washington math and English teachers split on AI, and clear policies and useful PD track with greater confidence
In a May 2026 survey of 222 Washington high school math and English teachers, 57.0% of English teachers and 37.6% of math teachers said their attitude toward teaching had worsened since AI appeared; 83.7% of English and 54.9% of math teachers reported academic integrity concerns. English teachers were nonetheless more confident teaching responsible AI use (45.5% vs 24.4%) and more likely to require AI experimentation (34.1% vs 14.6%). Teachers with clear institutional AI policies, or useful AI professional development, reported more confidence using AI and teaching students to use it. About one-sixth of responses mentioned moving writing to in-class pen and paper, and some teachers had stopped assigning homework.
Why it matters, evidence and limitations
- Why it matters
- Adds recent U.S. teacher-level evidence that subject matters and that policy clarity and relevant training go with teacher confidence, as districts nationwide write AI policies this year.
- Evidence and measured results
- Survey of 222 teachers in one university dual-credit programme, reported by its authors in an op-ed. Associations between policy, PD and confidence are correlational; sampling method, response rate and statistical tests are not given.
- Limitations and uncertainty
- Non-random sample from one programme; authors' commentary; correlational findings; republished article whose original appeared in Washington State Standard. Survey conducted May 2026.
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.
- Independent research
- Research conducted outside the implementing organization.
- Independent reporting
- Independent reporting with attributable sources but without a formal evaluation design.