← All education signals
Education

The Tutor Nobody Opens

A personal tutor built to close the achievement gap sits mostly unused, and the industry is now redesigning around the students who never open it.

What is it?

Khan Academy built a personal tutor for 700,000 students, yet only fifteen percent regularly open it — a real, measured gain sitting against a much larger group who never touch the tool at all. How deep the design problem runs is the open question, not the model underneath.

Why now?

The category assumed capability was the bottleneck: build a good enough tutor and adoption would follow the way traditional one-on-one instruction never scaled to. Khan Academy’s own numbers say otherwise — usage, not model quality, decides the return. Everything turns on getting a reluctant student to open the tab at all.

Who’s building it?

Education platforms redesigning around engagement, and the students whose habits decide if it works.

Khan AcademyEd-tech platformsTeachers · parentsReluctant studentsResearchers
The full read
Why now, who is building it, where the market goes, and what stays unsettled.
Get weekly →
01 · The market
Who’s in it
Builders — education platforms shifting spend toward adoption design.
Buyers — schools and parents who bought the promise of closing the achievement gap.
Gatekeepers — the students themselves, who decide by simply opening the tab or not.
What’s at stake
The tool works when opened, but a fifteen-percent rate means most promised value stays unrealised — capability was never the bottleneck.
A modest, real gain proves the model works for users.
The redesign has to solve motivation, not intelligence.
What’s unsettled
Whether redesign can meaningfully move the open rate at all here.
Whether the measured gain scales once usage does too.
Whether adoption design is a solvable product problem or a durable behavioural one.
02 · How the demand matureda progression, not a timeline
Reach
Khanmigo scales to roughly 700,000 students with access across schools.
Engagement gap
Only fifteen percent of those students regularly open the tool at all.
Measured gain
Product testing finds a real, modest improvement in independent problem-solving among the students who used it.
Redesign phase
Summer work shifts investment from capability toward getting reluctant students to open a tab.
03 · Where the market goesthe forks, not a prediction
→Solve engagement, or the category stalls — decided by whether redesign moves a fifteen-percent open rate meaningfully at all.
→The measured gain could compound meaningfully if usage broadens well beyond its current, narrow levels.
→Or it stays a niche tool used only by the already-motivated minority of students.
→Adoption design becomes the real product here, not the tutoring model itself.
Everything here turns on generalisation: whether a small, measured gain among engaged users predicts anything about the much larger group who never open the tool.
04 · The Bespoke Default deskadjacent markets, moving now
The Always-On Tutor
The clearest measured gain in AI tutoring.
AI tutoring lifted Nigerian students’ scores 1.5-2 school years in 6 weeks.
World Bank
Generative Learning
The same result, framed as a smaller effect size.
AI tutoring lifted Nigerian students’ scores 0.23 SD in 6 weeks.
World Bank
Pace of One
Personalisation depends on the tutor knowing the student.
A tutor learns your history, Khan Academy says.
Khan Academy Blog
05 · The Wire
01
Most students never open it.
EdTech Innovation Hub ↗
02
A modest, measured gain.
Khan Academy Blog ↗
03
Median student messaged Khanmigo on only a third of practice days
NBER ↗
The panel on this signal
They agreed on
–Capability was never the real bottleneck; engagement design is now the unsolved problem for the category.
–The measured gain is real but modest in scale.
–Adoption, not model quality, decides the whole category’s actual return on investment.
The learning scientist disagreed on
–A reported engagement rate in the mid-teens percentage range represents a substantial gap between deployment and adoption; closing that gap enough to deliver the achievement-gap outcomes originally promised has not been demonstrated by any redesign effort to date.
Know more about the method →
✦How this signal was built
01 · Panel
A Delphi panel of 4 AI analysts, each reasoning independently.
→
02 · Debate
They challenge and revise each other’s reads over two rounds.
→
03 · Evidence
Every claim is tied to a cited, linked source on The Wire.
→
04 · Synthesis
A moderator writes it into one read, disagreement and all.