Khan Academy
genAI conversational tutor at the scale of a school district
Khanmigo, Khan Academy's GPT-4 genAI tutor, reached 2.0 million users worldwide in 2024-25, including 770,000 students in US district partnerships, with signups growing 731% year over year.
Key points
- A genAI conversational tutor that guides the student without giving the answer.
- Khanmigo built on OpenAI GPT-4, with student and teacher versions.
- 2.0 million users in 2024-25, including 770,000 students in US districts.
- +731% signups year over year, evidence level B confirmed.
Objective
Get entire school districts to adopt Khanmigo as a student tutor and teacher assistant, in order to anchor Khan Academy in daily classroom use and convert free consumer usage into paid district partnerships.
The deployment
Khanmigo is a genAI tutor built on GPT-4. For the student, it guides step by step without giving the answer directly, in the spirit of Socratic questioning, and follows the context of the exercise in progress. For the teacher, Khanmigo for Teachers prepares lesson plans, drafts instructions, and helps with grading. The product is sold to districts (the Khan Academy Districts program) with a tracking dashboard, and remains freely available to individual teachers in more than 70 countries.
Results Proof B
Deployment figures published by Khan Academy in its official SY24-25 annual report (primary source from the subject brand), corroborated by a state government announcement (New Hampshire) confirming a real deployment at the state level.
How it works
Documented architectureThe stack in detail
- llm OpenAI GPT-4 language model powering the tutor and the teacher assistant since the March 2023 launch
- plateforme Khanmigo (integration in-house) Khan Academy's product layer: tracking of the exercise context, Socratic prompts, student and teacher versions
- outil Garde-fous pedagogiques et de securite rules preventing the answer from being given directly and filtering content unsuitable for minors
- outil Tableau de bord district aggregate usage tracking for district partnerships (Khan Academy Districts program)
How it runs, concretely
For ops teams-
1District provisioning client
The district signs the partnership, creates student and teacher accounts, and sets up access.
-
2Student interaction AI
The student works through an exercise and calls on Khanmigo, which asks questions and guides rather than answering.
-
3Teacher support marketing
The teacher generates lesson plans and instructions through Khanmigo for Teachers and tracks student activity.
-
4Guardrails and tracking data team
Khan Academy applies guardrails on outputs and tracks aggregate usage for the district.
The quality of step-by-step guidance: the tutor must steer without giving the answer. If the LLM releases the solution or makes a mistake, the educational value and the teacher's trust collapse.
How your customers perceive this type of use
Sourced studiesLes consommateurs n'acceptent pas les chatbots par defaut : 64% prefereraient que les entreprises n'utilisent pas d'IA dans leur service client (Gartner, 2024) et pres d'un utilisateur sur cinq du service client par IA n'en retire aucun benefice (Qualtrics, 2025). L'acceptation se construit sur trois conditions mesurees par Salesforce : savoir qu'on parle a une IA, pouvoir escalader vers un humain, comprendre la logique de l'agent.
Acceptance conditions
- Etre informe qu'on parle a une IA et non a un humain (pres de 75% le demandent, Salesforce 2024)
- Un chemin d'escalade clair vers un agent humain (45% plus enclins a utiliser l'agent IA, Salesforce 2024)
- Une logique de l'agent clairement expliquee (44% plus enclins, Salesforce 2024)
Red lines
- Rendre l'humain injoignable : c'est la premiere inquietude des consommateurs sur l'IA dans le service client (Gartner 2024) et 50% craignent que l'IA les coupe du contact humain (Qualtrics 2025)
- Remplacer le service client par l'IA sans alternative : 53% envisageraient de partir chez un concurrent (Gartner 2024)
Sources: Salesforce 2024 · Gartner 2024 · Qualtrics 2025
How to replicate
Inference, not sourcedData prerequisites
- structured educational content aligned to the curriculum
- context of the exercise in progress (question, answer, level)
- strict guardrails on LLM outputs (do not give the answer, no content unsuitable for minors)
Org prerequisites
- sales model for institutions (districts, schools)
- student data governance and compliance
- process for continuous evaluation of educational quality
Possible stack
- LLM (OpenAI, Anthropic, or equivalent)
- Socratic prompt and guardrail layer
- tracking dashboard for the institution
The plan, step by step
- Step 1Define the central educational guardrail (guide without giving the answer) and the subject scopeDeliverable: Tutor spec + answer evaluation set
- Step 2Connect the LLM to the educational content and the context of the exercise in progressDeliverable: Working prototype on one subject
- Step 3Evaluate educational quality and lock down compliance for minors' dataDeliverable: Evaluation report + validated legal framework
- Step 4Pilot in a school with volunteer teachers, student and teacher versionsDeliverable: Teacher feedback + real usage metrics
- Step 5Build the institution dashboard and the sales model for schoolsDeliverable: Institutional offer ready, sales cycle underway
First step: Define the central educational guardrail (the tutor guides without giving the answer) and validate it on one subject before opening a pilot to a school.
The same pattern, proven in other industries
Amazon
genAI shopping assistant
Booking.com
genAI trip planning assistant
Burberry
genAI copilot for consultants (search, image recognition, complete-the-look)
Sources
- S1 Khan Academy Annual Report SY24-25 Primary archive pending
- S2 Khan Academy to extend its AI services, at no cost, to New Hampshire educators and students (NH Department of Education) Established press archive pending
- S3 What's New for the 2025-26 School Year: Big Updates from Khan Academy Districts Interested party archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.