Khan Academy Adds Interactive AI Diagrams and Teacher Practice Tools

Learn how Khan Academy's new Gemini-powered visual tutoring and teacher-led practice tools aim to improve student outcomes while maintaining classroom control.

Monday, August 31, 2026

Key Takeaways

  • An independent NBER study found that using Khan Academy with Khanmigo improved middle school math scores by 0.14 standard deviations over one year. This matched the gains from using Khan Academy without AI assistance.
  • Student engagement is the main hurdle for AI tutoring. Research shows the median student messaged the AI tutor in only 17% of sessions where they made a mistake.
  • Leading AI models fail to generate mathematically accurate diagrams 26.1% of the time because of limitations in spatial reasoning and geometric constraint mapping.
  • Under the redesigned Khanmigo Practice My Knowledge workflow, teachers must review and approve AI-generated questions before assigning them to students.

Khan Academy is upgrading its artificial intelligence tutor, Khanmigo, for the 2026 school year. Developed with Google, the tools let the AI generate interactive math and science diagrams while giving teachers full veto power over AI-created practice questions. As school districts adopt artificial intelligence, these updates aim to keep teachers in control of classroom instruction.

What Happened

The updates follow a six-month project with Google.org, which placed six Google engineers on Khan Academy’s product team, as detailed in the official Google announcement. Students using Khanmigo are no longer limited to text-based chat. The AI tutor can now generate dynamic geometric shapes, charts, and coordinate plots that students can manipulate, according to Khan Academy's product release. For example, if a student drags a vertex on a shape, the AI recognizes the movement and updates the lesson.

At the same time, the platform redesigned its "Practice My Knowledge" workflow. While Gemini drafts practice questions, teachers must review, edit, or reject them before they reach students. As noted in our coverage of K-12 AI governance, human oversight is a priority for school administrators who want to avoid AI hallucinations. This model ensures educators retain control over assignments and curriculum materials, as shown in case studies on teacher-controlled AI.

The Bigger Picture

Visual tools help students learn, but current AI models often struggle with mathematical accuracy. A 2026 study on math diagram generation by the AIMS Collaboratory found that models like Gemini and GPT-4o generate mathematically accurate diagrams only 73.9% of the time. The researchers noted that AI frequently fails at basic spatial reasoning and geometric constraints. To fix this, developers are testing specialized coordinate solvers and agentic quality loops to verify AI images before students see them.

The academic impact of AI tutoring is also under study. A two-year randomized trial across 18 middle schools by the National Bureau of Economic Research (NBER) found that students using Khan Academy with Khanmigo had achievement gains of 0.14 standard deviations over a school year. However, these gains were nearly identical to those of students using Khan Academy without AI. Student engagement was the primary issue; the median student only messaged the AI tutor in 17% of the sessions where they made an error. In contrast, an eight-week study published in Zenodo indicated that targeted integration of Khanmigo improved core math competencies in secondary students. As we previously reported, school districts need to encourage active student interaction to get results from AI.

What This Means for Families

These changes mean student homework is more interactive but still requires supervision. Interactive diagrams help visual learners grasp math and physics concepts that are hard to explain in text. However, since AI diagram generation is still developing, parents and teachers should watch for errors in visual aids.

Teacher-approval workflows also help protect student privacy and academic accuracy. The system prevents the AI from generating off-topic questions without a teacher’s review. Still, public reports from Khan Academy do not fully explain whether district curriculum uploads or student data are used to train Google's commercial Gemini models. Families should check their school district’s data privacy agreements.

What You Can Do

  • Remind your child to ask Khanmigo questions when they get stuck, since research shows students rarely start chats on their own during math errors.
  • Tell your student to double-check geometric shapes and coordinate plots generated by the AI, as visual models still make spatial errors.
  • Ask your school's administration to confirm that student data and curriculum uploads in Khanmigo are protected from Google's public model training.
Share: