The research supports checking understanding, but does not validate this quiz or establish the effectiveness of premium private AI mentorship. Three distinctions inform the desk’s approach:
| Evidence reviewed | What it supports | What it does not establish |
|---|---|---|
| A 2025 randomized study of high-school mathematics found that an unrestricted AI interface improved assisted practice but reduced subsequent unassisted performance; a pedagogically constrained version mitigated that negative effect. | Assess learning separately from performance while a tool is helping. | That all AI use harms learning, or that the same effects occur in private coding mentorship. (pmc.ncbi.nlm.nih.gov) |
| Stanford’s National Student Support Accelerator standards address tutor preparation, formative assessment, relationships, continuity and progress monitoring. | Examine teaching processes and learner progress, not credentials alone. | That a supplier’s documentation proves implementation, or that school-based tutoring results transfer automatically to this niche. (nssa.stanford.edu) |
| AI4K12’s draft guidelines organize AI concepts across grade bands from K–2 through grades 9–12. | Distinguish learning AI concepts from entering an advanced programming course. | A universal starting age, compulsory Python prerequisite or entitlement to use a consumer chatbot. (ai4k12.org) |
A February 2026 Harvard podcast also discusses judging AI within the wider context of a child’s relationships, interests and activities. That is useful expert perspective—not experimental evidence for a particular tutoring package. (news.harvard.edu)
Sources cited on this page
- Generative AI without guardrails can harm learning: Evidence from high school mathematics - PMC ↗pmc.ncbi.nlm.nih.gov
- Tutoring Quality Standards | National Student Support Accelerator ↗nssa.stanford.edu
- Grade Band Progression Charts – AI4K12 ↗ai4k12.org
- Preserving learning in the age of AI shortcuts — Harvard Gazette ↗news.harvard.edu