In schools across the country over the last two years, there has been a strange pattern. Administrators met to talk about rules for a technology that most of them had only used briefly. Updates to syllabi were made quickly by teachers. Parents asked things that no one could quite answer. In the middle of it all, there wasn’t much information about what ChatGPT was doing to students’ learning, habits, or ability to think through problems.
The SCALE Initiative at Stanford and OpenAI are working together on new research to try to close that gap. The project, which is based at Stanford’s Accelerator for Learning, will look at how ChatGPT is used in K–12 classrooms across the country and around the world. It will look at everything from basic adoption patterns to whether using certain types of AI actually helps students do better in school. The idea behind it seems very simple, but there is a lot of work that goes into it.
Professor Susanna Loeb, faculty director of SCALE, framed the problem bluntly. AI tools are flooding K-12 classrooms, she noted, and education leaders are being asked to make consequential decisions in a data vacuum. That phrase — data vacuum — is doing a lot of work. Because it’s not just that the research has been slow. It’s that the decisions weren’t waiting for research. They were already being made.
What makes this study different from earlier attempts to measure AI’s role in education is partly structural. OpenAI will share data from real K-12 classrooms directly with Stanford researchers, giving them a window into actual usage patterns rather than self-reported surveys or small-scale trials. That matters. A lot of what we think we know about students and AI comes from asking students about students and AI, which has obvious limitations.
One earlier Stanford study from 2023, for instance, found that cheating rates among high schoolers had not measurably increased since ChatGPT’s launch — roughly 60 to 70 percent of students reported some form of cheating behavior, a number consistent with pre-ChatGPT surveys. It was a reassuring finding in some ways, though researchers were careful to note that the picture could shift as more students become familiar with the tools. The honest read was: don’t panic yet, but keep watching.
This new research goes deeper than cheating. The group wants to find out if certain features, like ChatGPT’s fairly new “study mode,” affect how well students learn new things or improve skills like self-control. Since those are harder to measure than test scores, they have likely been mostly ignored in public discussions about AI in schools.
This is also more important in the real world than some academic research is. For example, schools in Hong Kong at first didn’t want to use generative AI in the classroom, but then they changed their minds and started using it in a controlled way. This is a pattern that has been seen in school districts all over the United States. Policies were changed before there was enough evidence to back them up. I can see why, since things were moving so quickly, but it put teachers and students in a strange spot where the rules kept changing and no one knew why.
Being honest about what this study won’t solve is important. Even though there is a lot of solid data on how ChatGPT changes learning metrics, outcome studies don’t really answer the more difficult questions about creativity, dependence, and what it means to think through a problem without help. The same student who wrote their own essay and got the same grade as someone who used ChatGPT to write their essay is not the same student. There is still real disagreement about whether and how much that difference matters.
The work that Stanford and OpenAI are doing together is not the end of the story. It’s more like the start of an honest conversation—one that schools should have started having a long time ago, before all the tools were common and rules were set in stone based on assumptions that no one had tested.
