· Nov 6, 2026
Khan Academy’s Kristen DiCerbo: Student interests did not improve Khanmigo tutoring metrics
Markdown · JSON · Original source
Chief Learning Officer Kristen DiCerbo reports that adding a list of student interests to Khanmigo’s instructions did not improve measured engagement, math solving, or conversation length in an A/B test of Tutor Me. The test ran for two weeks across 77,613 threads and included students whether or not they had interests. DiCerbo says the test measured prompting Khanmigo to use interests, but did not track when or how it actually used them. Cognitive engagement was not significantly improved; math solving and conversation length showed no measurable change. Cost per thread was lower without interests included. Khan Academy removed the feature, while DiCerbo stresses that this result concerns this particular approach, not whether interests matter in learning. She says future experiments could test different instructions and whether interests help particular students, content, or points in an instructional sequence.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
· Article
Source details
Original source date: Nov 6, 2026 (2026-11-06T21:36:31+00:00).
Date basis: Date stated by the publisher.
Added to Escape Meridian: 2026-11-08T00:32:51.302Z. First observed: 2026-11-06T21:45:23.134Z. Retrieved: 2026-11-08T00:32:04.917Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
The test did not check implementation fidelity: it did not examine conversations to establish when Khanmigo actually used student interests. The test included students whether or not they had interests, and did not track when or how interests were used; its results concern prompting the tutor to use interests, not verified use in responses.
Open question
Would testing different instructions or particular students, content, or points in an instructional sequence show that student interests help in some cases?
Source-stated
Kristen DiCerbo reports that adding student interests to Khanmigo’s instructions did not improve engagement, math solving, or conversation length in the Tutor Me test.
Adding a list of student interests to the AI instructions did not improve engagement, math solving, or conversation length.
Source-stated
The A/B test ran for two weeks across 77,613 threads and included students whether or not they had interests.
The test ran for two weeks across 77,613 threads.
Source-stated
The test measured prompting Khanmigo to use interests, but did not track when or how it used them.
The test included all students (whether or not they had any interests), and we did not track when or how Khanmigo actually used the interests.
Source-stated
DiCerbo reports that cost per thread was lower without interests included.
Cost per thread: This was lower without interests included (0.072 vs. 0.069, which is small but adds up over hundreds of thousands of threads; 95% chance of no interest to win).
Source-stated
Khan Academy removed the interests feature based on the test results.
Based on these results, we removed the interests feature from Khanmigo.
Source-stated
DiCerbo limits the conclusion to this approach and says future experiments could test other ways of using interests.
It means that this particular approach—adding a list of interests to an AI tutor’s instructions and injecting them into prompts—did not help students engage more deeply or solve more problems.
Source and contributor context
Khan Academy · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
Media declared by the original publisher; availability checked separately from the editorial reading.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Khan Academy
Education deployment with explicit evaluation and implementation limitations.
Editorial provenance
Version 1. Publication ID: 2e16d3ad-6eb6-49e1-85b4-feb8e0dd4cd5.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-08T00:32:51.302Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 7c7b4d5575c5c52e03cba8663e745b5aa273098c6a171033073485d7196740be. Source text hash: be6e5e5022087248a535aeb84094441b2399bec6017e006ab7f9f26d77b80070.
Revision and source-check history
Version 1 · 2026-11-08T00:32:51.302Z
Khan Academy’s Kristen DiCerbo: Student interests did not improve Khanmigo tutoring metrics
Chief Learning Officer Kristen DiCerbo reports that adding a list of student interests to Khanmigo’s instructions did not improve measured engagement, math solving, or conversation length in an A/B test of Tutor Me. The test ran for two weeks across 77,613 threads and included students whether or not they had interests. DiCerbo says the test measured prompting Khanmigo to use interests, but did not track when or how it actually used them. Cognitive engagement was not significantly improved; math solving and conversation length showed no measurable change. Cost per thread was lower without interests included. Khan Academy removed the feature, while DiCerbo stresses that this result concerns this particular approach, not whether interests matter in learning. She says future experiments could test different instructions and whether interests help particular students, content, or points in an instructional sequence.
Original source
· Article
Source details
Original source date: Nov 6, 2026 (2026-11-06T21:36:31+00:00).
Date basis: Date stated by the publisher.
Added to Escape Meridian: 2026-11-08T00:32:51.302Z. First observed: 2026-11-06T21:45:23.134Z. Retrieved: 2026-11-08T00:32:04.917Z.
Limits
The test did not check implementation fidelity: it did not examine conversations to establish when Khanmigo actually used student interests. The test included students whether or not they had interests, and did not track when or how interests were used; its results concern prompting the tutor to use interests, not verified use in responses.
Open question
Would testing different instructions or particular students, content, or points in an instructional sequence show that student interests help in some cases?
Source-stated
Kristen DiCerbo reports that adding student interests to Khanmigo’s instructions did not improve engagement, math solving, or conversation length in the Tutor Me test.
Adding a list of student interests to the AI instructions did not improve engagement, math solving, or conversation length.
Source-stated
The A/B test ran for two weeks across 77,613 threads and included students whether or not they had interests.
The test ran for two weeks across 77,613 threads.
Source-stated
The test measured prompting Khanmigo to use interests, but did not track when or how it used them.
The test included all students (whether or not they had any interests), and we did not track when or how Khanmigo actually used the interests.
Source-stated
DiCerbo reports that cost per thread was lower without interests included.
Cost per thread: This was lower without interests included (0.072 vs. 0.069, which is small but adds up over hundreds of thousands of threads; 95% chance of no interest to win).
Source-stated
Khan Academy removed the interests feature based on the test results.
Based on these results, we removed the interests feature from Khanmigo.
Source-stated
DiCerbo limits the conclusion to this approach and says future experiments could test other ways of using interests.
It means that this particular approach—adding a list of interests to an AI tutor’s instructions and injecting them into prompts—did not help students engage more deeply or solve more problems.
Source and contributor context
Khan Academy · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
Media declared by the original publisher; availability checked separately from the editorial reading.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Khan Academy
Education deployment with explicit evaluation and implementation limitations.
Editorial provenance
Version 1. Publication ID: 2e16d3ad-6eb6-49e1-85b4-feb8e0dd4cd5.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-08T00:32:51.302Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 7c7b4d5575c5c52e03cba8663e745b5aa273098c6a171033073485d7196740be. Source text hash: be6e5e5022087248a535aeb84094441b2399bec6017e006ab7f9f26d77b80070.
uncertain · 2026-11-08T00:32:51.302Z · anthropic/claude-opus-5.5
Publication: 2e16d3ad-6eb6-49e1-85b4-feb8e0dd4cd5
attribution — pass: The reading correctly attributes the post to Kristen DiCerbo, Khan Academy Chief Learning Officer, and keeps the findings and conclusions framed as her and Khan Academy's report rather than as established fact.
By Kristen DiCerbo, Khan Academy Chief Learning Officer
summary — uncertain: The facts are accurate, but a reader without background may not follow it. Khanmigo is never identified as Khan Academy's AI tutor. Tutor Me, A/B test and threads are not explained. The draft also drops the source's central framing: personalizing to student interests is a common hope for AI tutors, and DiCerbo calls the result counterintuitive. Smallest repair: identify Khanmigo as Khan Academy's AI tutor, add that common-hope framing, and gloss A/B test as the source does.
Our findings were counterintuitive to the common wisdom: using student interests didn’t increase engagement or make it more likely that a student would solve a problem.
date — pass: originalPublished (2026-11-06T21:36:31Z, from the publisher page) matches the feed entry. Nothing indicates an earlier first publication.
support — pass: Each claim has a verbatim, source-stated excerpt. That covers the null results, the two-week test across 77,613 threads, the lower cost, the feature's removal, the limitation on implementation fidelity and the scoped conclusion. The question follows from the source's future-work section without asserting an answer. No unsupported benefits or costs are implied.
In particular, we did not check implementation fidelity.
media — pass: The selected media is the verified direct link to the original Khan Academy blog post. It is the most direct way to access the source.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
Education deployment with explicit evaluation and implementation limitations.
Explore Khan AcademyEducation deployment with explicit evaluation and implementation limitations.
Explore Khan AcademyEducation deployment with explicit evaluation and implementation limitations.
Explore Khan AcademyEducation deployment with explicit evaluation and implementation limitations.
Explore Khan AcademyEducation deployment with explicit evaluation and implementation limitations.
Explore Khan Academy