An IEP is supposed to be a portrait of one specific child — every goal, every accommodation, every service reflecting careful observation of their strengths, their challenges, and what they need to move forward. The paperwork that comes with building that portrait is substantial.
An NPR report from May 20th draws on a survey from the Center for Democracy and Technology that found 57% of special education teachers used AI to help develop IEPs in the 2024–25 school year, up from 39% the year before. And 15% said they relied on AI entirely — no teacher input, just AI output.
Why it’s happening — and why it makes sense
What special education teachers are dealing with is worth taking seriously before passing judgment.
Writing an IEP isn’t a 20-minute task. It can take hours — hours spent documenting present levels of performance, writing measurable annual goals, cataloguing services, getting sign-offs, attending meetings, revising, and translating everything into language parents can understand. And that’s on top of teaching. Special education teachers are already among the most overworked educators in any school, often managing caseloads far beyond what’s reasonable.
The CDT survey found that many teachers said AI helped them spend less time on paperwork and more time actually teaching. Time spent on compliance documentation is time not spent with students.
So when AI makes the documentation faster, it’s not surprising that teachers use it. What is surprising is how fast adoption has moved. Going from 39% to 57% in a single school year is a steep climb. Either the tools got better, the pressure got worse, or both.
The tension inside the document
The CDT report noted that AI models, which work by recognizing patterns in existing text, are “to a certain extent, inherently incompatible” with a process that legally requires individualization. An AI trained on thousands of IEPs might produce a document that looks right — correct sections, appropriate terminology, and goals that sound measurable — but reflects the average student rather than this one.
IEPs have legal weight. They’re binding agreements between families and schools. If the goals in a student’s IEP were generated by a language model making educated guesses, and those goals turn out to be wrong for that kid, that’s not a documentation problem. That’s a services problem. It affects what a student gets, for how long, and what it looks like when things aren’t working.
There’s also the privacy question. The CDT report flagged concerns about teachers using consumer-grade AI tools — ChatGPT, Claude, others — to draft documents that contain sensitive student information. Many of these tools weren’t built for FERPA compliance. Some districts have approved tools like MagicSchool AI or Playground IEP that are designed with that in mind, but many teachers are using whatever’s accessible.
This isn’t just a special education story
I work in a high school classroom, not in special education. But the pattern here is familiar.
Documentation that starts with a legitimate purpose can quietly become its own burden. Progress notes, behavior logs, learning objective write-ups, and rubric justifications — the list of things teachers are asked to produce has expanded significantly over the past decade, and AI has made most of it faster.
The problem is that some of that documentation isn’t just administrative. Some of it is the professional judgment made visible. When I write feedback on a student essay, I’m not filling in a template — I’m recording what I actually noticed about how that student is thinking. If I hand that off to AI and clean up the output, I’ve produced something that looks like professional judgment but isn’t. The student still gets feedback. But it’s feedback based on what the average student does with that type of prompt, not what this student did with it.
The IEP story makes that problem concrete. When 15% of special education teachers are handing the whole document to AI, they’re not being lazy. They’re almost certainly exhausted, behind, and trying to keep up with a system that demands more paperwork than any one person can produce. The problem is the conditions — not the teachers.
But the solution can’t be to generate better-looking AI output and call it individualization.
What reasonable looks like
There’s a version of AI-assisted documentation that works. It’s the version where AI handles structural scaffolding — formatting, reminders of required sections, flagging missing elements, and helping translate jargon into plain language — while the teacher fills in the substance.
The difference: AI generates a list of possible goals for a student with a reading processing difficulty. The teacher reviews, cuts what doesn’t fit, adjusts what’s close, adds what the AI missed, and writes the present-level narrative from their own observations. That’s faster than starting from scratch. The judgment is still the teacher’s.
Districts using purpose-built tools are starting to figure this out. Playground IEP and similar platforms are designed to assist within the IEP workflow rather than replace it — flagging compliance gaps, offering goal banks, and making the logistical work lighter. That’s a different use case than pasting student details into a consumer chatbot and pressing generate.
The CDT report isn’t arguing that teachers shouldn’t use AI for IEPs. It’s arguing that 15% going all the way — no teacher input, just AI output — is a threshold worth paying attention to.
Takeaway for Teachers
If you use AI for any documentation that’s supposed to reflect your professional judgment about a specific student — IEPs, learning plans, and formal feedback — take a few minutes this week to look at what you’ve produced recently. Ask yourself: does this document describe this kid, or does it describe a version of this kid that AI inferred from general patterns?
If it’s mostly the second thing, that doesn’t mean you did something wrong. It probably means you were overwhelmed and needed a lifeline. But it’s a useful distinction to make — and the line to draw is AI as scaffold, not AI as substitute.
The paperwork isn’t going to get lighter. The professional judgment behind it still has to come from you.
If you’re thinking through questions like this one — where AI helps and where it substitutes for something that should stay human — my book goes deeper. The AI Doesn’t Know Your Students is available on Amazon and at shouldiuse.ai/book.
David Jacobson is a high school history teacher. He writes about AI, education, and the messy intersection of the two at shouldiuse.ai.
