District leaders have more edtech choices than ever, and every product comes with claims about what it can do for students.
We think one question should come first: Does it improve learning?
As districts evaluate their edtech options, four things can help separate promising claims from meaningful evidence. Look for a product that:
- Is rigorously evaluated by independent third-party researchers
- Shares practical implementation guidance grounded in real classrooms
- Is clear about what it can and can’t do
- Keeps improving based on evidence from students and educators
These priorities also align with areas highlighted in the Department of Education’s recent “Dear Colleague” letter on evaluating edtech.
They are useful questions to ask of any product, including ours. Here’s how Khan Academy measures up, and where we’re still learning.
1. Rigorously evaluated by independent third-party researchers
District leaders should be able to review the evidence behind an edtech product. That is why we cite independent, third-party studies of Khan Academy (and publish our own efficacy research too). Four randomized controlled trials (RCTs) run by independent researchers who were not commissioned by nor paid by Khan Academy have tested Khan Academy’s impact on math. As a benchmark to interpret the results that follow, the median effect across 747 RCTs of education programs is 0.10 standard deviations (SD) (Kraft, 2020).
Visual based on Kraft (2020) benchmarks for causal studies of preK–12 achievement
Three RCTs found gains. In Hamilton County, Tennessee, about 6,900 middle schoolers below grade level who used Khan Academy during a daily intervention period scored 0.06 SD higher over two years, rising to 0.08 SD in year two. In Arlington ISD, Texas,, students in grades 3 to 6 gained 0.12 SD. In grades 7 and 8, where practice averaged about 4 minutes a week, there was no effect. In El Salvador, students in grades 3 to 6 gained 0.09 to 0.11 SD. The fourth RCT, which looked at 157 schools in Brazil, found no effect on math, but the study was conducted amid widespread internet outages and device problems.
The three studies with gains had one thing in common: strong implementation. Each had protected practice time and support. That pattern is what our implementation guidance below is built on.
Access to an effective edtech tool is not the same as effective implementation.
As a practical benchmark, we recommend aiming for students reaching Proficient or Mastered on about two Khan Academy skills each week, or roughly 60 skills over the course of the school year. Recent research on how practice relates to learning is consistent with these targets. A 2026 study in PNAS of more than 200,000 students, conducted by Khan Academy researchers and three university researchers, found that practice improves learning: about 0.0031 SD per Khan Academy skill brought to Proficient, or roughly 0.18 SD for 60 skills.
In districts where Khan Academy is working well, we tend to see a few things in common: a clear instructional goal, regular practice, teacher professional learning, clear ownership, and ongoing review of usage and learning data.
We also have causal evidence that this support matters. In a randomized trial in Uttar Pradesh, India, 74 schools had access to Khan Academy. Only 24 were randomly assigned facilitators who visited twice a week to protect practice time, troubleshoot, and monitor progress. Students in those schools practiced about 47 minutes a week and scored 0.44 to 0.47 SD higher in math after seven months.
We see a similar pattern in districts that prioritize implementation support: more than 20% of learners reach our recommended goals, compared with about 10% overall. We share more about those patterns in “Secrets of Success in Our Most High-Use School Districts.”
District teams can also draw on implementation resources, educator training, and district case studies to see not just what to use, but how to use it well.
3. Clarity about what the edtech product can and can’t do
Khan Academy is built to help students learn through mastery-based practice. It can help identify readiness for grade-level content, recommend what to work on next, provide instruction and feedback, and make progress visible to students, teachers, and district leaders. That practice is based on instructional content created and reviewed by experts on our content team.
Our chief learning officer has written about why districts should start with the educational problem rather than with the technology.
We are also clear about what Khan Academy cannot do on its own. Making the product available is not enough. Stronger results come when districts connect it to a clear instructional goal, give students consistent time to practice, and support teachers in using it well.
We hold our AI tools to the same standard. For students, Khanmigo is designed to tutor rather than simply give answers. For teachers, AI can help with planning, creating materials, and adapting instruction while educators stay in control of what reaches students and how it is used.
We are also explicit about AI’s limitations. It can be wrong, misleading, or inappropriate. We documented issues with Khanmigo’s math accuracy in the past and the steps we took to improve it (a post on math). We’ve also been open about where Khanmigo has fallen short of our expectations for engaging students and the work we’ve done to improve it.
On safety and student data, we publish specifics. We do not sell personal information or allow third-party advertising. Large language model providers are not allowed to train on student or teacher data, and student names are not shared with LLM providers when teachers ask for classroom insights.
For districts evaluating AI tools, that kind of clarity should be expected.
4. Keep improving based on evidence from students and educators
Sometimes improving an edtech product means admitting that it is not working well enough.
In 2025, we worked with about 5,100 students across five U.S. districts to reimagine Khan Academy and Khanmigo for classrooms. What we learned changed the product.
For example, Khanmigo was originally designed notto give a student the answer. Classroom data and teacher feedback showed us that the rule was too rigid. Now Khanmigo still encourages students to try first, but after an incorrect attempt, it can become more direct and help students understand what went wrong.
Before we make a change widely available, we look at whether it helps students learn and stay engaged. Some changes have helped. Others have shown no measurable effect.
We share those product-testing results too, including what worked and what didn’t, so districts can see how the evidence is shaping the product.
That is the standard we want to keep meeting: testing in real classrooms, measuring what matters, learning from the results, and continuing to improve for students and the educators who support them.
So how do we grade ourselves?
We are proud of the growing evidence behind Khan Academy, but there is still more research to do. We have learned a lot about strong implementation, and we will keep making that guidance easier for districts to use. We also know that transparency means being clear about what Khan Academy and AI can do, what it cannot do, and where we still need to improve.
What could stronger learning look like in your district?
Explore how Khan Academy Districts brings together mastery-based learning, insights into student progress, and implementation support to help your schools put evidence into practice.
