A Candid Number in a Hype-Heavy Market
In a sector where every vendor claims its AI tutor is transforming learning, Khan Academy just did something unusual: it admitted that most students who can use Khanmigo do not. Only 15 percent of students with access to the AI tutor regularly engage with it. That figure cuts against the prevailing narrative that AI tutoring is sweeping classrooms, and we think the candor is more useful than another round of adoption press releases.
The honesty matters because Khan Academy is one of the most credible names in the space, a nonprofit with deep classroom relationships rather than a venture-backed startup chasing growth metrics. When it concedes that access does not equal use, every other AI tutoring vendor's engagement claims deserve a second look. The gap between a tool being available and a tool being used is exactly where edtech budgets quietly evaporate.
The Usage Data Behind the Decision
The numbers are not small in absolute terms. Khanmigo has logged more than 108 million interactions since its 2023 rollout and averages roughly 269,000 interactions on a typical school weekday. That is real engagement at scale. But spread across the population of students who have access, the regular-use rate still lands at 15 percent, which tells a story about depth versus breadth.
This is the kind of granular telemetry that separates serious edtech operators from the rest. Khan Academy is not guessing about engagement; it is measuring it down to the chat level and using that data to drive a product overhaul. For enterprise learning leaders, the lesson is direct. If you cannot see who is actually using a learning tool and how, you cannot tell whether your investment is working, and you certainly cannot fix it.
From Passive Helper to Proactive Coach
The redesigned Khanmigo changes its fundamental behavior. Instead of waiting for a student to ask for help, the new version guides students through assignments proactively. It offers differentiated support before and after a student attempts a problem, adapts based on whether the learner is seeing material for the first time or reviewing it, and detects prerequisite skill gaps that quietly sink comprehension. The shift is from a chatbot that sits idle until summoned to a coach that intervenes at the right moment.
That design choice reflects a hard-won truth about learners. Most students do not know what they do not know, so a tool that only responds to questions misses the students who need it most. Kristen Eignor DiCerbo, Khan Academy's chief learning officer, captured the nuance, noting that "some chats help students move forward more than others." The redesign is an attempt to engineer more of the chats that help.
Practice Still Beats Conversation
Notably, Khan Academy is not positioning the AI tutor as a replacement for the work of learning. DiCerbo was explicit that "the learning that happens through practice remains the foundation." That framing is a quiet rebuke to the more grandiose claims in the market, where some vendors imply that conversing with an AI is itself sufficient instruction. Khan Academy is treating Khanmigo as a support layer around practice, not a substitute for it.
We find that restraint encouraging. The most durable edtech products tend to be the ones that respect how learning actually works rather than the ones that promise to shortcut it. Anchoring an AI tutor to deliberate practice, and using engagement data to make that practice stick, is a more defensible thesis than betting that students will learn by chatting.
What District and Enterprise Buyers Should Take Away
The redesigned Khanmigo rolls out to all district partners in summer 2026, following a year of pilots with select districts. That cadence, pilot, measure, then scale, is exactly the discipline buyers should demand. Khan Academy did not push the new design to everyone at once; it tested, gathered data and is now expanding deliberately.
For CIOs evaluating AI learning tools in any setting, including corporate training, the Khanmigo story is a useful template. Insist on real usage data, not license counts. Favor tools that intervene proactively rather than waiting to be asked. And be skeptical of any vendor that cannot tell you, with numbers, how many of your people actually use what you bought. Khan Academy just showed what that honesty looks like.
