The Great AI Teaching Experiment:
We Know, What We Don't, and What We Should Do While We Figure It Out
Image generated by AI using ChatGPT with DALL·E.
Teachers are saving six weeks a year with AI tools. Sounds amazing, right? Well, maybe. Here's what a new study tells us—and what it definitely doesn't.
A comprehensive new study from Gallup and the Walton Family Foundation surveyed over 2,000 K-12 public school teachers about their AI use, and the headline finding is striking: teachers using AI weekly save an average of 5.9 hours per week—equivalent to six weeks per school year. That's what researchers are calling the "AI dividend," and it represents time that teachers can reinvest in providing better student feedback, creating more individualized lessons, and yes, getting home to their families at a reasonable hour.
But before we declare victory in the battle against teacher burnout, we need to ask some harder questions about what these findings really tell us—and what they don't.
The Good News First: The AI Dividend is Real (At Least for Some Teachers)
Let's start with what the study gets right. The data shows that 60% of teachers have already used AI tools this school year, with particularly high adoption among high school teachers (66%) and early-career educators (69%). Those who dive in aren't just dabbling—they're seeing real benefits.
The teachers using AI most frequently report significant time savings across multiple tasks: making worksheets and assessments, handling administrative work, and preparing lessons. More encouraging still, they're not just working faster—64% report higher quality in the materials they modify for students, 61% generate better insights about student learning data, and 57% say AI improves their grading and feedback quality.
Listen to how some teachers describe the impact:
"It gives me back the time I need to make lessons more engaging and better aligned with standards."
"I spend the extra time on more meaningful feedback for my students."
"It also allows me to get my work done during my contracted time, so I am not having to stay late after work."
These aren't abstract productivity gains—they're descriptions of professionals reclaiming both their craft and their lives.
But Hold Your Horses: Why We Should Take These Results with a Grain of Salt
Here's where we need to pump the brakes. While the study offers valuable insights, it suffers from some significant limitations that should make us cautious about drawing broad conclusions.
The Self-Selection Problem: The teachers reporting these dramatic benefits are, by definition, the ones who chose to adopt AI tools and use them regularly. Of course they're enthusiastic—they're the early adopters who were already motivated to experiment with new technology. The 40% of teachers who haven't used AI at all might have very different experiences if they tried, or very good reasons for their reluctance.
The "Trust Me" Data: All those impressive time savings? They're based entirely on teachers' self-reported estimates. Anyone who's ever asked someone how long their commute takes knows how unreliable such estimates can be. We don't have objective measurements of actual time spent or independent verification of quality improvements.
Missing the Most Important Question: Here's the big one—we have zero evidence that students are actually learning better. Teachers report feeling more efficient and producing higher-quality materials, but the study provides no data on student engagement, achievement, or outcomes. A teacher might save hours on lesson planning, but if students are less engaged or learning less effectively, have we really improved education?
This isn't to say the study is wrong—it's to say it's incomplete in crucial ways.
The Bigger Picture We're Missing: What Happens When the Honeymoon Phase Ends?
We're clearly in the "shiny new toy" phase of AI adoption in education. The teachers using these tools most enthusiastically are likely experiencing the natural excitement that comes with discovering something that genuinely helps with long-standing frustrations. But what happens when the novelty wears off?
Consider some longer-term questions the study doesn't address:
Skill atrophy: If AI handles more routine tasks, do teachers lose important pedagogical muscles they might need later?
Over-dependence: What happens when the AI tools fail, change dramatically, or become unavailable?
Academic integrity challenges: How do we maintain authentic assessment and learning when both teachers and students have access to increasingly sophisticated AI assistance?
The equity gap: What about the teachers and schools that don't have access to these tools or training in how to use them effectively?
The study notes that 68% of teachers haven't received any AI training from their schools, and only 19% work at schools with AI policies. This suggests we're in a wild-west phase where individual teachers are figuring things out on their own—hardly a recipe for systematic improvement.
The "Wait and See" Dilemma: We Can't Afford to Just Pause
Here's our challenge: we can't conduct a five-year randomized controlled trial while students are graduating every year. The train has left the station—AI tools are already in classrooms, and they're not going away. We need frameworks for responsible adoption now, even while we're still gathering evidence about long-term impacts.
This creates a delicate balance. We can't rush headlong into AI adoption based on preliminary, self-reported benefits. But we also can't afford to ignore tools that might genuinely help overworked teachers and underserved students.
How This Study Fits in the Research Landscape
Before outlining a path forward, it's worth examining how these Gallup findings align with—or diverge from—other recent research on AI in education. The picture that emerges is fascinatingly contradictory.
Where there's corroboration: The study's adoption rates align roughly with other recent surveys. RAND found that 25% of teachers used AI for instructional planning in 2023-24, while the Gallup study reports 60% have used AI "for their work" more broadly in 2024-25. Other surveys suggest that 50% of teachers now use AI for lesson planning, and EdWeek Research found that 43% of teachers had received AI training by fall 2024—all suggesting rapid adoption acceleration.
The reported benefits also find some support. Carnegie Learning found that 42% of AI-using teachers reported reduced administrative task time, with 25% seeing benefits in personalized learning. These efficiency gains mirror the Gallup study's time-saving claims.
Where there's contradiction: But here's where things get complicated. Pew Research, surveying over 2,500 teachers in fall 2023, found strikingly different attitudes: only 6% of teachers thought AI does more good than harm in K-12 education, while 25% said it does more harm than good. Even more telling, high school teachers—who the Gallup study identifies as the heaviest AI users—were the most negative, with 35% saying AI tools do more harm than good.
This creates an apparent paradox: how can the same group of teachers be both the biggest users and the biggest skeptics? The answer may lie in timing. Fall 2023 is ancient history in AI development terms—ChatGPT had been publicly available for less than a year, and the tools accessible to teachers were far more primitive than today's offerings. We may be looking at teachers who were initially skeptical of limited tools but have since discovered more sophisticated applications.
But there's a deeper problem here: we keep citing these outdated studies as if they describe current reality. The result is a discourse built on evidence that may no longer be relevant—what we might call "research lag bias." By the time rigorous studies are completed and published in a field moving this quickly, they're often describing a technological landscape that no longer exists. Yet these become the "authoritative" sources that shape policy and practice.
The student outcomes question: Most tellingly, while some meta-analyses suggest AI can improve student learning outcomes—one recent study found "a significant positive effect of AI on students' academic performance, with an effect size of 0.924"—these studies typically examine controlled AI tutoring systems, not the general classroom AI tool use that Gallup measured.
The disconnect between teacher efficiency gains and student outcome evidence remains the crucial gap across all studies. We have mounting evidence that AI can save teachers time and effort, but far less evidence that this translates to better student learning.
A Practical Path Forward While the Jury's Still Out
Given this mixed research landscape, what should we actually do? Here's a framework for moving forward thoughtfully:
For School Leaders:
Develop AI policies immediately. Only 19% of schools have them, which is unacceptable given the current level of adoption.
Invest in proper teacher training. Don't leave educators to figure this out alone.
Focus on transparency and ethical guidelines from the start, rather than trying to retrofit them later.
Demand better research that includes student outcome measures, not just teacher satisfaction surveys.
For Teachers:
Experiment thoughtfully, not blindly. Try AI tools, but track your own results honestly—both time spent and student engagement.
Share what works AND what doesn't with colleagues. The field needs honest accounts of both successes and failures.
Maintain your core pedagogical skills even while using AI assistance. Don't let the tools replace critical thinking about teaching and learning.
For Everyone:
Push for equity in AI access and training. The benefits shouldn't accrue only to well-resourced schools and tech-savvy teachers.
Keep the focus on student learning, not just teacher efficiency. Time savings mean nothing if students aren't learning more effectively.
Demand longitudinal research that follows students and teachers over time, measuring real educational outcomes rather than just satisfaction scores.
Six Weeks Saved Means Nothing if Students Don't Learn More
The Gallup study gives us a tantalizing glimpse of AI's potential to address one of education's most persistent problems: teacher workload and burnout. The idea that technology might finally help teachers reclaim time for the human parts of their job—building relationships, providing thoughtful feedback, inspiring curiosity—is genuinely exciting.
But we can't let enthusiasm outpace evidence. The question isn't whether AI can save teachers time—this study suggests it can, at least for some. The question is whether it can help students learn better, think more deeply, and prepare more effectively for the complex challenges they'll face.
That's the question we still need to answer. Until we do, let's proceed with both optimism and caution, experimenting responsibly while demanding better evidence. Our students deserve nothing less than our most thoughtful approach to the tools that might shape their educational future.
After all, the real measure of any educational innovation isn't how much time it saves teachers—it's how much more our students learn, grow, and flourish. On that crucial question, the jury is still very much out.
Who is doing the training when we don’t even know what works, whether AI helps with student learning, and the models and tools are transforming so quickly?
You're right that we don't have all the answers. But we do know some things: which tasks AI handles well, which ones it doesn't, and how to craft useful prompts. Teachers who've been experimenting have practical insights worth sharing.
The training doesn't need to be "here's the definitive approach." It can be "here's what we've tried and what we've learned so far."
The alternative to learning together isn't waiting for perfect knowledge, it's letting adoption happen randomly while some schools move ahead and others fall behind. We can acknowledge uncertainty while still sharing what we're discovering.