Schools Are Buying AI Tutors. Most Skip the Question That Actually Matters.
Last updated:
If your child is behind in reading and a school or district announces it is adopting an AI tutoring program, the announcement arrives wrapped in the word personalized. Personalized learning. Personalized support. Personalized instruction. The word is doing a lot of work, and it is worth pausing to ask exactly what it is personalizing to.
North Carolina’s state legislature just gave the education community a rare, clarifying look at how these decisions get made. A Senate proposal to spend $10 million on Khan Academy’s Khanmigo AI tutoring platform, awarded without competitive bidding, drew immediate scrutiny from educators and AI researchers who questioned whether the evidence base warranted the price. The final conference budget, released July 1, 2026, landed at $500,000, a 95 percent reduction. NC Newsline described the value of the tool as “debatable.”
The skepticism is warranted. But the lesson is not about this particular contract. The lesson is about the question that is almost never asked before a school buys a product like this one.
TL;DR
North Carolina’s Senate proposed $10 million for Khanmigo AI tutoring in a no-bid contract; the final conference budget allocated $500,000, after educators and researchers questioned whether the evidence justified the cost.
Khanmigo has one published randomized controlled trial (2025, algebra) showing directional benefit. There is no peer-reviewed evidence base for its use with students who have reading, attention, or processing differences.
IDA 2025 confirmed reading difficulties involve phonological processing, auditory processing, working memory, and processing speed simultaneously. Khanmigo, a Socratic curriculum chatbot, does not target any of these systems.
Brain imaging studies (Shaywitz, Yale; Temple, Stanford) show that the right intervention physically rewires reading pathways. “Right” means matched to the specific processing gap, not AI-powered in a general sense.
The question every parent and every legislature should ask before adopting an AI learning tool: which specific cognitive processing system does this target, and what is the independent evidence that it closes the gap for students with this profile?
North Carolina lawmakers cut AI tutor funding by 95 percent before the contract paid out. The skepticism was warranted. The question most coverage missed is the one that matters most to parents of struggling learners.
Common questions
Is Khanmigo good for kids who struggle with reading?
Khanmigo is designed as a Socratic curriculum tutor, and there is a 2025 randomized controlled trial showing directional improvement in algebra scores. There is no published evidence that it addresses the processing systems driving most reading difficulty: phonological processing, auditory discrimination, working memory, and processing speed. For a child whose reading struggles trace to one or more of those systems, an AI curriculum chatbot is not targeting the actual bottleneck. A screener is a starting point for identifying which systems need support. A screener is not a diagnosis; if your child needs formal accommodations such as an IEP or 504 plan, or you suspect a vision, hearing, or medical cause, a professional evaluation is the route to those supports.
Why did North Carolina cut the Khanmigo funding so sharply?
The original $10 million Senate proposal was awarded without competitive bidding, which drew procedural objections from educators and public officials. NC Newsline reported that AI experts questioned whether the tool’s value justified the expenditure, describing it as debatable. The budget conference committee reduced the allocation to $500,000 nonrecurring for licenses and an AI course over two years. The contract must still pass both chambers and be signed by the governor. The cut reflects both procurement concerns and skepticism about whether the evidence base warranted the original investment.
How do I know if an AI learning tool is right for my child?
Start by knowing which processing system is the actual constraint. Reading difficulties involve combinations of phonological processing, auditory processing, working memory, and processing speed, as confirmed by the IDA’s 2025 definition. Once you know which system is the bottleneck for your child, you can ask whether any given AI tool targets that system specifically and what independent research backs that claim. An AI tool that skips the cognitive profile step is offering personalized-sounding delivery to an audience it does not know.
My child is behind in school. Where do I start?
Start with the processing gap. A multi-system analysis identifies which cognitive systems are falling short, whether phonological processing, auditory processing, working memory, or others, so you know what kind of targeted practice your child needs and which tools are actually worth trying. Brain imaging research by Shaywitz (Yale) and Temple (Stanford) shows that children with reading difficulties who receive the right intervention develop the same reading pathways as typical readers. “Right” means matched to the specific system. Knowing the gap makes every other decision clearer.
How a $10 million AI tutor contract became a $500,000 question
The sequence in North Carolina is instructive. Senator Mike Lee, a Republican from New Hanover County, introduced a proposal in the state Senate to direct $10 million to Khan Academy for Khanmigo licenses, awarded through a no-bid contract. The proposal was revised to $5 million split between Khanmigo and another AI education platform, MagicSchool, before a budget conference committee settled on $500,000 nonrecurring, directed to the state Department of Public Instruction for Khanmigo licenses and an AI course over two years.
Alongside the Khanmigo allocation, the budget requires school districts and charter schools to develop AI policies, tasks the NC State Friday Institute with educator professional development, and mandates teacher AI training by June 2028. Governor Josh Stein still has to sign the final budget. The Khanmigo contract, if it survives, will move forward with a fraction of the original funding and none of the competitive vetting that a purchase of this type would normally require.
Educators who raised concerns about the no-bid structure were asking the right procedural question. But there is a substantive question underneath it that matters even more for the parents of struggling learners: what does Khanmigo actually do, and does it address what causes a child to fall behind?
Author Quote"
“Reading isn’t one skill wearing a trench coat. It’s several systems — language, attention, memory, processing speed — working together. Fix only the one everyone’s heard of, and the others stay invisible until they stall a child out later.”— Laura Lurns, LearningSuccess.ai
"
What most coverage misses
Reporting on the North Carolina Khanmigo story focuses on the no-bid procurement process and the 95 percent budget cut. Both are legitimate angles. What the coverage largely omits is the underlying cognitive question: even a well-procured, competitively bid AI tutoring contract would not address the processing systems that drive most reading difficulty. The International Dyslexia Association’s 2025 definition identifies phonological processing, auditory discrimination, working memory, and processing speed as the systems involved simultaneously in reading difficulties. Khanmigo is a Socratic curriculum chatbot. The procurement question and the evidence question are separate; both matter, and most coverage treats the first as though answering it settles the second.
What Khanmigo is and what the evidence actually shows
Khanmigo is a Socratic AI tutor built on top of Khan Academy’s curriculum. It guides students through math and other subjects using questions rather than direct answers, the same conversational approach a good human tutor uses when walking a student through a problem. A 2025 randomized controlled trial found promising improvements in algebra scores for students who used it. That is one study, with a directional finding, in a single subject. It is not an evidence base for adoption across a state’s struggling learner population.
More importantly, the study was not conducted with students who have reading differences, attention differences, or the kinds of processing challenges that are at the root of most academic difficulty. Khanmigo is a curriculum delivery tool. It works by having a conversation. It does not assess or train phonological processing. It does not build auditory discrimination, working memory capacity, or processing speed, the systems that the International Dyslexia Association’s 2025 definition confirmed are involved simultaneously in reading difficulty.
Reading doesn’t fail in one system. “Reading isn’t one skill wearing a trench coat,” as the Learning Success framework puts it. “It’s several systems — language, attention, memory, processing speed — working together. Fix only the one everyone’s heard of, and the others stay invisible until they stall a child out later.” Phonics is necessary. It was never the whole answer, and a conversational AI built for curriculum delivery is not phonics instruction. It is something different again.
North Carolina’s budget negotiators cut the Khanmigo allocation by 95 percent because they could not find an evidence base strong enough to justify $10 million. That instinct is sound. The question worth borrowing from their deliberation is not “how much should we spend?” It is “which specific processing systems does this tool actually target?”
Key Takeaways:
1
The NC cut reflects real evidence gaps: North Carolina’s legislature reduced Khanmigo funding from $10 million to $500,000 after educators and AI researchers questioned the evidence base. One 2025 algebra RCT is not an evidence base for state-wide adoption across diverse learner populations.
2
Khanmigo is curriculum delivery, not cognitive training: Khanmigo is a Socratic chatbot for curriculum content. It does not assess or train phonological processing, auditory discrimination, working memory, or processing speed, the systems IDA 2025 confirmed are involved simultaneously in most reading difficulty.
3
The right question comes before any purchase decision: Before any AI tutoring tool is adopted for struggling learners, the question “which specific processing gap does this target?” should be answered. A multi-system cognitive profile makes that question answerable for your child.
The question that should come before any AI tool purchase
When a school announces an AI learning platform for students who are behind, three questions belong in every parent conversation with administrators. First: what specific cognitive processing gap does this tool target? Second: is there independent, peer-reviewed research showing it closes that gap for students with this profile? Third: what would the same resources fund if directed toward approaches with a stronger evidence base?
The IDA’s 2025 definition made explicit what cognitive scientists had established over decades of research: reading draws on multiple systems simultaneously, including phonological processing, auditory processing, working memory, and processing speed. When a child struggles with reading, the bottleneck is somewhere in that system. Identifying which system is the actual constraint is the work that has to happen before any intervention is chosen. It is the work that no amount of AI curriculum delivery substitutes for.
Brain imaging studies at Yale (Shaywitz) and Stanford (Temple) confirmed that children with reading difficulties who receive the right targeted intervention develop the same reading pathways as typical readers. “Right” means matched to the specific processing gap. It does not mean engaging, or modern, or AI-powered in a general sense. An AI tutor that skips cognitive profiling is a Socratic conversation with a child who needed something more specific.
A multi-system processing analysis identifies which of those cognitive systems is falling short for your child, in plain language, before you have to decide which program or tool to trust. A screener is a starting point, not a diagnosis. If your child needs formal accommodations such as an IEP or 504 plan, or you suspect a vision, hearing, or medical cause, a professional evaluation is the route to those supports. But knowing the specific gap changes every question you ask, including the ones a state legislature should ask before signing a $10 million contract.
Author Quote"
“Phonics is necessary. It was never sufficient. Treat it like the whole answer and you’ll get a child who sounds out words and still can’t read.”— Laura Lurns, LearningSuccess.ai
"
The obstacle here is not bad technology. Khanmigo is a competent Socratic tutor for curriculum content. The obstacle is a framing error repeated at every level of the education system: that AI plus engagement equals learning, that personalized-sounding language proves the targeting is real. North Carolina’s budget negotiators stumbled toward the right instinct when they cut the contract 95 percent, even if the stated reason was procedural rather than cognitive. The right instinct, applied to your child, is the same one: ask what system this addresses before you trust the label. Learning Success gives you the map. A multi-system analysis of your child’s processing profile tells you which cognitive systems need support and what targeted practice for those systems looks like, in plain language, before you have to decide which tool or program to trust. Find out what’s actually in the way.
Is Your Child Struggling in School?
Get Your FREE Personalized Learning Roadmap
Comprehensive assessment + instant access to research-backed strategies
International Dyslexia Association — 2025 definition of dyslexia (multi-system causation; phonological processing, auditory processing, working memory, processing speed)
Khanmigo algebra RCT (2025) — directional improvement in algebra scores, single published study
Shaywitz et al. (Yale) & Temple et al. (Stanford) — fMRI studies of reading-intervention brain change
Mark Seidenberg, Language at the Speed of Sight — multi-system reading science
