Overworked and Understaffed: Special Ed Teachers Turn to AI for Help

Editor’s note: NPR uses only the first names of minors in this story because it discusses their learning disabilities and placement in special education.

BAY POINT, Calif. — The sun would just be rising when teacher Mary Acebu began her days. She’d blast music on the way to work to get energized and get to her classroom by 6:30 to prepare for her students’ arrival at 8. Often, it’d be dark by the time she headed home, sometimes with paperwork in tow.

Like so many special education teachers around the country, this was Acebu’s life for much of the 10 years she’s been teaching at Riverview Middle School, in this small, unincorporated northern California town.

“I don’t do that anymore,” she says with a laugh.

That’s because Acebu has been experimenting with artificial intelligence for the last two years to get through paperwork more quickly and says it’s helped her instead use precious time for student interaction. “I have time to talk to the kiddos and really build those relationships,” she says, “instead of sitting here in front of my computer.”

For years, schools nationwide have struggled with hiring and retaining special educators. In the 2024-25 school year, 45 states reported special education teacher shortages, and staff turnover is worse in schools that largely serve low-income students, like Riverview.

Some special educators say part of what makes them feel overworked is legally required paperwork layered on top of regular teaching duties. Acebu is one of a growing number of those teachers around the nation using AI to help speed up that paperwork — including for writing individualized education programs (IEPs). Educators and families maintain these detailed documents that outline goals and services students need to meet those goals at school.

According to a recent survey by the nonpartisan Center for Democracy and Technology (CDT), 57% of special education teachers polled nationwide said they used AI to help develop individualized plans for their students in the 2024-25 school year. That’s up from 39% the previous school year.

Along with the survey results, the CDT warned of privacy, legal and ethical risks around using AI. Other research, however, including from the University of Virginia (UVA) and the University of Central Florida (UCF), has shown that when used appropriately, AI can help special education teachers craft IEPs of equal or higher quality than when teachers produce them alone.

And the time saved can benefit students, too. “The more face time a student with a disability has with a teacher, that often yields better outcomes for them, both educationally, functionally — just across the board,” says Olivia Coleman, a researcher and professor at UCF who has been studying the role of AI in special education.

Acebu says that rings true in her classroom. She points out King, one of her eighth graders, as an example. “He was a non-reader, beginning of seventh grade. He’s reading now.” That, for Acebu, is the point of IEPs — to put what’s on paper into practice for her students. She says that is only possible with intentional, hands-on work in the classroom.

What IEPs are and why they matter

Every seventh and eighth grader in Mary Acebu’s class learns differently — some work independently, some in pairs, others with headphones on and yet others with speech-to-text technology. Those differences are captured in each child’s IEP, a document required by federal law for each of the over 8 million students with disabilities in this country.

Every IEP includes annual goals tailored to each student’s present needs, but importantly, “also where you want them to go within the next year,” says Danielle Waterfield, Coleman’s research partner at UVA.

Both Coleman and Waterfield say while many teachers report feeling bogged down by the work that goes into developing IEPs, teachers also recognize they are a necessary tool for students with disabilities to get a quality education.

Acebu says that to develop those goals, teachers must know each student’s learning style intimately. “The key term is ‘individualized.’ No two kids are the same,” she says. For special educators, the process involves hours of meetings and a deep knowledge of complex education law and policy.

It used to take Acebu around 45 minutes to develop three or four IEP goals per student. She points to a big, blue binder at least 5 inches thick on her bookshelf that contains California’s education standards. “It used to be flipping through all those pages,” to find the right standard to match unique student goals, she says.

Then came AI.

Using AI — with a ‘human touch’ 

A couple of years ago, Acebu began taking courses on how to safely and effectively use AI. Around the same time, her district, Mt. Diablo Unified, entered agreements with companies that offer education-focused AI tools including MagicSchool AI and Google. They promise to protect sensitive student data, a primary concern for those who warn against the risks of using AI in schools. A growing number of districts are adopting such products, though only a few states have official AI education policies.

Recently, using a district-vetted tool, Acebu customized chatbots for her school and trained them on state standards, assessments and other special education data. She now uses her “little assistants” for a wide range of tasks, from creating personalized worksheets to developing IEP goals.

And then, she says, “you’re double-checking everything. Like you have to put that human touch, that’s the final step.”

In their research, Coleman and Waterfield found special education teachers nationwide are using AI to help write IEP goals, track student progress, synthesize data and create differentiated learning materials, among other things.

Acebu is uniquely equipped to use tech-tools: She just earned her doctorate in instructional technology and is on her district’s AI task force, which is developing an official AI policy.

Some of Acebu’s less tech-savvy colleagues, however, were skeptical, including Paul Stone, who has been a special educator at Riverview for 22 years.

Then the number of students he serves shot up.

“I don’t want to say it’s killing me, but it has put a huge stressor on my mental health and my life,” Stone says of his work this year. “It would be kind of nice if there were two jobs, like one paperwork job and one working with the kids.”

So, a few weeks ago, after a tutorial from Acebu, he gave her chatbot a shot. He was surprised by the results.

“It’s an amazing time-saver so far,” he says. Stone has used AI for a number of things including producing simple summaries of complicated data to present to parents at IEP meetings. “I mean, it’s not like ‘that’s it, I’m done.’ I still have to go through and check it all.”

He and Acebu both say it could help them, and other educators, avoid burnout. Yet, Ariana Aboulafia, who was the lead author of CDT’s report, calls AI tools “a Band-Aid” for special education teachers who feel overworked.

Using AI in special education — with guardrails

Band-Aid or not, more teachers are using AI around the country. There are a litany of concerns about its use, especially in special education, which is highly regulated. “Student privacy is number one,” says Acebu. “Don’t put information there that’s gonna identify your students.” CDT’s Aboulafia adds that while the risks around privacy may be reduced if a school is using a vetted vendor, data breaches could still make that information vulnerable.

But not all teachers are using district-approved tools. Coleman, Waterfield and CDT’s research all found that educators around the country are using AI both formally and informally — from free consumer platforms like ChatGPT and Claude to district-approved tools like MagicSchool AI, Google Gemini and Playground IEP, among others. To help teachers navigate this complicated landscape, Waterfield and Coleman developed a “decision tree” for ethical AI use.

Another consideration is the fact that AI models can be biased, including against people with disabilities, says Aboulafia, who leads the Disability Rights in Technology Policy Project at CDT. In addition, she worries AI models built on pattern recognition are, “to a certain extent, inherently incompatible with a process that legally requires individualization.”

Aboulafia is most concerned about the 15% of teachers CDT’s survey found have been relying entirely on AI to develop IEPs. There must always be a “human in the loop,” she says.

Acebu, who happens to be her district’s teacher of the year, says these days, she comes to class just 30 minutes before her students, and leaves just after the last bell. This has improved her work-life balance and the quality of her teaching.

King, the eighth grader in her class who has evolved into a confident reader, also goes to math class now without any additional support.

“That’s the dream of every special educator,” she says, beaming. “But guess what? That takes a lot of hard work.”

AI tools, Acebu says, have given her more time for that kind of hard work.

Edited by: Nirvi Shah
Visual design and development by: LA Johnson