Last year, U.S. schools spent about $2.5 billion on AI tools. By 2033, that number is expected to top $15 billion. Somewhere in a district office near you, someone is deciding which of those tools gets a contract — and according to a new Stateline investigation, they’re often doing it with less information than the company selling them the product.
“There’s always an asymmetry between what the providers know and what the districts know,” one official told Stateline. Ed-tech purchasing has always had this problem. What’s different with AI is speed: the tools are changing faster than any district’s ability to evaluate them, and unlike a textbook, you often can’t just flip through an AI tool and see what it actually does.
This isn’t a one-off complaint. Instructure’s 2026 Evidence Report — a survey of classroom technology broadly, not just AI — found that most consumer ed-tech in schools today lacks any verified proof that it actually helps students learn. Tools get adopted because they’re popular, because a rep gave a good demo, or because “everyone else is using it.” Whether they work is a separate question nobody’s actually answering.
A case study in what “no way to verify” looks like
You don’t have to look far for a concrete example. Alpha School — the AI-driven “two-hour learning” model now expanding toward 50 campuses nationwide — has built its entire pitch around a headline claim: students learn twice as fast.
That number comes entirely from Alpha’s own analysis of its students’ NWEA MAP Growth scores. Alpha submits the data. NWEA records it. Alpha publicizes it. NWEA itself hasn’t audited it, no independent researcher has been given access to check it, and there’s no control group to compare against. According to Tech Times, Alpha’s own team promised to hand over raw data files for an independent audit — as of this writing, three months after that promise, the files still haven’t arrived.
To be clear: this isn’t an accusation that the number is false. It’s the opposite problem — nobody outside Alpha is in a position to say whether it’s true. And that’s exactly the gap Stateline and Instructure are describing at the scale of an entire industry, playing out in one specific, checkable claim.
What “verified” would actually require
If a district can’t tell whether an AI tool works, the honest answer isn’t “trust the vendor” or “wait for a study that may never come.” It’s building the evaluation into how the tool gets used in the first place — so the evidence exists whether or not anyone later asks for it.
That’s the idea behind COMPASS, Auxesis’s AI operating system for facilitated learning. Before a session, a Facilitation Brief tells the human teacher or tutor what to focus on. During the session, a person — not the AI — does the actual teaching. After, a five-minute note captures what happened, in plain language, tied to a real student and a real concept. None of that data lives in a black box. It’s not a marketing claim aggregated at the end of the year; it’s an ongoing, readable record of what was taught, to whom, and whether it landed — built for a parent, a teacher, or a district to actually look at.
That’s a smaller promise than “twice the learning.” It’s also one you can check.
The real question for anyone buying AI right now
Before any district signs a contract, before any parent pays for a subscription, the Stateline and Instructure findings point to the same practical question: not “does this claim sound impressive,” but “can I see the evidence myself, or am I taking the vendor’s word for it?”
Auxesis exists because we think that question deserves a real answer — not a promise that’s still three months overdue.
If your district or your kid’s school is using an AI tool right now — do you actually know what evidence backs it up, or are you taking it on faith? We’re curious what people have found when they’ve asked.
Sources: Stateline, “Schools spend billions on AI, but struggle to figure out what’s worth buying,” Aug 12, 2026 · Instructure 2026 Evidence Report · Tech Times, “AI Private Schools Promise Twice the Learning: Experts Cannot Verify That Claim,” Jul 9, 2026
Leave a Reply