Understanding Math ≠ Doing Math: The Harvard Research That Explains Your Kid's Struggle
Explanation-based teaching creates comprehension. Only practice creates skill. Harvard's Ying Xu found students who watched AI tutors understood each explanation but couldn't do the next problem alone. Here is why — and what works instead.
Your child watches the tutor solve it. They nod. They say "I get it." Then the next problem comes and they're stuck again. Harvard has a name for this — and it's not your child's fault.
There is a specific research finding that explains why so many hours of tutoring, YouTube videos, and AI-assisted homework leave your child no better off than before. The finding comes out of Harvard's Graduate School of Education, and it separates two things most people treat as the same: understanding a math problem and being able to do it.
The distinction is not semantic. It is neurological. And it explains why the current explosion of AI math tutors, video content, and homework helpers is producing students who feel confident right up until the moment they sit down to a test alone.
The Harvard finding: comprehension ≠ capability
Ying Xu at Harvard's Graduate School of Education studied children learning math through AI tutors. Her finding: when the tutor explained a problem, students understood the explanation. Their subjective sense of "I get it" was high. But when tested on a very similar problem immediately after — no tutor, same day, same room — a large fraction could not solve it independently.
Xu, Y. — Harvard Graduate School of Education, 2024
Students learning math via AI-explained tutorials reliably understood each individual explanation but could not independently solve subsequent problems of the same type. Explanation-based teaching creates comprehension; only structured practice creates skill. The two are neurologically distinct.
Why "I get it" isn't the same as "I can do it"
When your child watches someone else solve a problem, their brain is running comprehension circuits — the same regions that light up when they watch a movie or follow a story. Comprehension circuits are cheap. They fire fast. They feel like knowing.
Doing the problem yourself runs different circuits: retrieval, working memory, and motor planning if you're writing. These circuits only strengthen through use. Watching a tutor solve twenty problems strengthens comprehension of those specific problems. It does not strengthen the retrieval circuits your child will need at 8:47am on test day.
If your child says "I get it" but still can't do the next problem, they are not lying and they are not lazy. Their comprehension is real. Their skill is not built yet.
Why this pattern is worse now than five years ago
In 2019, a struggling student's default help was a paid tutor once a week — expensive, but structurally forced them to practice alone the other six days. In 2025, a struggling student's default help is an AI chatbot that will explain any problem, any time, for free. The child feels supported. They no longer practice alone. Comprehension is available on demand. Skill formation is not happening.
The result: a generation of students with the highest subjective sense of "understanding" in history, and standardised math scores that have declined every year since 2019 (OECD PISA 2022 results, US NAEP 2023 results). The scores are not a paradox. They are exactly what Xu's research predicts.
Sweller, J. — Cognitive Science, 1988
Working memory is limited to ~5–7 items and depletes rapidly during multi-step problems. Only automatic retrieval — built through repeated self-directed practice — bypasses this limit. Passive learning (watching, reading, being explained to) does not build the automatic retrieval that time-pressured problem-solving requires.
Three signs your child is stuck in the explanation loop
- "I get it" during help, blank stare during solo work. High comprehension confidence, low independent execution.
- Homework takes hours, tests are worse. Extended tutor-assisted homework produces above-grade comprehension and below-grade tested skill.
- Repeated topics keep needing re-explaining. Comprehension does not consolidate to memory without the retrieval work that only self-directed practice provides.
Test the difference in 5 minutes
Open the free ×11 lesson. Read the technique. Close the lesson. Try 65 × 11, 78 × 11, 47 × 11 without looking back. If you can, you built skill. If you can't, you had comprehension only.
Try the ×11 lesson (free) →Practice-first math: how Vedic techniques structurally build skill
Vedic Math avoids the explanation-loop trap by design. Every lesson is short (5 minutes to learn), the technique is a compact pattern (fits in working memory), and the immediate next step is the child solving a problem themselves — not watching another example. The pattern locks in through self-generated answers, not through more explanations.
This is not our opinion — it is the design constraint the research imposes. Sweller's cognitive load research, Rivera's neural-migration scans, Xu's explanation-vs-skill distinction, and the deliberate-practice literature all point to the same design: short chunks, immediate self-directed practice, spaced repetition. Vedic Math happens to fit this design because the sutras were built for oral transmission and mental execution, not for classroom lecture.
What "practice-first" looks like in a session
- Minute 0–3: concept + one worked example. Total explanation time: under three minutes.
- Minute 3–13: child solves 8–10 problems themselves, no scaffolding after the first two.
- Minute 13–15: spaced-repetition review of an older technique to keep it retrieval-active.
Common questions from parents
Q: Should we stop using the AI tutor?
Not entirely. Use it for the "I don't understand what the question is asking" moments. Do not use it as the default before your child has attempted the problem themselves.
Q: What about video math courses?
Same rule. Watching is comprehension; solving is skill. If a course is 80% videos and 20% practice, expect proportional results.
Q: How much practice is enough?
Rivera's neural-migration research suggests 8 weeks of daily 15-minute practice moves arithmetic from effortful to automatic. That is the target: not "understanding," not "getting it" — automatic retrieval that survives test day.
Try 4 practice-first lessons (free) → — No signup. Short lessons, immediate practice, no video lectures.
https://www.thevedicmind.com/learn/lesson/l1-multiply-11