As schools enter the fourth academic year since ChatGPT hit the scene, artificial intelligence tutoring tools are increasingly finding their way into classrooms.

The number of K-12 students using nonprofit Khan Academy’s Khanmigo tutoring tool, for example, jumped from 40,000 to 700,000 between the 2023-24 and 2024-25 school years, according to Kristen DiCerbo, Khan Academy’s chief learning officer. And that number is expected to continue rising to over 1 million students in 2025-26, DiCerbo said. 

DiCerbo, who has worked in the ed tech sector for 20 years, said Khanmigo’s increase between 2023-24 and 2024-25 “was the biggest one-year jump that I have seen in terms of adoption of an education technology.”

Still, as Khanmigo and other AI tutoring tools proliferate, evidence is still scant as to whether they can actually improve student outcomes. 

Here are three key questions that ed tech researchers and industry experts say school and district leaders should consider. 

What kind of AI tutoring tools are available for students?

AI tutors beyond Khanmigo that specifically cater to K-12 include Amira, which uses a cartoon avatar as a reading tutor and provides step by step assistance to students while recording their sessions so teachers can review and evaluate problem areas. Other AI-powered tools EarlyBird, Bamboo Learning and Imagine Learning use similar kinds of assessment techniques.

Publicly available AI apps have also embraced tutoring options that anyone can use. 

For example, OpenAI in late July launched a “Study Mode” in Chat GPT that uses interactive prompts to ask questions and guide students through a problem rather than providing a quick answer. Google has introduced a rival mode for Gemini called “Guided Learning,” and Anthropic’s Claude now offers a tutoring tool for college students.  

The release of these new features comes as some teachers have expressed concerns over students using AI tools to cheat. And in fact, the Pew Research Center found earlier this year that students are increasingly using ChatGPT for their schoolwork, with 18% of surveyed teens saying it’s acceptable to tap into AI for assistance with essays. 

What are the problems?

Research has long backed tutoring — at least with humans — as an effective way to raise student achievement, said Robbie Torney, senior director of AI programs at Common Sense Media, a nonprofit research organization.  

The challenge, however, is that there is little to no evidence that generative AI-powered tutors can do the same, Torney said. 

Torney described AI tutoring as “one of those sort of holy grails in the K-12 space of like, ‘Well, what would it look like to be able to have a tool or have a program that could help students make progress?’” This, he said, is why there’s growing interest in these AI tutoring apps.

For Chris Agnew, director of the Generative AI for Education Hub at Stanford University, there is a strong sense of urgency to help educators understand the efficacy of AI tools through research. 

The lack of research is coinciding with both rapid development of AI tools and strong encouragement from the Trump administration for school districts to use AI in their classrooms, Agnew said. Currently, however, districts are “flying blind, because the best data they have is surveys” or anecdotes from teachers who either strongly support or reject using the tools, he said.

The GenAI for Education Hub plans to analyze the use of OpenAI’s Study Mode in schools, and also how teachers and administrators use ChatGPT. “This is really important, because most AI innovation currently is coming from the private sector,” said Agnew. 

Khan Academy has yet to test the efficacy of its own Khanmigo through a randomized control trial study due to the expense and other challenges in running such studies, DiCerbo said — though the nonprofit still plans to conduct that “gold standard” research.