Funders rarely get perfect evidence before making investment decisions. Nowhere is that tension sharper right now than in AI-enabled tutoring.
In a new research brief, Mackenzie Flynn, Director of Client Engagements, and Emily McLane, Director of Research, examine what the research supports, where the gaps remain and where funders can shape how AI-enabled tutoring develops.
Opposite ends of the evidence spectrum
High-impact tutoring is validated by more than 150 randomized controlled trials, with results often ranging from half a year to more than a year of learning gains. AI is moving faster than research can keep up: Stanford’s SCALE initiative identified only 20 high-quality causal studies on the effects of AI on students or educators.
The brief maps three ways AI is being applied in tutoring today:
- Direct support for students. On-demand practice and feedback. The catch: the students most likely to use AI tutors are higher achieving, while those who may benefit most are the least likely to engage.
- Strengthening human tutors. The most promising evidence so far. In NSSA’s Tutor CoPilot trial, students with AI-supported tutors were 4 percentage points more likely to master content, and up to 9 points more with less-experienced tutors.
- A lever for cost and scale. AI may lower per-student cost without lowering quality, the constraint that has kept high-impact tutoring resource-intensive.
A $24 billion market is forming before the evidence is in
New federal policy could direct up to $24 billion a year toward education expenses, including tutoring, starting in 2027. As of June 2026, 28 states had opted into the Federal Scholarship Tax Credit program. Caregivers will become buyers, many for the first time, with few signals about quality.
For funders: invest in the questions the market won’t answer on its own, shape quality signals before the money moves, and fund evaluation that breaks out results by student population and access.
For organizations: design for the students least likely to opt in, and treat AI as a complement to human tutors, not a replacement.