Last updated October 27, 2026. The New York Times published its report on Alpha's results on October 23, 2026. I will update this page if Alpha responds or the figures change.
Nobody can yet say whether Alpha School is worth $40,000 or more a year, because most of the evidence is the school's own. The New York Times reported on October 23, 2026 that only 42 percent of its students grew twice as fast in both math and reading, which is Alpha's own promise. How your child works alone matters most, so ask the eight questions below before you pay.
The verdict
Is Alpha School worth it? Unproven. Its results come mostly from its own analysis of one kind of test, and the Times found that fewer than half of its students met its headline promise. That does not mean your child would not do well there. It means the brochure cannot tell you.
Does the answer change with age? Yes. The younger the child, the more I would want to see an adult doing the teaching. The older the child, the more the question becomes whether they can work alone for hours.
One thing to do tonight: Give your child ten minutes of homework and stay out of the room. How they start, what they do when they get stuck, and whether they come and find you tells you more than any school's average will.
I have not visited an Alpha campus and I have no inside information about it. Everything I say about the school comes from the sources linked below, and I say which kind each statement is: what a source found, what I make of it, or what I would do.
What is an AI school, and what does a day at Alpha look like?
"AI school" is a loose label. Alpha, the best known, says on its homepage, which I read on October 27, 2026, that academics are "completed in the mornings," with afternoons for life skills and interests. Its staff are called guides, and the page says "Our AI tutor gives students one-on-one personalized education."
The detail that matters most to a parent comes from Chad Aldeman's column in The 74, published January 8, 2026. He reports about two hours of core academics a day, one guide for every eight students, and adaptive programs such as i-Ready, DreamBox, Khan Academy, IXL and Zearn, with AI used to map which lessons a student still needs. The column quotes Alpha's Joe Liemandt as critical of chatbots. So the "AI" here is mostly adaptive practice software, not a chatbot answering questions. Aldeman also discloses that he is a policy adviser for NWEA, which makes the MAP tests that Alpha's results rest on, so I read his column as informed rather than neutral.
On scale and cost, Newsweek's September 14, 2026 map found Alpha campuses in 16 states, all charging tuition, many not yet open. The Times, as summarized below, put the network at roughly 50 campuses and about 1,500 students. The counts differ between reports, so ask any campus you are considering for its own current list.
My reading: for a parent, "AI school" mostly means a school where adaptive software does most of the direct teaching and an adult coaches. That tells you what to compare it with. It is not a human-free future. It is a particular way of arranging a day. Most schools use AI in a far more modest way, which I describe in How Is AI Actually Used in Schools?.
What did The New York Times find about Alpha's results?
The Times article is behind a paywall that blocked my tools, so I read two summaries that quote it: AI Weekly's report on October 24, 2026 and Benjamin Riley's response on October 25, 2026. Riley is a critic of Alpha, which is worth knowing when you read him. As those two quote the Times:
- Alpha's "2x" promise held for only some students. Riley quotes the Times: "only 42 percent progressed twice as fast in both math and reading."
- Weaker math students did worse. AI Weekly quotes the Times: "About a fifth of Alpha's private school students scored at or below the 50th percentile nationally in math last fall," and "only 37 percent of that group met the network's promised 2x growth by winter."
- "2x" has a specific meaning. AI Weekly reports that reviewers say Alpha divides a student's growth on the test by the growth expected for a typical student. If a student grows 14 points when 7 are expected, Alpha calls that 2x.
- Some families left. AI Weekly reports that the Times describes parents withdrawing children "after realizing that the model demands an extraordinary level of intrinsic motivation and self-discipline."
Two cautions on reading those numbers. The 42 percent and the 37 percent describe different groups of students, so they do not add up or average. And growth on a test is a measure of test growth, not of everything a school does. Riley makes the same point in his own words, calling test scores "rough proxies for real learning." Neither summary included a response from Alpha, and I did not find one in other coverage I read.
My reading: the figure that should matter most to you is the math group. An average hides the child who looks like yours. If your child is behind, or finds school hard to start, ask for the results of children who started where yours did.
What does Alpha itself claim, and has anyone independent checked it?
Alpha's homepage says "Alpha classes learn twice as much as their peers and rank in the top 1% nationwide." The page I read does not name the test or the comparison group. Aldeman's column reports the company's figure that students gain on NWEA's MAP Growth test at 2.6 times the rate of similarly scoring students. He says the figures come from Alpha's own materials and identifies no independent evaluator. He calls the rates "plausible" but advises "a healthy dose of skepticism," and notes that Alpha's data on lower-performing students rests on a small sample. Newsweek's own summary is that "some education experts caution that evidence for fully AI-centered schooling remains limited."
A regulator has also looked at a related proposal. In January 2025, Pennsylvania's Department of Education rejected an application for Unbound Academy, an online charter school tied to Alpha co-founder MacKenzie Price, writing that "There is no compelling evidence that Unbound Academic's proposed programs will enable students to meet academic standards," as Chalkbeat reported on January 29, 2025. That was an online charter application in one state, not Alpha's private campuses, so it does not tell you how those campuses perform. It does show what a public body asked for and did not get.
I could not find an independent evaluation of Alpha's results in my search on October 27, 2026. That is a statement about my search, not proof that none exists.
Does the research on AI tutors support a school run this way?
Partly, and not the part that is being sold. Two well-known studies show what the evidence can and cannot carry.
In a randomized trial published in Scientific Reports on June 3, 2025, 194 Harvard students in an introductory physics course learned one topic with a purpose-built AI tutor and another in an active-learning class. The tutor group did better on the follow-up test. The authors write that "we do not presume that structured AI tutoring will always outperform in-class active learning in all contexts," and the study covered one course at one university over one semester.
In a field experiment of nearly 1,000 high school math students in PNAS, students with a plain GPT-4 interface improved during practice (48 percent in grades) but did worse than students who never had it once access was removed (a 17 percent reduction). A version with teaching safeguards largely avoided that harm.
Neither study tested a school day built around software, for young children, over years. What they support is narrower: AI helps learning when a person has designed the learning around it, and the design matters more than the label. I cover both in more depth in AI Tutors for Kids, and the wider question of whether AI can stand in for a teacher in Will AI Replace My Child's Teacher?.
What should I ask before paying for an AI school?
Use these at any school that runs much of the day on software, Alpha or not. Print them and take them to the visit. For each, I give what a good answer sounds like.
- What are the results for children who started where mine did? A good answer names the test, the grade, the year and the number of children, and splits results by starting level. "Our students grow 2x" is a claim. A table by starting level is an answer.
- Who checked those results, and can I see the data? A good answer names an outside evaluator or offers the underlying figures. If the school mentions an accreditation, ask what the accreditor looked at. A logo and a test result are different things.
- How many children left in the last two years, and why? A good answer is a number and a plain reason. The Times reported families leaving because of the self-discipline the model demands, so this question is worth asking out loud.
- What happens here on the day my child will not work? A good answer names a person and a plan. A reward system is not a plan. This is the question I weight most (see below).
- Which adult knows my child, and how often will I hear from them? A good answer gives a name, a ratio and a rhythm, such as a weekly note. Ask what training that adult has and whether they teach or supervise.
- What does my child do on the software, and what happens to the data? A good answer lists the programs, says what is collected, who it is shared with, and how to have it deleted. Ask for the privacy policy in writing and read it.
- What fills the rest of the day, and is it taught or only supervised? A good answer is specific: who leads it, what the children produce, how you would know it was working.
- If we leave in January, what happens next? A good answer covers the refund terms in writing, the records your child takes with them, and how they would fit back into a standard curriculum. For an older child, add how colleges will read the transcript.
The questions for an ordinary school, on how it uses AI generally, are in 10 Questions to Ask Your Child's School About AI.
What I See in Schools
The pattern I see most often with adaptive software in ordinary schools is that it works best for the children who already have the habit of working alone, and that the teacher's most important job becomes finding the ones who do not. That is my observation from working in and around schools, not a finding from the studies above. It is also why the Times detail about self-discipline reads to me as the heart of the story and not a footnote.
Here is the question I would ask:
The Bad Day Question is the one I would ask any school that runs on software: what happens here on the day my child will not work? Nothing in a brochure answers it. A person in the building does.
Every child has that day. A school's answer to it shows you what it is made of.
Judge an AI school by its worst-served child, not its best-served one.
Which children might suit it, by age?
This is my judgment, not a finding. I have seen no study that sorts children by age for this model.
Under 8. I would be most cautious. Young children's reading, talking and play depend on adults responding to them, and the part of early reading that has to happen out loud is a good example (see Can AI Teach Kids to Read?). Ask how much of the day is spent with an adult, not a screen.
8 to 12. These are the years in which children learn to start hard things on their own. A model that assumes the habit may not build it. Ask how the school teaches self-direction, not whether children have it already.
13 to 15. Teens can work independently for stretches, and they can also find ways around a program. The crutch effect in the math study is the risk to ask about.
16 to 18. Ask how the school's transcript and qualifications will be read by the colleges or employers your child may apply to, and get the answer in writing.
For any age, the child's temperament counts for more than the age. A child who likes clear goals and working alone may find the structure freeing. A child who needs company, or finds starting hard, may find it lonely.
What should I say to my child about it?
Bring your child into the decision, and keep it open. You could say: "We're looking at a school where most of the learning happens on a computer in the morning. What would you want to know before we decide?"
If your child is happy where they are and worried about the idea, you could say: "Nothing is decided. I'm asking questions so that we know what's real and what's marketing."
If your child is excited, you could say: "I love that you're curious. Let's visit and ask what happens on a day when someone doesn't feel like working."
When do school rules or professional advice come first?
If your child has an individualized education program, a 504 plan or other additional needs, speak to their current special education team before anything else, and get in writing what support any new school will provide. Rules for private schools and for withdrawing a child from a current school differ by state, so check with your state's education department before you act. In the UK, the equivalent is a fee-paying independent school, and the same questions apply. If your child's wellbeing is the reason you are looking, a pediatrician or your family doctor comes before any school.
What should I do next?
Ask the eight questions, visit on an ordinary day, and talk to a family who left as well as one who stayed. The decision is yours, and a school that welcomes the questions is telling you something. If you are weighing this against staying where you are, Should Kids Use AI in School? covers how ordinary schools are handling the same question.
If your child's school is weighing how much of the day to hand to software, this is the strategy work I do with schools, and it is worth sending to your head of year or principal. I also write a Sunday newsletter that keeps teachers and parents up to date without the panic, and parents are very welcome.
Dan Fitzpatrick is a keynote speaker, Forbes contributor and five-times bestselling author, a former teacher and assistant headteacher, and the founder of The AI Educator, which has trained more than 150,000 educators across 30-plus countries. Read Dan's story.
