AI Inequity Is Developing in Schools
A survey indicates that policies and training are lagging in systematically problematic ways.
- By
- September 11, 2026
- CBR - Artificial Intelligence
In June 2023, just months after the widespread release of ChatGPT, roughly 20 percent of school principals in the United States said teachers at their school were using generative artificial intelligence. Two years later, that was up to 90 percent.
But many schools haven’t built the framework needed to use AI in the classroom, finds research by Chicago Booth’s Christopher Campos and University of Rochester’s John Singleton. Their study reveals what they call “a new form of digital inequality in K–12 schools: not unequal access to AI tools, but unequal development of the policies, routines, training, and infrastructure that govern school-level use.”
This inequality revealed itself in a survey that the researchers sent to more than 70,000 public and private K–12 schools, which they assembled from two federal databases. Campos and Singleton emailed principals and offered compensation—a guaranteed payment that varied in amount across recipients as well as entry into a raffle for a $600 gift card—to complete a roughly 15-minute questionnaire between December 2025 and February 2026. The final sample included about 1,200 principals spanning traditional public, charter, and private schools across the country.
Because invitations were tied to individual schools, the researchers could merge each response with demographic and neighborhood data and reweight the sample to represent US schools nationally. And by randomizing the amount offered in return for participation, they were able to determine whether there was a difference in responses among principals who were more eager or less eager to answer the survey.
AI is largely used to boost productivity within existing educational structures, the study indicates. Eighty-eight percent of principals said AI “never" or “rarely” replaced direct instruction. They said that students used AI for homework (37 percent of schools surveyed), drafting essays (35 percent), brainstorming (31 percent), and building study guides (30 percent), while teachers leaned on it for administrative work, lesson planning, and grading. As for the respondents themselves, they used AI to off-load busywork and prepare for meetings.
Learning on the fly
A survey of US school principals finds that the share of teachers using AI rose sharply in just three years, but AI use outpaced any formal training in it. Both use and training were lower in more economically disadvantaged schools.
But central to the researchers’ findings is that the use of AI outpaced the scaffolding around it. Teacher adoption preceded formal training. A majority of principals (58 percent) reported having no written AI policy in place; most (76 percent) said their schools held no contract for AI-powered tutoring; and only a third had tapped pandemic-era funds for AI purchases. Overall, these “patterns point to broad but shallow integration that is only weakly embedded in instructional practices,” the researchers write.
That shallowness is not evenly distributed. Combining seven measures, such as written policy and teacher training, into a single “integration index,” Campos and Singleton documented two persistent gaps. First, schools serving more economically disadvantaged students scored lower on levels of integration than schools with more advantaged students. (This finding held whether disadvantage was measured by lunch subsidies, test scores, or neighborhood income.) Second, integration in charter and private schools trailed that in traditional public schools. (A 2001 study by the Department of Education, cited by Campos and Singleton, found a similar lag in computer and internet adoption among charter and private schools in the late 1990s.)
Several additional factors predicted greater AI integration within schools. Principals who believed AI improves learning represented the single strongest predictor, but also relevant were teachers who championed the technology, available budget, and the level of discretion principals had over investment in AI. But none of these effectively accounted for why charter, disadvantaged, and private schools fell behind on integration.
What did make a difference was a higher level of organizing around AI policy. The single most important variable they uncovered was scale, Campos and Singleton find. District size explained about a third of the disadvantage gap. Public schools tend to be embedded in large districts in a way charter and private schools are not.
This points to a difficult truth for policymakers: Interventions such as educating principals, nudging teachers, or opening budgets to allow more discretionary spending may raise the level of integration but won’t help close the gap between these different school types.
Narrowing the gap likely requires a system or district to adopt AI and supplement it with professional development and guided practices, “as opposed to teachers independently figuring things out,” says Campos. In separate, ongoing research, he and Singleton find that when districts adopt teacher-facing AI (built for educators rather than students) and supplement the tools with development and guidance, students learn more.
Whether addressing this AI gap will narrow educational-achievement disparity is a question the study does not address, but the researchers are pointed about their findings: “The evidence here shows that the conditions for unequal effects are already present.”
Christopher Campos and John Singleton, “AI Diffusion Gaps: Unequal Integration of AI Across K–12 Schools,” NBER working paper, June 2026.
More from Chicago Booth Review
More from Chicago Booth
Your Privacy
We want to demonstrate our commitment to your privacy. Please review Chicago Booth's privacy notice, which provides information explaining how and why we collect particular information when you visit our website.