A lot of teachers first met artificial intelligence the same way they met previous initiatives—in a staff meeting, on a slide deck, and with a promise that it would make life easier. They heard promises of easy lesson planning, automated grading, faster feedback, and reduced paperwork.
In reality, the bell rang, students arrived, and the day unfolded the way it always does, with too much to do, too little time, and not enough support for the hardest parts of the work.
For years, policymakers and school leaders have tried to make teaching “more efficient,” and some early survey evidence does suggest that teachers who frequently use AI report meaningful time savings.
Here’s the problem: As AI becomes the latest tool sold as a fix in K-12 education, we’ve been stuck on the wrong question.
The question isn’t whether AI can save teachers’ time; it’s whether AI can make the job feel more manageable. Those are not the same thing and confusing them is one reason so many well-intentioned reforms fall flat. A tool might save minutes off grading and still leave teachers feeling overwhelmed, because the strain of teaching isn’t just about task volume but whether teachers feel capable of doing the work that matters most.
New research we conducted points to a more realistic answer. In a nationally representative survey of more than 400 K-12 teachers, we examined how teachers’ confidence using AI for instructional purposes relates to perceived workload, anxiety, and overall mental well-being.
Here’s what stood out: AI didn’t directly reduce teachers’ workload. There also wasn’t a straight line from “use an AI tool” to “feel better.” The relationship was indirect. Teachers who felt more confident using AI also tended to feel more capable in the parts of teaching that matter most, engaging students.
That matters because engagement has become one of the more taxing parts of the job. Students in post-pandemic classrooms exhibit more disengagement, more behavior challenges, and more mental health needs, often without proportionate supports. It is hard for teachers to teach when half the class is distracted by the technology at their fingertips and other students are dysregulated on a daily basis.
In our data, teacher well-being tracked closely with a small set of factors: confidence (including confidence using AI), perceived workload, and anxiety. Teacher demographics and the context in which they taught explained very little.
But the most useful story is in how the pieces connect. In our structural model, teachers with higher AI pedagogy self-efficacy—that is, confidence in selecting and integrating AI tools into instruction—also reported stronger self-efficacy for engaging students and managing instruction. Engagement-related self-efficacy was linked to lower perceived workload, and lower workload strongly predicted lower anxiety. And lower anxiety predicted better mental well-being. To put it simply, AI may matter when it helps teachers feel more capable, and that sense of capability can reduce strain, not because the work disappears but because teachers feel more in control of it.
One finding from our research is especially relevant for school leaders: Not all confidence worked the same way. Confidence engaging students had a clear relationship to perceived workload. Confidence around instructional strategies alone did not.
School leaders need to take this into account when developing and providing training for teachers. If AI is going to support teacher well-being, leaders should focus less on AI as a productivity tool and more on AI as a tool that helps teachers adapt, differentiate, and sustain student engagement in the classroom.
So, what does this mean for how school leaders should approach AI?
First, stop treating AI like a product rollout.
One-off workshops won’t build real confidence. Teachers need repeated, practice-based support for using tools for tasks like lesson planning, feedback, and differentiation. They need ongoing coaching and time to compare what works. Unlike most of the new products and initiatives teachers encounter, AI is here to stay and will likely shape education and society as much as the introduction of the internet has.
Second, make the rules clear.
Unclear or inconsistent guidance on how teachers can use AI can create anxiety. Teachers need to know what’s allowed, what’s discouraged, and where the ethical boundaries are—especially when it comes to student data, grading, and communications with families.
Third, protect teachers’ time.
AI should not become “one more thing” on teachers’ plates. And to the extent that AI can save teachers’ time, that shouldn’t be permission to give teachers even more to do.
Our data are a snapshot of one point in time and are based on teacher self-reports, so we are not claiming that AI will lead to better mental health. But our findings do suggest that AI is a tool that may help teachers feel more control in their professional lives.
But we can offer a practical takeaway: Teacher well-being is closely tied to whether teachers feel effective at the work that matters most. Confidence using AI may strengthen that effectiveness in ways that reduce workload strain and anxiety.
AI is not going to fix all the challenges teachers face, but it holds potential to make the job more sustainable. However, if schools want teachers to get the most out of the technology, invest in real training, provide clear guardrails, and protect the time it gives back. Used well, AI can help teachers feel more capable at engaging students. And when teachers feel capable, they are more likely to stay. That’s how we make teaching a profession people will still choose 10 years from now.
