AI in K-12 Classrooms: What a National Survey of U.S. Teachers Tells Us About Who’s Using AI and Who’s Not

By /

Most of the research I cover on this blog comes from controlled experiments, lab studies, or small qualitative projects. They’re valuable, but they rarely tell us what’s happening across an entire education system. The RAND Corporation’s report on AI use in K-12 classrooms fills that gap.

Diliberti, Schwartz, Doan, Shapiro, Rainey, and Lake (2024) surveyed more than 1,000 teachers and over 200 school districts in fall 2023, then followed up with interviews with district leaders. The result is one of the first national snapshots of how American teachers were engaging with generative AI at that point.

I should say upfront: fall 2023 already feels like a different era in AI terms. The tools have changed, adoption has accelerated, and the policy conversation has matured considerably since then. But this report still matters because it gives us a baseline. It tells us where K-12 education stood roughly a year after ChatGPT’s release, and many of the patterns it documents, especially around equity and training, are almost certainly still with us.

The numbers were modest. As of fall 2023, only 18 percent of teachers reported using AI tools in their teaching. Another 15 percent had tried them at least once. That left a clear majority who hadn’t touched AI in any professional capacity. But the trajectory was already pointing upward. “Nearly three-quarters (73 percent) of current AI-using teachers said that they expect to use AI products and tools more next school year (2024-2025) than they do this school year (2023-2024)” (p. 5). Even some nonusers anticipated getting started. The direction was unmistakable.

I’d bet the 2025 numbers look very different. An updated version of this RAND survey, one that reflects where we are in 2026, would be incredibly valuable. The field needs current data.

Who Was Using AI and Why It Matters

The adoption patterns are worth paying attention to, because they likely haven’t changed as much as the overall numbers. English language arts and social studies teachers were more likely to use AI than those in elementary or STEM roles. The authors suggest a practical explanation: these teachers already build and adapt their own materials regularly. If you’re someone who customizes lesson plans, adjusts reading levels, and creates your own assessments, AI fits naturally as a drafting partner. The workflow already exists. AI accelerates it.

I think there’s something deeper here too. Teachers who routinely create original content are already comfortable with iterative work. They draft, revise, and adapt constantly. That mindset aligns well with how AI works best in education: as a starting point, not a finished product. Several teachers in the study described using AI exactly that way, to generate a first draft or brainstorm ideas, then refining everything themselves.

That pattern echoes what Cheng et al. (2025) found in their study on AI and writing performance. Students who brought their own questions and purposes to the AI interaction got significantly better results. Agency shaped the outcome. The same logic applies to teachers. Those who approach AI with clear instructional goals and a willingness to revise get more out of it than those who expect a polished product from a single prompt. That was true in 2023, and I suspect it’s even more true now that the tools are more capable and the temptation to accept AI output at face value is stronger.

AI in K-12 Classrooms

Districts Were Leaning Toward Support, Not Restriction

One of the more encouraging findings in the report was the orientation of district leaders. “These interviewees were more focused on how to use AI well rather than on how to restrict or block its use” (p. 11). Given how many districts had moved toward outright bans in early 2023, that shift was significant by fall of that year. The panic was already subsiding for many, replaced by a growing interest in training, guidance, and responsible experimentation.

By the end of the 2023-2024 school year, 60 percent of districts either had provided or planned to provide teacher training on AI. That sounded promising at the time. But the details mattered enormously then and still do now. What kind of training? How deep? A one-hour webinar on ChatGPT prompts is very different from sustained professional development that builds pedagogical judgment around AI use.

I wrote about this problem when covering Bilbao-Eraña and Arroyo-Sagasta’s (2025) study on AI literacy for pre-service teachers. Their 8-hour intervention improved awareness and attitudes but didn’t build trust. Trust requires deeper engagement with ethics, reliability, and governance. Choi, Jang, and Kim (2023) found something similar: ease of use was the strongest predictor of whether teachers would adopt AI tools, and trust needed deliberate, sustained attention. If district training stays at the surface level, and there’s no reason to think that problem has disappeared since 2023, adoption numbers may grow without the pedagogical depth to support them.

The Equity Gap in the Data

Here’s the part of the report that should worry us most, and the part I suspect has changed the least. The data showed that historically advantaged districts were far more likely to offer AI training:

“Assuming districts’ current plans come to pass, by the end of the 2023-2024 school year, 65 percent of majority-White districts will have provided training compared with only 39 percent of districts serving mostly students of color” (p. 11).

A 26-percentage-point gap in training access. And it mapped onto another troubling pattern. Teachers in high-poverty schools were more likely to use AI to generate lesson plans. On the surface, that sounds practical. These teachers are often stretched thin, and AI can help with workload. But if the materials AI produces aren’t high quality, and if the teachers using them haven’t received training to evaluate and refine AI output, we end up with a situation where the most vulnerable students receive the least vetted instructional content.

Perkins and Roe (2025) raised a parallel concern in their chapter on the future of assessment. They argued that the idea of AI as a universal equalizer is “more myth than reality” and that infrastructure gaps create fundamentally different AI experiences depending on context. The RAND data from 2023 put numbers behind that argument. AI in well-resourced districts looked like supported experimentation with guardrails. AI in under-resourced districts risked becoming unguided reliance on machine-generated content. Same technology. Very different conditions around it.

Has that gap closed in two years? Maybe partially. But equity gaps in education tend to be structural, and structural problems don’t fix themselves through technology adoption alone. A 2026 version of this survey would tell us whether the training gap has narrowed or widened. I genuinely hope RAND is working on it.

The authors were clear about what needed to happen: “Further research on the quality of AI-generated classroom content is an essential next step in verifying its use in classroom settings” (p. 14). That call was urgent in 2024. With AI tools now significantly more powerful and more widely adopted, it’s even more pressing today.

What This Report Still Tells Us

The RAND report confirms something that runs through virtually all the research I cover: the technology matters far less than the pedagogy around it. AI arrived in K-12 classrooms. Adoption grew. Districts began to offer training. Policy started forming.

Two years later, all of that has accelerated. But the underlying challenges this report identified, uneven training, equity gaps, surface-level professional development, the lack of quality research on AI-generated classroom materials, those don’t resolve on their own just because the tools get better.

The teachers in this survey who used AI well shared a common profile: they already created their own materials, they treated AI output as a starting point, and they revised what it gave them. That’s exactly the kind of intentional, pedagogically grounded approach that produces real value. And it’s the kind of practice that needs ongoing institutional support, not a single training session in 2023 and nothing after.

If you haven’t tried AI yet, this report should encourage you. Eighteen percent was where things stood more than two years ago. The numbers have grown since then, and the teachers who started building their own approach to AI early, critically and with their students’ learning at the center, are almost certainly better positioned now. The tools will keep evolving. The need for thoughtful, informed teachers won’t.

Reference

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top