AI Tools School Districts Approved in 2026: What Districts Actually Paid
NationGraph / Blog / K-12 EdTech
Seven named AI vendors, eight published prices, and the 47 districts that wrote a policy without buying anything.
Published Sep 15, 2026
Quick Answer
School boards approved seven named AI vendors between March 2025 and September 2026, and the eight published prices run from $712.39 to $37,500. The largest is Crown Point Community School Corporation in Indiana, which voted $37,500 of Digital Learning Grant money for the MagicSchool Enterprise platform on June 22, 2026. MagicSchool appears in more district records than any other tool, in 15 districts, 8 of which took a board vote or cut a purchase order. Most districts have not bought anything: of 76 entities with a documented AI decision, 30 adopted a policy and 16 approved a tool.
According to NationGraph's analysis of school board minutes and meeting video, agenda packets, and purchase orders, at least 76 public entities across 27 states took a documented AI action in the last 18 months. Almost none of it reaches a bid feed. A district AI purchase is usually a consent-agenda line under $40,000, below the threshold that triggers a competitive solicitation, and the policy decisions that precede it produce no procurement document at all. Anyone selling into K-12 procurement is looking at a category that moves through minutes.
Which AI tools have school boards actually approved?
Seven: MagicSchool, Brisk Teaching, Khanmigo, SchoolAI, OpenAI's ChatGPT Edu, Curipod, and Diffit. Twenty-nine of the 76 entities named a vendor at all. The other 47 took an AI action without naming a product.
MagicSchool leads on raw count, but the stage column matters more. Of its 15 districts, 8 reached a board approval or purchase order, five are still piloting, one is at proposal, and one, Rio Elementary CA, saw two trustees pull the MagicSchool contract off the consent agenda on June 24, 2026. That is the only visible case of a district stopping a purchase already on the agenda.
Brisk Teaching converts at a higher rate, 5 of its 6 districts, and holds the longest contract in the corpus, a multi-year agreement with Closter Public School District in New Jersey running through April 2029. Khanmigo is the only consistently student-facing tool: Washington Elementary and Tucson Unified in Arizona both approved it for student use, and Flagler County in Florida is the one case in the census that shows a pilot converting to a rollout.
Two districts bought two tools at once: Keyport NJ approved SchoolAI Classroom Pro and MagicSchool as separate motions on the same August 2025 night, and Haverford Township PA did the same with SchoolAI and Brisk. These are not evaluations that ended in a single winner. Google Gemini for Education and Microsoft Copilot appear in no retrieved record, and the reason is below.
What does an AI tool cost a school district?
Between $712.39 and $37,500 per agreement, based on the eight records carrying a verbatim dollar figure. Eight prices out of 76 records is the most important thing to know about this table.
The spread is district size, not product tier. Mill River Unified VT approved $712.39 for MagicSchool Enterprise and Crown Point IN approved $37,500 for the same product, a factor of 53 on one platform. Neither record states how many teachers the license covers, and none in the census does.
Jefferson Academy in Utah is the row worth studying. It shows $1,673.25 for software and a separately approved $500 for "professional employee training and development" on the same date, so training ran about 30 percent on top and was priced as its own line. Pennsylvania Cyber Charter School's $3,999 Curipod approval bundles licensing and professional development into one figure, so the same cost exists there but is not separable.
One purchase in eight names its funding source, Crown Point's Digital Learning Grant, and no record cites ESSER, Title IV-A or E-Rate. Grant awards remain a leading indicator of an unsolicited purchase, but in this corpus the money question is close to unanswerable.
What do these AI tools cost at list price?
Five of the seven publish no district price at all. Only Khan Academy posts a per-unit institutional number, $10 per student for districts already using the SAT Suite in 2026-27. Everything else at the district tier is quote-only.
The two vendors that publish a unit price use different units. MagicSchool lists Plus at $12.99 a month or $99.96 a year and prices Enterprise by quote. Brisk publishes Educator Pro at $99.99 a year and states that district pricing is per student enrollment. A per-teacher tool and a per-student tool cannot be compared on sticker until you know the district's staff-to-student ratio.
One piece of arithmetic, with a warning. Crown Point's $37,500 is 375 times MagicSchool's published $99.96 annual individual rate, and Fulton's $11,250 is 112.5 times it. Those are ratios, not seat counts. Enterprise is volume-discounted and adds dashboards, SSO and a custom privacy agreement the individual plan lacks, so the real teacher count behind each figure is higher. No record states it.
The most consequential list-price fact is not a number. On November 19, 2025, OpenAI launched ChatGPT for Teachers, free for verified US K-12 educators, and has since extended that window to June 2028. Two districts here bought paid ChatGPT access before that date: Hurst-Euless-Bedford ISD TX approved a ChatGPT Edu purchase plus an Enterprise API item on June 9, 2025, and Lower Merion PA approved $12,960 for an Edu license through the reseller Edvative Learning on September 15, 2025. Five months and two months later, the teacher-facing version was free, and no district in the census has approved paid ChatGPT access since.
Are districts buying AI tools or writing AI policies?
Writing policies, by close to two to one. Thirty of the 76 entities adopted an AI policy, 16 approved a tool purchase, and another 12 have a policy in draft.
Governance is moving faster than procurement, and the sequencing is not what a vendor would expect. Saint Peter Public Schools in Minnesota described its order in a November 2025 board meeting: "Phase one includes uh teachers, a group of teachers who will be researching ways that artificial intelligence [can be used] ... move into ways to implement with students and essentially get to policy at some point." (clip 00:24:40) Pilot first, students second, policy last. Oyster River in New Hampshire is running the same sequence, reviewing its policy alongside feedback from a fourth-grade MagicSchool pilot.
Part of this split is real and part is a disclosure artifact. A policy adoption is always minuted, because adopting a policy is the board action. A software purchase under a district's competitive threshold often appears as a consent-agenda line with no vendor name, or not at all. Both effects push the same way, and this table reflects our corpus rather than the market.
A district that has adopted an AI policy has already had the hard internal conversation: a named owner, a position on student versus staff use, usually an approved-tools concept in the language. That is a qualified account, not an early one, and timing against the budget cycle matters more once a policy exists.
Where do school district AI policies come from?
Most write a standalone new policy: 14 of the 34 entities where the route is visible. The second most common route is not a district at all. Six adopted model language from a policy service, named in the minutes as NEOLA, the California School Boards Association, New Jersey POLICY ALERT No. 234, and the Missouri School Boards' Association.
That matters more than it looks. When six districts adopt near-identical language from the same subscription service, the effective author of their AI policy is an association, not a superintendent. Two New Jersey districts here, Pitman Boro and Toms River Regional, adopted "Policy 2365 Acceptable Use of Generative Artificial Intelligence" under the same number, which is the signature of a shared model. A product compatible with the NEOLA or CSBA template is compatible with a lot of districts at once.
Five districts wrote no AI policy at all, folding AI into an existing integrity or conduct policy, and three more extended an acceptable use policy. Only two wrote theirs in-house: Loudoun County VA, where "Policy 5430 ... was written by Department of Teaching and Learning staff," and Springfield MA, through a task force.
What do school district AI policies actually restrict?
Specific uses, not the technology. No district in this census banned generative AI. What exists is narrower and more useful.
Cabrillo Point Academy District in California adopted an AI policy in May 2025 containing an explicit "PROHIBITION ON USE OF AI NOTETAKERS," with monitoring, enforcement and a signed staff acknowledgment attached. That is a ban on a product category, aimed at staff rather than students, and it is the most specific prohibition in the corpus. Sunset Ridge School District No. 29 in Illinois wrote AI eyewear into its recording-devices policy, covering "recording by students using any device including artificial intelligence, AI glasses, aka metagasses." (clip 02:36:20)
The other two are assignment-level. Buckeye Union High School District in Arizona placed AI inside its assessment policy as a prohibited use, with a procedure for when "an employee suspects that a student has used technology, including artificial intelligence, in violation of this policy." Pitman Boro in New Jersey restricted students using AI to complete assignments.
Academic integrity is the most frequently stated reason, named at six entities including Buckeye Union, Pitman Boro and Dallas School District OR. Student data protection is second. Southeast Fountain School Corporation IN put the tradeoff in one line: "Our goal is to protect student data while preparing students in a future where AI is a common tool." (clip 00:15:49) Dallas is the only entity in the census to cite a federal statute by name, 20 U.S.C. § 1232h. Expect the integrity conversation before the privacy one.
Which school districts are evaluating AI tools right now?
Twenty-five of the 76 entities sit at pilot, proposed, studying, discussed or grant-applied stage. These are pre-RFP positions: a decision is forming and no solicitation exists. The 14 with the most substance are below.
Sort on the stage column. A district running a named pilot has already chosen a product and is building the internal case, which is late. A district running a survey has chosen nothing. La Canada Unified CA is polling parents so the results "can help inform the board's discussion and shape its policy," and San Mateo Union HSD ran "a mini focus group centered on artificial intelligence." Those are the cheapest stages at which to win a relationship, and they are invisible to bid boards by definition.
Two rows deserve attention. Newton Public Schools MA recorded grant writing for a Brisk pilot continuation and a "Venture capital testing partnership with Renaissance learning" in one January 2026 memo, a district acting as a test site rather than a buyer. Rio Elementary CA is the pulled-contract row: an item reaching the consent agenda and coming back off means a competitor is inside and has hit resistance.
Why don't district AI purchases show up in purchase order data?
Because the vendor names are ordinary words. Searching a purchase order corpus for "MAGIC SCHOOL" returns Scotch Magic Tape back-to-school promotions, the Apple Magic Keyboard and The Magic School Bus. Roughly 25 of 30 results for that query were false positives, removed by hand. Four real AI vendor POs survived.
This is structural. Any AI vendor whose name is built from common English words is unfindable in spend data. "Gemini" and "Copilot" are worse than "MagicSchool," and both usually sell inside a larger Google Workspace or Microsoft 365 agreement rather than as a separate line, so even a clean match would not isolate the AI component. Their absence here is a search artifact compounded by a bundling artifact, not a market finding.
The approach that works is the one for any pre-RFP category: read the minutes. Board agendas name the product in the item title, which is where purchase order data is weakest.
Methodology
Figures come from board minutes, meeting video transcripts, agenda packets, purchase orders and one budget document, each carrying a verbatim passage. Forty-one of the 76 records have no clickable source URL because the underlying minutes were indexed without one, so where a figure rests on an unlinked record the passage is quoted instead.
The census covers 76 entities in 27 states, dated between March 17, 2025 and September 9, 2026. Every search returned a capped or paged result, so 76 is a floor rather than a total.
Two things here are computed rather than stated: the ratio of an approved sum to a published list rate, labeled as a ratio, and the per-vendor entity counts, recomputed from the record file. Three records carry a dollar figure truncated at the source, at Paterson NJ, West Jefferson Hills PA and Rio Elementary CA, and those amounts are withheld. Price tables exclude any record without a verbatim figure.
Counts reflect disclosure patterns, not market share: Pennsylvania supplies 14 of 76 entities because its districts publish indexed text minutes, not because it buys more AI.
Related reading
- How to Sell to School Districts: A Complete Guide
- How Education Technology Companies Navigate K-12 Procurement
- Timing Is All You Need: K-12 District Budget Cycles for EdTech Sales
- How to Monitor School Board and City Council Spending
- 8 Key Data Sources for Pre-RFP Government Signals
- How to Track Government Contracts Before the RFP in 2026
Frequently Asked Questions
Which AI tools are approved for school districts?
Seven named vendors appear in this census: MagicSchool, Brisk Teaching, Khanmigo, SchoolAI, OpenAI's ChatGPT Edu, Curipod, and Diffit. MagicSchool leads with 15 districts, 8 of them at board approval or purchase order. Only 29 of 76 entities named a vendor at all.
How many school districts have an AI policy?
Thirty of the 76 entities here have adopted one and 12 more have a policy in draft, across 27 states, roughly twice the number that approved a tool purchase. The national figure is higher: this census is a floor drawn from indexed public records, not a complete count.
Have any school districts banned AI?
Not in this census: zero of 76 entities banned generative AI outright. Four restricted specific uses. Cabrillo Point Academy District CA prohibited AI notetakers, Sunset Ridge SD No. 29 IL wrote AI glasses into its recording policy, Buckeye Union HSD AZ made AI a prohibited use in assessments, and Pitman Boro NJ restricted AI for completing assignments.
Who writes school district AI policies?
Usually the district, but not always. Fourteen wrote a standalone new policy and two wrote theirs in-house. Six adopted model language from a policy service, named in the minutes as NEOLA, the California School Boards Association, New Jersey POLICY ALERT No. 234, or the Missouri School Boards' Association.
Are AI tools approved for teachers or for students?
Overwhelmingly for teachers. The exceptions are Khanmigo, approved for student use at Washington Elementary and Tucson Unified AZ and expanded to all math students at Flagler County FL, plus two MagicSchool records covering student use. San Juan Unified CA ran a 22-student pilot.
How do school districts pay for AI tools?
Only one priced record names its funding source: Crown Point Community School Corporation's $37,500 Digital Learning Grant. No record cites ESSER, Title IV-A or E-Rate for an AI purchase. Spring Hill, Tennessee noted that any future AI licensing costs would go through normal procurement and budgeting.
Find your first opportunity in minutes
