As AI use in education grows more popular, there is a strong desire among some education leaders and researchers to understand what that usage looks like. A recent Gallup poll, for example, found that 6 in 10 teachers reported using AI for their work. Rather than relying on self-reported data to answer this question, the Stanford study is based on usage logs from the AI platform SchoolAI.
“We all know that humans are flawed at reporting our own behavior accurately,” Chris Agnew, director of Stanford’s Generative AI for Education Hub, who worked on the project, said.
The analysis tracked 9,000 U.S. teachers who first joined SchoolAI between Aug. 1 and Sept. 15, 2024, and collected data on their usage for 90 days. In that time, not all teachers used AI continuously. According to a report released this month, 16 percent used the platform only once, 43 percent were short-term users and 41 percent became regular users, logging in between eight and 49 days out of the total 90. Only 1 percent, the “power users,” logged into the platform on 50 or more days.
Agnew said that over 40 percent of participants becoming regular or power users is slightly higher than typical software adoption rates. According to software experience management platform Pendo, software keeps about 30 percent of its users after three months.
The findings suggest teachers are using AI on an as-needed basis rather than incorporating it into daily or weekly workflows. In any given week, about a third of participants used the platform. The lack of a sharp drop-off or spike shows variation in who used the platform from week to week.
Teachers not only differed in frequency of use but also in the tools they favored. SchoolAI offers student-facing chatbots, teacher productivity tools like lesson generators, grading aids and quiz builders, and teacher chatbot assistants. While lighter users often spent significant time with student-facing chatbots, the teachers who used the platform more regularly shifted more and more of their time toward teacher support features. Power users bypassed most student tools from the start, spending over 80 percent of their time on teacher productivity tools and chatbots.
Agnew said the trend reflects a “human in the loop” approach to AI in education.
“The teacher [can] take the output of the AI and then filter it with all their depth of experience to then determine, how does this inform their practice, their classroom,” Agnew said. “Versus building an AI tool that goes direct to students, goes direct to a younger brain — somebody who's not an adult, somebody who is building their expertise and judgment, building their skill.”
The data also showed a clear pattern in when teachers use AI. AI’s time-saving abilities in grading and lesson planning might suggest that teachers would use them after hours, when those tasks usually take place, Agnew said. However, the data showed that most teachers used AI tools on weekday mornings. While the survey did not collect information about why that might be, Agnew said it could point to teachers using AI to brainstorm ideas or prepare materials before class.
“Myself, being a former teacher, what I could imagine is, this is a partner for them in their teaching practice as they are orienting themselves to the school day,” he said.
The researchers note that tracking active days doesn’t measure the impact of the work done with AI. The next phase of the project will analyze the content of teacher-AI interactions and explore how students engage with the platform, which may follow different patterns.
