14 states have published zero AI education guidance for schools. We scored the other 37. Here's what we found.
Every score, every source, every rubric weight. Free to cite with attribution to Kevin J. Roberts and Henry Dan.
Scored on a 7-criterion rubric adapted from the TeachAI Toolkit (Code.org, ISTE, Khan Academy, WEF), the UNESCO AI Competency Framework (2024), and the AI4K12 Five Big Ideas (AAAI + CSTA). Rated by Kevin J. Roberts, M.A. and Henry Dan, B.S. (Data Science). Full methodology ↓
"Schools are using AI to watch students. Most haven't figured out how to teach them with it. That's the wrong order."— Kevin J. Roberts, America's AI Report Card, April 2026
Your child used AI on their homework last night. Their teacher has no framework for it. Their state has almost certainly never asked the most important question: what should students still do themselves before they turn to AI?
This is not a future problem. It is happening in every classroom in America right now, without guidance, without standards, and without anyone responsible for the answer.
Across 37 scored jurisdictions, Assessment Adaptation averages 5.73 out of 10 and Student Effort / Cognitive Development averages 6.11. The other five criteria average 6.65 or higher.
Average score across all 37 scored jurisdictions. How to grade students in an AI world — what counts, what doesn't, how to know the difference — is the single most neglected topic in all of American AI education policy. 26 of 37 jurisdictions (70%) score 6 or below on this dimension.
From the researcher (Kevin J. Roberts): If a student submits an AI-written essay, no state document scored in this study specifies how to grade it. Of 37 scored jurisdictions, 26 score 6 or below on Assessment Adaptation, and no state scored 9 or 10.
Does state guidance address what students should still do themselves before turning to AI? Does it ask what happens to a student's brain when they outsource thinking too early? 23 of 37 jurisdictions (62%) score 6 or below on this dimension.
Vermont is the only state whose published guidance (January 23, 2026, 50-page document) explicitly names "cognitive offloading" as a risk and provides developmental guardrails for it. The other 36 scored jurisdictions never asked the question. The Effort Crisis was written because no one was.
This criterion is adapted from UNESCO's "Human Agency" competency and TeachAI Principle #7 (Prioritize Human Connection).
Every other criterion averaged 6.65 or higher. These two did not.
Across 37 scored jurisdictions: AI Literacy 7.08, Critical Thinking 6.86, Teach vs. Ban 6.84, Teacher Training 6.81, Equity 6.65. Student Effort 6.11. Assessment Adaptation 5.73. The gap between the highest and lowest dimension averages is 1.35 points.
36 states published formal AI education guidance and were scored on a 7-criteria weighted rubric (1–10). 14 states have not published guidance and are categorized by tier. 36 + 14 = 50 US states. Puerto Rico is also included as a scored US territory. A score of 7.0+ means a state is actively teaching students to think with AI. Below 6.0 means guidance is primarily focused on restriction and compliance.
Note on rankings: Rankings cover 36 U.S. states plus Puerto Rico (37 scored jurisdictions total). New Jersey ranks #37 — dead last with a 2.30. Ohio dropped to #36 at 4.40. The 14 U.S. states that published nothing carry no rank at all. Ranked last and unranked are two different things.
36 states scored + 14 states without guidance = 50 US states
Puerto Rico is also included above as a scored US territory. The 14 US states below have not published formal guidance.
⏳ Legislation Passed — Guidance Pending
Idaho — SB 1227 signed, guidance due July 2026 · Iowa — bill moving, deadline 2028 · Texas — Responsible AI Governance Act passed
🔄 Emerging Activity — No Published DOE Guidance
Arkansas, Florida, Illinois, Kansas, Maryland, Nebraska, New Hampshire, New York, South Carolina, South Dakota — task forces, city-level guidance, university frameworks, or coalition documents. No formal state DOE document.
⚠️ Safety Legislation Only — No Educational Framework
Pennsylvania — passed laws protecting students from AI deepfakes and harmful chatbots. No guidance on how to teach students to use AI.
The National Picture — All 50 States + Puerto Rico
All 50 States + Puerto Rico — Alphabetical
Find your state. Scored states show their rank and score. Unscored states show their status.
All 50 states + Puerto Rico. Scored states show their rank and score. ⏳ = legislation passed, guidance pending. 🔄 = emerging activity, no published DOE document. ⚠️ = safety legislation only, no educational framework.
* Rankings include Puerto Rico as a scored jurisdiction. New Jersey is #37 of 37 scored jurisdictions, including Puerto Rico. Ohio is #36.
Only five states score 8+ on cognitive effort, the work students must still do themselves to build working brains. Massachusetts, Vermont, New Mexico, Utah, and West Virginia lead. The rest treat student effort as an afterthought.
Only three states legislated AI mandates: Ohio, Tennessee, and Utah. Two of them rank in the bottom three on guidance quality. Mandating action does not mean mandating quality.
states issued AI guidance voluntarily. Districts can ignore it.
Among 183 documented early-adopter districts, 86% offer sustained teacher AI PD. Those 183 represent under 1.5% of the roughly 13,000 U.S. districts.
of U.S. districts are documented AI early adopters
In the CRPE deep-analysis subsample of 79 adopters, only 27% share AI-course information with students. Only 8% have adjusted learning standards to reflect AI.
of adopters adjusted learning standards for AI
Gaggle monitors about 6 million students across ~1,500 districts. GoGuardian reaches 7,000+ per a 2021 Senate investigation. Eight times more districts surveil students than teach them AI.
districts monitor with AI, vs. 183 teaching with it
Iowa City built required AI curriculum without state help. San Ramon Valley students built their own AI study app. Bullitt County students built an anti-bullying AI tool.
leading while the other 99% figure it out
18 of 37 jurisdictions have not updated AI guidance since 2024. Oregon has not updated since 2023. GPT-4o, Gemini 1.5, Claude 3, and DeepSeek all shipped in the meantime.
running on guidance from 2024 or earlier
183 publicly documented AI early-adopter districts are tracked in the CRPE 2025-26 Early Adopter Database. 183 districts is less than 1.5 percent of the roughly 13,000 U.S. school districts. Their AI implementation decisions, and the cases where those decisions have been withdrawn, are the only publicly documented record of district-level K-12 AI at scale.
Home to Seckinger High School, the nation's first AI-themed high school. Built a K–12 AI Learning Continuum and is expanding districtwide. Two years ahead of most peers.
Iowa has no state AI guidance. Iowa City didn't wait. The school board required AI curriculum districtwide and added quarter-long AI electives for 7th and 8th graders.
All students grades 7–12 are required to take an AI ethics course. The most mandatory student AI education program identified in this study.
Built a secure platform where students access ChatGPT, Gemini, and Claude trained on Washington State guidelines, with student identifiers removed. 2025 AWS Champions Award.
Students built an anti-bullying AI counseling tool and a school-wide homework assistant. The most student-driven AI district in the CRPE database.
AI assistant Sofia answers parent questions in 100+ languages. Board officially declared SAUSD an AI Forward District in April 2024.
Launched the "Ed" AI chatbot in spring 2024. The vendor company collapsed in June 2024. A whistleblower reported student data was misused. LAUSD serves 520,000 students, making this the largest U.S. district to deploy and then withdraw an AI chatbot.
NC ranks #3 nationally (7.90). CMS still bans ChatGPT on teacher devices and plans to lock it on take-home devices. They also deploy Evolv AI metal detectors and Gaggle surveillance.
Additional Districts Referenced in The Effort Crisis
The following districts were identified through independent research for The Effort Crisis and are not part of the CRPE Early Adopter Database used for this study's primary district analysis.
Mayor Michelle Wu announced BPS would become the first major city school district to require AI literacy for graduation. Backed by a million-dollar seed grant, the program trains an AI Ambassador at each high school and integrates AI literacy across subjects with a focus on ethics and critical thinking. A bold vision, though proposed alongside cuts to 265 classroom teachers.
One of the most practical AI frameworks in the country. Teachers label every assignment with a color: red means no AI, yellow means AI for specific purposes only, green means AI is encouraged. The first public high school to partner with OpenAI. Teachers know exactly what to expect. So do students.
Created a detailed AI framework covering professional development, student guidelines, and responsible use standards, with meaningful parent and teacher engagement built into the process from the start.
All 50 states were analyzed. 36 states and Puerto Rico had published formal AI education guidance and were scored on a 7-criteria weighted rubric (1–10). 14 states had published nothing and received a failing grade. All data sourced from publicly available documents compiled by aiforeducation.io, TeachAI, and individual state DOE websites. District data from CRPE's 2025–26 Early Adopter Database.
Does guidance encourage students to use and learn with AI, or focus on restriction?
Does guidance develop student judgment about when and how to use AI?
Does guidance address what students should still do themselves?
Does guidance treat AI literacy as a curriculum goal with standards?
Does guidance include specific, mandatory professional development plans?
Does guidance address how assessment changes in the presence of AI?
Does guidance address digital divides and equitable access to AI tools?
14 states have not published any formal AI education guidance document from their state department of education. They were not scored because there was nothing to score. The absence itself is a finding, but not all absence looks the same.
Legislation Passed, Guidance Pending — Idaho, Iowa, Texas
These states passed laws requiring AI education policy. Idaho's SB 1227 mandates guidance by July 2026. Iowa's bill sets a 2028 district deadline. Texas passed the Responsible AI Governance Act with an advisory council. All three will be scored when formal guidance is published.
Emerging Activity — Arkansas, Florida, Illinois, Kansas, Maryland, Nebraska, New Hampshire, New York, South Carolina, South Dakota
These states have real activity underway — task forces, city-level guidance, university frameworks, or coalition documents — but none have published a formal state DOE document. New York's NYC guidance covers the country's largest district but not the state. Maryland's MSDE is actively developing guidance. Florida's University of Florida-led task force has published resources, but official DOE adoption is still pending.
Safety Legislation Only — Pennsylvania
Pennsylvania passed Act 125 criminalizing AI-generated deepfakes and chatbot safety legislation protecting minors. But there is no guidance for how students should use AI as a learning tool. The state has answered the question of how to protect students from AI. It has not answered the question of how to prepare them for it.
Full dataset, rubric, and state-by-state breakdowns available upon request. Request the data →
Methodology
Transparency is not a footnote. Everything below is public on purpose — including the limitations.
Framework Citation — What We Adapted
The 7-criterion scoring rubric is adapted from three established frameworks:
Each of our 7 criteria maps to specific TeachAI principles and UNESCO competency domains. The full crosswalk is published in the Methodology sheet of our downloadable dataset.
Scoring Rubric — What the Numbers Mean
Each of the 7 criteria is scored 1–10. The table below defines what low scores and high scores look like in each dimension so any reader can independently evaluate or re-score a state's guidance document. The full scoring rubric with all anchors is in the downloadable dataset (Scoring Rubric tab).
| Criterion (Weight) | Score 1–3: Failing | Score 4–6: Developing | Score 8–10: Leading |
|---|---|---|---|
| Teach vs. Ban (20%) |
Document leads with prohibitions, consequences, and detection. AI is framed primarily as a cheating tool. | Balances caution with some guidance on use. Acknowledges AI as a learning tool but stops short of active integration. | Explicitly teaches students to use AI as a thinking partner. Provides differentiated guidance by subject, grade level, and context. |
| Critical Thinking (15%) |
No mention of evaluating AI outputs or questioning AI reasoning. Students are assumed to be passive recipients. | Mentions AI accuracy or bias in general terms. Students advised to "check AI output" without specific guidance. | Teaches systematic AI output evaluation. Covers bias detection, hallucination, source verification. Includes classroom implementation examples. |
| Student Effort / Cognitive Dev. (15%) |
AI use is unrestricted or limited only by academic integrity rules. No guidance on what students should still do themselves. | Mentions preserving student work but offers no framework. Relies on individual teacher discretion. | Explicitly addresses cognitive offloading. Names tasks AI should not replace (drafting first ideas, working through confusion). Vermont is the only state to name it explicitly. |
| AI Literacy (15%) |
No instruction on how AI works. Students learn to use AI tools without any understanding of their mechanics or limitations. | Mentions AI literacy as a goal. May reference standards (AI4K12) without integrating them into curriculum guidance. | Structured AI literacy scope and sequence across grade bands. Students understand how models are trained, what they can and cannot do, and how to use them critically. |
| Teacher Training (15%) |
No mention of professional development. Teachers expected to navigate AI policy without preparation. | References PD as a goal or recommends external resources. No district-level commitment or rollout plan. | Mandates sustained, multi-session PD (CRPE definition). Includes implementation timeline, coaching support, and accountability for teacher readiness. |
| Assessment Adaptation (10%) |
Assessments unchanged. AI policy is layered on top of existing test-and-grade infrastructure with no redesign guidance. | Recommends teachers "rethink" assessments in general terms. No examples or frameworks provided. | Provides concrete models for AI-resistant assessments: oral defense, iterative drafts with documentation, process portfolios, in-class demonstrations. |
| Equity (10%) |
No mention of access gaps, rural/urban divides, or how AI guidance reaches under-resourced districts. | Acknowledges equity as a concern. No specific provisions or funding mechanisms to close the access gap. | Names specific populations, provides implementation tiers by resource level, and includes data privacy protections for low-income and rural students. Alaska is the strongest example. |
How States Are Classified
Every U.S. state and territory falls into one of four categories in this study. Classification is based solely on publicly available documentation from state departments of education.
Scored (37 jurisdictions)
A formal AI education guidance document was published by the state department of education. Document was substantive enough to apply all 7 rubric criteria. Scored on a 10-point scale.
Legislation Passed, Guidance Pending (3 states)
State has enacted legislation requiring AI education policy, but no formal guidance document has been published yet. Idaho, Iowa, and Texas. Will be scored when published.
Why New Jersey is scored and Idaho is not
New Jersey's Department of Education published "AI in Education: Guidance and Considerations" in March 2024. A published guidance document, even a weak one, can be scored against the 7-criterion rubric. Idaho enacted SB 1227 in 2026 requiring a statewide AI framework by July 2026, but no guidance document has been published yet. Published guidance is the threshold for being scored. Legislation without a published document is tracked separately under "Legislation Passed, Guidance Pending."
Emerging Activity (10 states)
Real policy activity is underway — task forces, city-level guidance, university frameworks, or coalition documents — but no formal state DOE guidance document has been published. Not yet scorable.
Safety Legislation Only (1 state)
State has passed AI safety legislation (deepfakes, minors protection) but published no guidance on how students should use AI as a learning tool. Pennsylvania. Protection without preparation.
Rubric Ownership & Limitations — What We Own
Two-rater scoring. Both Kevin J. Roberts, M.A. (author of The Effort Crisis, academic coach) and Henry Dan, B.S. in Data Science with a machine learning emphasis (Post University, former Cambodian Math Olympiad team member and coach, co-founder of The AI Edge), independently scored all 37 jurisdictions against the pre-specified rubric. Discrepancies were reconciled by consulting the source document together. As a small-team study, we flag the rater panel size as a known limitation.
Tiebreaker rule. When weighted scores tie, the higher Student Effort / Cognitive Development sub-score wins. In the current rankings, Massachusetts leads at 9.05 and Vermont is second at 8.80. Vermont remains the only jurisdiction that explicitly names cognitive offloading in its guidance.
Reweighting sensitivity. The rubric does not structurally advantage Kevin's thesis. Dropping Student Effort (the criterion tied to The Effort Crisis) to zero weight does not move Vermont or Massachusetts out of the top 5. Equal-weighting all 7 criteria produces substantively identical top/bottom rankings. A reader who disagrees with our weights is invited to re-score with their own weights in the downloadable dataset.
Funding and disclosure. This study is self-funded. No external funding, sponsorship, grants, or organizational ties. The only financial interest connected to the findings is the sale of Kevin J. Roberts' book The Effort Crisis (Amazon, April 2026) and enrollment in The AI Edge summer program.
Mitigation and invitation.
Global Context
The United States leads the world in AI innovation. It does not lead the world in AI education. That gap is growing.
200M+
Chinese students now receive mandatory AI education
China's Ministry of Education issued mandatory AI curriculum standards in March 2025. Starting September 1, 2025, AI instruction became required in all primary and secondary schools — 200+ million students, ages 6–18, minimum 8 class hours per year. On April 8, 2026, China escalated further: the "AI+ Education" Action Plan, issued jointly by the Ministry of Education and four other government agencies, mandates AI integration from kindergarten through lifelong education and adds AI literacy to the national teacher qualification exam. Target: comprehensive AI education infrastructure for all citizens by 2030. Sources: People's Daily (April 2026); WebProNews.
27
EU countries now require AI literacy by law
Article 4 of the EU AI Act, effective February 2025, requires any organization deploying AI systems to ensure sufficient AI literacy of staff. This is a binding legal requirement across all 27 EU member states. Schools must demonstrate compliance. The U.S. has no federal equivalent.
Age 9
Singapore starts AI curriculum in primary school
Singapore's Ministry of Education integrates AI literacy beginning at Primary 4 (age 9-10) as part of their EdTech Masterplan 2030, with AI tools held back entirely at lower primary to protect foundational development. Their published AI-in-Education Ethics Framework is built on four principles: Fairness, Accountability, Transparency, and Safety. The U.S. has no national AI curriculum at any grade level.
All citizens
South Korea: Ambitious goal, stalled — then restarted
South Korea set an aggressive target: AI integrated across all K-12 schools by 2025. The plan stalled when Parliament reclassified AI textbooks as supplementary materials, leaving adoption to individual school principals. The new government, as of 2026, is pushing again — mandatory AI courses at national flagship universities, expanded K-12 AI programs, and significant infrastructure investment. The direction is clear even if the timeline slipped. The U.S. has neither a national goal nor a national timeline.
The countries leading in AI education are not the ones with the most AI companies. They are the ones with the most deliberate, coordinated education strategy.
The OECD's Digital Education Outlook 2026 found that 36% of lower-secondary teachers across OECD countries used AI in their work, ranging from under 20% in France and Japan to 75% in Singapore and the UAE. The U.S. does not even have a reliable number because there is no national system tracking it.
Read the Full Global Comparison →Kevin J. Roberts, M.A., is the author of The Effort Crisis: What Happens When We Hand Students a Shortcut Before They Build the Skill (April 2026), the book that identified the cognitive offloading problem this study measures. He is an academic coach, AI literacy educator, and co-founder of The AI Edge, a summer intensive that teaches students in grades 8-12 to use AI as a thinking tool rather than a thinking replacement.
Roberts has spent two decades studying the intersection of technology, cognition, and student behavior. His graduate research focused on the neuroscience of screen-based behavior, including the cerebral impact of excessive screen use on developing brains. He has presented Grand Rounds at Children's Hospital of Michigan (twice), spoken at 50+ conferences across the UK and Europe for organizations including CHADD, ADDA, ADHD Europe, and ADDISS, and appeared on ABC's 20/20, BBC Radio, and NPR affiliates. He is the author of six books including Cyber Junkie, Get Off That Game Now, Movers, Dreamers, and Risk-Takers, and Schindler's Gift.
Henry Dan, B.S. (Data Science), is co-founder of The AI Edge and co-rater for this study. A former member of the Cambodian Math Olympiad Team and math coach, Dan brings a machine learning background to the scoring methodology. He is fluent in Khmer, English, and Mandarin.
What Makes This Study Different
This is the first study to score state AI education guidance on a weighted rubric, not just catalog it. The 7-criteria rubric is adapted from established frameworks (see Methodology below) and applied to all 37 U.S. jurisdictions with published guidance.
The cognitive offloading criterion — whether states address what students should still do themselves before turning to AI — is the element most closely tied to Roberts' research for The Effort Crisis. Only Vermont's guidance (January 2026) names it explicitly.
The surveillance paradox finding — districts deploying AI to monitor students while restricting AI in classrooms — emerges from cross-referencing state-level policy with verifiable aggregate deployment data (Gaggle, GoGuardian, Securly).
Press inquiries, interview requests, dataset access: kevin@kevinjroberts.net · 248-867-3591
If this research was useful, please share it.
The more people who see these rankings, the harder it becomes for states to ignore the question.
Press inquiries, interviews, dataset access: kevin@kevinjroberts.net · 248-867-3591