AI for Special Education & Co-Teaching: IEP Workflows, Real-Time Accommodations, and Differentiation

This intersection — AI, special education, and co-teaching — has no established content territory. You can find resources on AI in education. You can find resources on co-teaching models. You can find resources on special education law and IEP processes. What you cannot find is a sustained treatment of how these three domains connect, why that connection matters, and what practitioners should do about it.

That gap is significant because the populations most affected by AI’s adoption in schools are the populations least represented in the conversation about it. Students with disabilities. English learners. The teachers who serve them — special education teachers managing caseloads of 20+ IEPs, co-teachers navigating shared classroom space with shared responsibility for student growth, instructional coaches trying to support both.

This guide addresses that gap directly. It draws on research on co-teaching outcomes, the legal framework for special education, and the emerging practice of using AI copilots to support — not replace — the professional judgment of special education teachers and co-teachers.

The Intersection That Nobody Covers

Three trends are converging, and the convergence point is where the need is greatest.

Growing IEP populations. The percentage of students receiving special education services under the Individuals with Disabilities Education Act has risen steadily. The National Center for Education Statistics reports that the number of students ages 3-21 receiving IDEA services increased from 6.4 million in 2010 to over 7.5 million in recent years. Each IEP represents not just a legal document but a workflow: referral, assessment, eligibility, goal writing, service delivery, progress monitoring, annual review.

Teacher shortage. Special education has been among the hardest-hit areas of the nationwide teacher shortage. Districts struggle to fill SPED positions, and the teachers who remain carry larger caseloads with less support. The structural pressure means that anything that can improve efficiency without compromising quality is not a luxury — it is a necessity.

Legal requirements for FAPE in LRE. The Individuals with Disabilities Education Act guarantees a free appropriate public education in the least restrictive environment. Co-teaching is a primary vehicle for delivering FAPE in LRE because it places students with disabilities in general education classrooms with both a general education teacher and a special education teacher. Research supports this placement: co-teaching yields faster growth than self-contained settings (Karge 2023b, Dieker 2007), and students with disabilities in general education settings show fewer absences and higher test scores (Anderson 2021, Cole et al 2021). Co-teaching greatly enhances the success of students with disabilities (Dieker 2007). Co-teaching is a step toward LRE.

The intersection is not theoretical. It is where the daily work happens. A co-teaching pair needs to plan together, deliver instruction together, monitor progress together, and adjust in real time. AI can support each part of that cycle — but only if the implementation is grounded in the specific needs of special education and the specific dynamics of co-teaching.

AI for IEP Workflow

IEP writing is one of the most time-intensive tasks special education teachers perform. It is also one of the most formulaic — many sections follow patterns that can be scaffolded with AI assistance while still requiring professional judgment for the individualized components.

AI copilots can support:

• IEP drafting — Generating draft language for present levels of performance, goal frameworks, and accommodation lists based on teacher input. The AI produces a starting point; the teacher individualizes based on their knowledge of the student.
• Goal writing — AI can generate multiple versions of a goal aligned to a specific standard, at different levels of complexity, with different measurement criteria. The teacher selects and modifies rather than writing from a blank page.
• Progress monitoring — AI can help organize and summarize progress monitoring data, identifying trends in student performance toward IEP goals and flagging concerns that may require goal revision.
• Compliance documentation — AI can check draft IEPs against common compliance requirements — are all required components present, are goals measurable, are accommodations specific — before the document goes to the team for review.

A specific workflow: IEP essay feedback. This workflow, drawn from the co-teaching and education (CTE) book, is straightforward and high-impact:

1. Paste a student’s essay into ChatGPT for grammar and structure analysis
2. Review the AI-generated feedback for accuracy
3. Add your own content-specific notes — comments on argument quality, evidence use, content understanding
4. Combine the AI-generated mechanical feedback with your teacher-generated content feedback into a comprehensive response to the student

This workflow works because it separates mechanical feedback (grammar, syntax, organization) from content feedback (argument, evidence, understanding). AI handles the former efficiently; the teacher handles the latter with the nuance that only a content expert can provide.

For a deeper treatment of AI in the IEP workflow, see Amplifying Special Education Teachers’ IEP Workflow with AI Tools.

Critical caveat: AI-generated IEP content must always be reviewed by the special education teacher. AI can draft, but it cannot individualize in the way that IDEA requires. Every IEP must reflect the unique needs of the individual student, and that determination requires professional judgment that no AI tool can provide. The teacher is the decision-maker. The AI is a drafting tool.

Real-Time Accommodations in Co-Taught Classrooms

One of the most powerful applications of AI in a co-taught classroom is the ability to provide accommodations and modifications in real time, as referenced in students’ IEPs. Co-Teaching Evolved (Rhoads, 2025) notes that “teachers can see student progress and performance in real time, they can use each student’s accommodations” (p.131).

What this means in practice:

• During parallel teaching (two groups, same content, different approaches), AI can generate differentiated materials for each group on the fly. The general education teacher leads grade-level content; the special education teacher delivers the same content with accommodations — reduced complexity, visual supports, scaffolded language — that AI helped prepare.
• During alternative teaching (one teacher works with a small group while the other manages the rest of the class), AI can generate practice sets, supplementary materials, or re-teaching resources for the small group, customized to their specific skill gaps.

The key insight is that real-time accommodation delivery requires preparation. AI does not generate content spontaneously during instruction; it generates content that teachers prepare before instruction, using AI to reduce the time that preparation takes. A co-teaching pair that can generate differentiated materials in 15 minutes rather than an hour has 45 more minutes for co-planning — which research identifies as the most critical factor in co-teaching effectiveness.

Both teachers must be versed in special education (p.138). This does not mean the general education teacher becomes a special education expert. It means that both teachers understand the accommodations and modifications documented in their students’ IEPs and can deliver them within the co-teaching structure.

For more on co-teaching structures and strategies, see the Co-Teaching Models Guide, Co-Teaching: Equitable and Inclusive Opportunities for Students, and Optimizing Co-Teaching Partnerships for Student Success.

Differentiation at Scale

“Differentiating instruction becomes more efficient, as AI can generate content tailored to diverse learner needs” (p.9). This single sentence captures the core value proposition of AI for special education and co-teaching. Differentiation has always been a matter of professional judgment and will. AI does not change the judgment — it changes the will, by making differentiation less labor-intensive.

AI-generated content at multiple reading levels. A co-teaching pair teaching a social studies unit on the American Revolution can use AI to generate the same core content at three reading levels: one at grade level, one below grade level with simplified syntax and vocabulary, and one well below grade level with visual supports and sentence frames. The content is the same; the entry point differs.

Scaffolded assignments. AI can generate assignment sheets with different levels of scaffolding: a fully scaffolded version with sentence starters and graphic organizers, a partially scaffolded version with organizational cues, and an independent version with minimal supports. The teacher determines which version each student needs; AI generates the versions.

The framework that makes this work is UDL. “Integrating generative AI as a co-teaching copilot within the frameworks of UDL and TPACK transforms the modern classroom” (p.8). UDL provides the instructional design framework — multiple means of engagement, representation, and action/expression. AI provides the production capacity to actually create the multiple versions that UDL requires. Without AI, creating three versions of everything is impractical for most teachers. With AI, it becomes achievable.

“AI copilots help teachers plan lesson plans, develop content…and differentiate tasks” (p.87). The differentiation is not the AI’s decision. The AI generates options; the teacher selects, reviews, and deploys. The teacher’s role shifts from producing every version of every material to curating and individualizing AI-generated drafts. This is a different kind of work, and it requires different skills — the skills of evaluation, adaptation, and professional judgment.

Multilingual Learner Support

English learners occupy a space that overlaps significantly with special education, and co-taught classrooms frequently serve both populations. AI tools offer particular capabilities for multilingual learner support that complement their differentiation functions.

Translation. AI copilots can translate instructional materials, directions, and content into multiple languages. Machine translation has improved substantially, though it still requires teacher review for accuracy, nuance, and cultural responsiveness. A teacher who checks AI-generated translations against their own knowledge — or a bilingual colleague’s — is using AI responsibly. A teacher who distributes unchecked translations is not.

Language acquisition scaffolding. AI can generate content with built-in language supports: simplified syntax alongside academic language, vocabulary definitions embedded in context, visual glossaries, and sentence frames that move from heavy support toward independence. These scaffolds align with the WIDA framework’s levels of language proficiency and support students at each stage of language development.

Content-accessible materials. For students in early stages of English acquisition, content-area materials written at grade level may be inaccessible regardless of content knowledge. AI can generate content-accessible versions that preserve conceptual complexity while reducing linguistic complexity — a distinction that matters enormously. Students should not be simplifying their thinking because the language of instruction is too complex. AI can help decouple conceptual demand from linguistic demand.

For strategies specifically focused on multilingual learners, see Amplifying Multilingual Learners with the Support of AI Tools. Also see Differentiated Instruction Online Instruction for Special Education and English Language Learners for approaches that apply across both populations.

Ethical Considerations with Vulnerable Populations

Working with students who have disabilities, who are English learners, or who belong to both populations raises ethical considerations that extend beyond general AI ethics. These are not optional overlays — they are central to responsible practice.

Data privacy for students with disabilities. IEPs contain some of the most sensitive information in a school’s possession: disability categories, assessment scores, behavioral data, medical information. When a special education teacher pastes IEP content into an AI copilot — even for legitimate workflow support — they may be disclosing protected information to a third party. District policies must address this explicitly: what IEP information may be entered, into which tools, under what conditions.

Bias in AI-generated IEP content. AI models reflect the data they were trained on, and that data contains biases about disability, race, language, and socioeconomic status. AI-generated IEP goals may default to lower expectations for students with certain disability labels. AI-generated accommodation recommendations may reflect common practice rather than individual student need. Every piece of AI-generated content must be reviewed through the lens of bias awareness.

Human oversight requirements. The principle is straightforward: AI should never be the final author of any document that carries legal weight for a student’s education. IEPs, 504 plans, evaluation reports, and progress notes must always reflect the professional judgment of the humans who are legally and ethically responsible for those documents. AI assists; humans decide.

Equitable access. AI tools, professional development on AI, and the time to learn and integrate them must be distributed equitably across schools and programs. If a district’s wealthier schools have AI-savvy co-teaching pairs and its under-resourced schools have neither the tools nor the training, the district has created a new equity gap on top of existing ones.

Co-Teaching Evolved’s ethics framework addresses these concerns under four categories: data privacy, age-appropriate use, plagiarism, and equitable access. Each category requires specific attention when applied to students with disabilities and multilingual learners — populations whose vulnerability to data misuse, biased algorithms, and unequal access is amplified.

For broader ethical considerations in AI use, see Harnessing Generative AI for a Comprehensive WASC Accreditation Visit and Case Managing for Special Education During Remote Learning for historical context on the evolution of special education technology and privacy concerns.

Getting Started

Do not try to implement every possibility at once. Start with one workflow, build competence, and expand from there.

The highest-ROI starting point: IEP essay feedback. This workflow (detailed above) is the best starting point because:

• It addresses a real pain point — providing detailed feedback on student writing is time-intensive
• It is technically simple — paste the essay, review the output, add your notes
• It clearly separates AI’s role (mechanical feedback) from the teacher’s role (content feedback)
• It benefits students immediately — they get faster, more comprehensive feedback
• It does not involve entering sensitive student information into AI systems (the essay itself is student work, not IEP data)

After the essay feedback workflow is established, expand to:

• Differentiated material generation for co-taught lessons
• IEP goal drafting with AI support
• Progress monitoring data synthesis

Key resources for implementation:

• Co-Teaching Models Guide — understand the structures before you try to enhance them
• Amplifying Special Education Teachers’ IEP Workflow with AI Tools — detailed IEP-specific AI strategies
• Graze and Tag: The Most Underrated Strategy in Your Co-Teaching Toolkit — co-teaching strategies that pair well with AI
• The Heart of Co-Teaching: Moving from Partner to Partnership — the relational foundation that makes co-teaching with AI effective

The sequence matters. A co-teaching pair that has not established trust, shared planning time, and a shared understanding of their students’ needs will not benefit from AI tools. AI supports effective co-teaching; it does not create it. Build the partnership first, then add the technology.

Common Barriers and Challenges

The lesson: Privacy violations don’t happen because people don’t care. They happen because the approved path is too slow or doesn’t exist. Strong AI policy prohibits entering student PII into non-approved tools and requires heightened scrutiny for tools touching IEP data — but those prohibitions only work if the district provides approved alternatives. A rule without a sanctioned path is just a sign on a wall.

The lesson: AI-generated IEP content is always a draft until the case manager signs it. Co-teaching pairs need an explicit protocol: drafts live in a ‘working’ folder, not the shared folder, and implementation doesn’t begin until the case manager’s review is complete.

This guide is part of the AI in Education Guide series. Related guides: AI for District Leaders, AI for Instructional Coaches, AI Tools Evaluation Framework. For comprehensive co-teaching strategies, see the Co-Teaching Models Guide.