Schools have acquired AI products rapidly with little evidence about their effects. The pattern predates this technology and follows from how educational purchasing is structured.
The buyer is not the user
Purchasing decisions sit with administrators and district offices, while the product is used by teachers and students.
Vendors therefore sell to criteria administrators can assess: coverage of standards, reporting dashboards, integration with existing systems, and price.
Whether the tool improves learning is not assessable in a procurement meeting, so it is represented by whatever evidence the vendor supplies.
Vendor evidence is structurally weak
Studies accompanying these products are usually conducted by the vendor, in schools that volunteered, comparing users against non-users.
Schools that adopt a new tool differ from those that do not in leadership, funding and staff capacity, and those differences affect outcomes independently.
Reported results therefore measure the willingness to adopt at least as much as the product, and the design cannot separate them.
Independent evaluation runs slower than procurement
A properly controlled study takes a school year at minimum, followed by analysis and publication.
By the time results appear the product has changed version, and the vendor can reasonably say the evaluated version is not the one being sold.
This is a particular problem for tools built on external models, where the underlying capability changes without any action by the vendor at all.
Funding cycles reward spending now
Grants and one-off allocations frequently must be spent within a defined period or returned.
That creates pressure to purchase within a timeframe far shorter than any evaluation, and favours products that can be deployed immediately.
Multi-year subscriptions signed under this pressure then constrain budgets after the funding that justified them has ended.
What a school can actually check
The assessments within reach of an ordinary school are narrower than an efficacy study and considerably more useful than a vendor claim.
These include running the tool against material the school already has, checking what data leaves the building and on what terms, confirming whether the content is accurate for the specific curriculum in use, and asking what happens to student records if the contract ends.
Schools that ask those four questions reject a substantial share of what is offered to them, which suggests the constraint is less about capacity than about who in the building is asked to look.