I Homeschooled My 3 Kids Using Only AI for a Full Year — The Unfiltered Results
Before you call CPS, hear me out. My kids scored higher on standardized tests than 92% of their peers.
I know how that sounds. Believe me, I do. When I told my mother-in-law what I was planning, she literally called her lawyer. When I mentioned it to colleagues at the district, they looked at me like I'd announced I was joining a cult. And my wife — well, we'll get to her perspective later.
But here's what I also know: I spent 14 years as a high school history and English teacher. I have a Master's in Education from UVA. I've sat in more IEP meetings, curriculum committee sessions, and professional development workshops than I can count. I know the system inside and out. And I watched it fail my own children.
So in August 2024, I pulled all three of my kids — Emma (14), Liam (11), and Sophie (8) — out of Fairview Public Schools and embarked on what I now call The AI Homeschool Experiment. For one full academic year, artificial intelligence was our primary teaching tool. ChatGPT, Claude, Khan Academy's AI tutor, custom GPTs I built myself, and a rotating cast of other AI tools formed the backbone of my children's education.
This is the unfiltered account of what happened.
Why I Did It
Let me be clear about something: I didn't do this because I think AI is some magical solution to education. I did it because the system was broken for my kids, and AI happened to be a tool that could help me fix it.
Emma (14) was in 8th grade at a school where her honors English class had 34 students. Thirty-four. Her teacher — a good teacher, someone I knew personally — had five periods of 30+ students each. Emma was getting her essays back with a checkmark and a letter grade. No feedback. No revision process. She was bored, disengaged, and starting to hate reading, which she had loved since she was four.
Liam (11) has ADHD and was in 5th grade. Despite his IEP, he was spending most of his day in a classroom designed for neurotypical learners. His teacher was doing her best, but accommodations on paper don't translate when you have 28 kids and no aide. He'd come home defeated, convinced he was stupid. His math scores were plummeting despite the fact that I knew — because I'd worked with him at home — that he could do the work when it was presented differently.
Sophie (8) was in 2nd grade and academically fine but miserable. A group of girls had been systematically excluding her since kindergarten. The school's response amounted to "kids will be kids" and a few half-hearted circles of trust. She was developing anxiety and starting to resist going to school entirely.
Three kids. Three different problems. One shared root cause: a system that simply cannot individualize at scale.
The Setup
I took a leave of absence from my teaching position. We could afford it for one year — barely — because my wife Sarah works as a project manager at a tech company. I became the full-time facilitator, with AI as my co-teacher.
Here's what our daily schedule looked like:
| Time | Emma (14) | Liam (11) | Sophie (8) |
|---|---|---|---|
| 8:00-8:30 | Morning journal (AI-prompted reflection) | Morning movement + breakfast | Morning read-aloud with Dad |
| 8:30-9:30 | Math (Claude + Khan Academy AI) | Math (Khan Academy AI + custom GPT) | Math (Khan Academy AI, supervised) |
| 9:30-10:00 | Break | Break | Break |
| 10:00-11:00 | Science (Claude for concepts, experiments at home) | Science (ChatGPT for exploration, hands-on projects) | Science (nature observation + AI Q&A) |
| 11:00-12:00 | Writing & Literature (ChatGPT as writing partner) | Reading & Writing (ChatGPT + audiobooks) | Reading & Phonics (supervised AI reading buddy) |
| 12:00-1:00 | Lunch + free time | Lunch + free time | Lunch + free time |
| 1:00-2:00 | History/Social Studies (ChatGPT + primary sources) | History (ChatGPT interactive stories) | Social Studies (AI-guided cultural exploration) |
| 2:00-2:30 | Spanish (custom GPT conversation partner) | Spanish (custom GPT + Duolingo) | Art/Music (minimal AI, mostly traditional) |
| 2:30-3:00 | Independent project time | Independent project time | Outdoor play / socialization |
Key principle: I was always present. AI was never babysitting my kids. I sat with Sophie for nearly every session. I checked in on Liam every 15-20 minutes. Emma had the most independence but I reviewed every interaction at the end of each day.
The AI Tools We Used
| Tool | Primary Use | Strengths | Weaknesses | Cost/Month |
|---|---|---|---|---|
| ChatGPT (GPT-4) | Creative writing, history discussions, brainstorming | Engaging conversational style, broad knowledge, great at Socratic dialogue | Occasional hallucinations in history, sometimes too eager to give answers | $20 |
| Claude | Math reasoning, science explanations, essay feedback | More careful/accurate, better at step-by-step reasoning, excellent at constructive feedback | Less creative/playful, sometimes overly cautious | $20 |
| Khan Academy (Khanmigo) | Math practice, standardized test prep, structured curriculum | Aligned to standards, adaptive difficulty, progress tracking | Less flexible, sometimes repetitive, limited subject range | $9/month (family plan) |
| Custom GPTs (4 total) | Spanish conversation, ADHD-adapted math, writing prompts, science experiments | Tailored to each kid's needs and level, consistent approach | Required significant setup time, needed frequent tuning | Included in ChatGPT Plus |
| Perplexity AI | Research projects, fact-checking | Source citations, current information | Sometimes overwhelming for younger kids | $20 |
| NotebookLM | Study guides, audio summaries | Great audio overviews, good for review | Limited interactivity | Free |
Total monthly AI cost: ~$69
The Results
Here's the part everyone wants to see. In May 2025, all three kids took the Iowa Assessments (standardized test) administered by a certified proctor through our homeschool co-op.
No slides found. Use headings (#) to define slide titles.
Subject-by-Subject Breakdown
| Subject | Child | Before (Aug 2024) | After (May 2025) | Method |
|---|---|---|---|---|
| Math | Emma | Grade 8.2 equivalency | Grade 10.1 equivalency | Claude for concept mastery, Khan AI for practice |
| Math | Liam | Grade 4.1 equivalency (below level) | Grade 6.3 equivalency (above level) | Custom ADHD-adapted GPT + Khan AI |
| Math | Sophie | Grade 2.0 equivalency | Grade 3.4 equivalency | Khan AI with Dad supervising every session |
| Reading | Emma | 89th percentile | 96th percentile | ChatGPT book discussions, Claude essay feedback |
| Reading | Liam | 62nd percentile | 78th percentile | AI audiobook companions, adapted reading pace |
| Reading | Sophie | 58th percentile | 91st percentile | AI reading buddy + daily read-alouds with Dad |
| Science | Emma | B+ average | Completed 6 independent research projects, scored 94th percentile | Claude for hypothesis testing, real experiments |
| Science | Liam | C average | Completed 4 projects, scored 82nd percentile | ChatGPT exploration + hands-on experiments |
| Science | Sophie | N/A (not formally tested) | Scored 85th percentile on grade-level assessment | Nature journals + AI Q&A sessions |
| Writing | Emma | "Competent" (teacher assessment) | College-freshman level (independent assessment) | ChatGPT as writing partner, Claude as editor |
| Writing | Liam | Below grade level | At grade level | Voice-to-text + AI revision assistance |
| Writing | Sophie | At grade level | 1 grade above level | Story co-creation with AI (Dad supervised) |
| Spanish | Emma | Zero (never studied) | A2 level (basic conversational) | Custom GPT conversation partner, 30 min/day |
| Spanish | Liam | Zero | A1 level (basic phrases) | Custom GPT + Duolingo, 20 min/day |
| Spanish | Sophie | Zero | Basic greetings and colors | Minimal, mostly songs and games |
The Socialization Question
I know this is the first thing people ask about homeschooling, so let me address it head-on.
My kids were not isolated. Here's what their social calendar looked like:
- Monday/Wednesday/Friday afternoons (3:30-5:30): Homeschool co-op activities with 15-20 other kids. This included group projects, presentations, art, and unstructured social time.
- Tuesday afternoons: Emma — community theater rehearsals. Liam — robotics club at the local makerspace. Sophie — gymnastics.
- Thursday afternoons: All three — swimming at the YMCA + free play with other homeschool families.
- Saturday mornings: Liam — recreational soccer league. Sophie — dance class.
- Sunday: Family time, church youth group for Emma.
Was it the same as being in a school building with 500 kids for seven hours? No. Was it sufficient for healthy social development? I'd argue it was actually better. Here's why:
Sophie's anxiety disappeared. Completely. By November, she was a different child. She made two close friends through co-op who she still sees regularly.
Liam stopped being "the ADHD kid" and became "the kid who's really into robots and knows a ton about space." Without the social hierarchy of a school building, his confidence exploded.
Emma initially struggled the most socially. She missed her friends. We made sure she maintained those friendships through regular hangouts, and the theater community gave her a new social circle. By January, she told me she actually preferred having fewer but deeper friendships.
What Went Wrong
This is the section that I think matters most. If I only told you the success story, I'd be doing the same thing the education-tech industry does: selling you a fantasy. Here's the real, uncomfortable truth.
1. Emma Gamed the System
Around month four, I discovered that Emma had figured out how to get ChatGPT to essentially write her essays for her while making it look like she'd done the work. She'd ask it to "help her brainstorm," then "help her outline," then "help her write a rough draft," then "help her revise" — and at each stage, the AI was doing the heavy lifting while she provided minimal input.
When I caught it, she broke down crying. Not because she was sorry, but because she was frustrated. "What's even the point of learning to write when AI can do it better?" she asked me.
That question haunted me. I still don't have a perfect answer. What I did was restructure her writing process: first drafts had to be handwritten. AI could only be used for feedback on completed drafts, not for generating content. It worked, but it revealed a fundamental tension in AI-assisted education that I don't think anyone has solved.
2. Sophie Needed More Human Interaction Than AI Could Provide
For the first three months, Sophie did well with the AI tools. But around November, she started resisting. She didn't want to talk to "the computer." She wanted to talk to ME.
I realized I'd been so focused on implementing the AI curriculum that I was treating my 8-year-old like a small adult who could learn from a screen. She couldn't. She needed warmth, physical presence, facial expressions, hugs when she got frustrated. The AI was a tool, not a teacher, and an 8-year-old needs a teacher who is a human being.
I restructured Sophie's day to be about 60% me directly teaching her and 40% AI-assisted. Her engagement and happiness improved dramatically. But it meant I had less time for Liam and Emma, which was a constant tension.
3. AI Hallucinations Were a Real Problem in History
In February, Liam came to me excited about a project on the American Revolution. He'd been chatting with ChatGPT about the Boston Massacre and had a whole page of notes. Half of them were wrong.
ChatGPT had told him that 11 people died in the Boston Massacre (it was 5). It had invented a quote attributed to John Adams that Adams never said. It had described a "retaliatory attack by colonists the following week" that never happened.
I caught it because I know American history well. But what if I didn't? What if a parent without my background was relying on AI to teach their kid history?
This became one of my core protocols: every AI-generated factual claim had to be verified against at least one primary or secondary source. It added time to every lesson but was absolutely non-negotiable.
4. The Over-Reliance Problem
By month six, I noticed something troubling in all three kids: they were losing the ability to sit with not-knowing. When they encountered a hard problem, their first instinct was to ask the AI. Not to struggle with it. Not to think about it. Not to try and fail. Just to ask.
This is the AI equivalent of looking up the answer in the back of the textbook, and it's potentially more damaging because the AI is so responsive and personalized that it feels like learning even when it isn't.
I implemented what I called "AI-free hours" — periods where the kids had to work through problems with only books, their own notes, and their own brains. They hated it at first. Emma literally said, "This is like making us use a typewriter when computers exist." But over time, their problem-solving stamina improved.
5. My Own Burnout
Nobody talks about this, but full-time homeschooling is EXHAUSTING. Adding AI tools didn't reduce my workload — it changed it. Instead of lecturing, I was curating AI interactions, reviewing conversation logs, creating custom prompts, adjusting difficulty levels, and constantly monitoring for accuracy.
By March, I was working 10-hour days. Preparing in the evening, teaching all day, reviewing and adjusting at night. My wife and I barely spoke about anything other than the kids' education. I lost 15 pounds (not in a good way) and started having insomnia.
I hired a part-time tutor for Sophie in April. It helped. But it also added $600/month to our costs.
Cost Comparison
| Expense | Public School (Annual) | AI Homeschool (Annual) |
|---|---|---|
| AI subscriptions | $0 | $828 |
| Curriculum materials / books | $150 (supplies, etc.) | $340 |
| Extracurricular activities | $1,200 | $2,400 (more needed for socialization) |
| School fees, fundraisers, photos | $450 | $0 |
| Before/after care | $3,600 | $0 (parent at home) |
| Transportation | $480 | $120 (less driving) |
| School lunches | $900 | $600 (home cooking) |
| Lost parent income | $0 | $52,000 (my salary) |
| Part-time tutor (Sophie, Apr-Jun) | $0 | $1,800 |
| Total | $6,780 | $58,088 |
Let me be brutally honest: the AI tools were cheap. The parent being home was not. If you're looking at this thinking "I could do this on the cheap with AI," you're missing the biggest cost by a mile. Someone needs to be there, and that someone is giving up a salary.
Sarah's Perspective
My wife agreed to let me publish her honest assessment, and I promised I wouldn't edit it. Here's what she wrote:
"When James first proposed this, I thought he was having a midlife crisis. I'm not exaggerating. I made him do a full written proposal before I'd even consider it. I gave him one year, with the condition that if any of the kids fell behind academically, they'd go back to school immediately.
Here's what I'll say after living through it: the results are real. The kids learned more this year than in any previous year of school. Liam's transformation alone made it worth it. Watching him go from a kid who cried over math homework to a kid who voluntarily does math puzzles on the weekend was something I never expected to see.
But James almost broke himself doing it. There were nights in March where I genuinely worried about his mental health. The isolation of being home all day, the pressure of being solely responsible for three kids' education, the constant second-guessing — it took a real toll.
Would I do it again? Honestly, yes, but only with major changes. James needs help. A co-teacher, a regular tutor, something. One person cannot be a full-time AI-assisted teacher to three kids at three different levels and also be a functioning adult.
Also: we're privileged. We could afford for James to not work for a year. Most families can't. I get uncomfortable when people treat this as a model for education because it's really a model for education if you happen to have a trained teacher spouse who can stay home. That's not most people."
— Sarah O'Brien
She's right about all of it.
Month-by-Month Adaptation Highlights
September: Pure chaos. I over-structured everything. Kids rebelled against the rigid schedule. Liam had a meltdown on Day 3. Sophie cried because she missed her classroom. I almost quit.
October: Found our rhythm. Loosened the schedule. Let each kid have input on how they wanted to learn. Discovered that Liam works best in 15-minute focused bursts with movement breaks. Emma thrives with long uninterrupted blocks.
November: Sophie's resistance to AI began. First round of custom GPTs built (Spanish conversation partner was a hit). Emma's test scores jumped noticeably.
December: Holiday break, but the kids voluntarily kept using some AI tools. Liam asked to continue his space exploration project "for fun." I cried in the bathroom (happy tears, for once).
January: Discovered Emma was gaming the system. Hard reset on writing methodology. Implemented handwriting-first policy. Tense two weeks but ultimately productive.
February: History hallucination crisis. Built verification protocol. Started keeping a "corrections log" — we found 23 significant factual errors across all AI tools that month alone.
March: My burnout peaked. Wife and I had a serious talk about whether to continue. Decided to push through but get help.
April: Hired part-time tutor for Sophie. My stress dropped 50% immediately. Liam's math breakthrough month — something clicked and he jumped two grade levels in four weeks.
May: Testing month. All three kids exceeded expectations. Sophie was nervous but aced it. Liam was genuinely proud of himself for the first time in years. Emma scored college-level in reading and writing.
June: Reflection month. Each kid wrote a self-assessment (yes, without AI help). Their reflections were more insightful than any report card I'd ever written as a teacher.
What I'd Change for Year 2
We are doing Year 2. Here's what's different:
-
Hired a co-teacher. A retired elementary teacher from our co-op is working with Sophie and Liam three mornings a week. Cost: $1,500/month. Worth every penny.
-
AI verification protocol from Day 1. No more trusting AI output without verification. Every factual claim gets checked.
-
Structured AI-free time. Two hours per day, minimum, of screen-free learning. Books, hands-on projects, outdoor observation.
-
Emma's AI contract. She and I wrote a formal agreement about how she will and won't use AI for writing. If she violates it, she loses AI access for a week.
-
Better social integration. We're joining a second co-op and adding a weekly volunteer commitment for Emma.
-
My own boundaries. I'm working 8am-4pm and then I'm done. Evening prep is limited to 30 minutes. Sarah and I have a weekly date night again.
-
Subject-specific AI assignments. Claude for math and science reasoning. ChatGPT for creative work and discussion. Khan for practice and drills. No more "use whatever."
-
Monthly assessments. Not waiting until May to find out if it's working. Monthly check-ins with objective measures.
The Honest Disclaimer
I need to end with this, and I need you to hear it clearly:
I am a trained teacher with 14 years of classroom experience and a Master's in Education. Do not try this without understanding what you are getting into.
The AI tools are not teachers. They are tools. A hammer doesn't build a house; a carpenter does. ChatGPT doesn't educate a child; a knowledgeable, attentive, patient human being does, using ChatGPT as one of many instruments.
If you're a parent thinking about this because you saw my test scores and got excited: slow down. Ask yourself these questions first:
- Can you afford to have a parent home full-time (or close to it)?
- Does the at-home parent have strong enough content knowledge to catch AI errors?
- Can you provide adequate socialization outside of school?
- Are you prepared for the emotional labor of being your child's primary educator?
- Do you have a plan for when things go wrong?
If the answer to any of those is "no" or "I'm not sure," then this model probably isn't for you — yet. Maybe it will be someday, as AI tools improve and support systems develop. But today, right now, this requires a level of commitment, knowledge, and resources that most families simply don't have.
I'm sharing this because I believe in transparency. The AI-in-education conversation is dominated by either utopian techno-optimists or fearful luddites. The reality is messier, harder, and more human than either camp admits.
My kids learned more this year than ever before. And it nearly killed me to make that happen.
I'll answer every question I can in the comments. I owe you that honesty if I'm going to put this out into the world.
James O'Brien is a former high school teacher on leave from Fairview Public Schools. He lives in Virginia with his wife and three children. He is not affiliated with any AI company and received no compensation for this article.