Take one general-education assignment and adapt it to match the accommodation and modification language in a student's IEP — without putting any IEP content into a consumer AI tool. Reading-level adjustments, chunking, sentence frames, and visual scaffolds.
Take one assignment you would otherwise teach as a single document and produce an IEP-friendly version that supports the most common accommodation categories — reading-level adjustment, chunked tasks, sentence frames, and visual scaffolds — without disclosing any specific student's IEP content to a third-party AI tool.
Every special-education teacher and every general-education teacher who has students with IEPs in their classes knows the drill. The IEP says the student gets "shortened assignments," "chunked instructions," "sentence frames for written response," and "reading materials at instructional level." The general-education assignment was not built that way. So you spend 30-45 minutes per assignment per student rebuilding the task to honor the accommodation. Multiply by the number of IEP-served students times the number of assignments per week and the math becomes brutal.
AI can compress that 30-45 minutes to 5-10 if you give it the original task and the generic accommodation type to apply — not the student's specific IEP. The output is a parallel version of the assignment that you review against the IEP language, edit, and use. The model is producing the scaffold; you are matching it to the student.
For each of these, AI is producing a draft. Your professional judgment maps the draft to the specific student. The IEP itself never enters the chat.
Here is the template. Notice the absence of student name, IEP number, disability category, or any identifying information.
You are an instructional designer helping a [GRADE LEVEL] [SUBJECT] teacher
adapt an assignment for a student who has the following accommodations
in their plan. The student is not identified to you.
GENERIC ACCOMMODATIONS TO APPLY:
- Reading materials at approximately [TARGET READING LEVEL] level
- Chunked instructions: each direction a single numbered step
- Sentence frames for any extended written response
- Vocabulary box with the 6 most important terms defined
in plain language at first use
- Estimated completion time approximately 60% of standard task
ORIGINAL ASSIGNMENT:
[Paste original assignment text here. No student work, no student name.]
OUTPUT REQUIREMENTS:
- Produce a parallel version of the assignment that applies all five
accommodations above.
- Preserve the same learning objective as the original.
- Do NOT remove the cognitive demand. Adapt the access, not the rigor.
- Format: clean, single-page printable.
- End with a one-paragraph teacher note: which exact moves you made
and why, so I can verify the adaptation matches the plan.
Run that prompt with one assignment. The output is a parallel assignment, scaffolded for access. Your job is to read it against the actual IEP language for your student and confirm the scaffolds match. If the IEP says "chunked into no more than three steps" and the AI produced six chunks, you collapse them. If the IEP specifies a particular reading level, you check the output against a readability tool (the Flesch-Kincaid score in your word processor is a fast first check).
The IEP is one of the most heavily protected documents in a student's education record. Under FERPA (20 U.S.C. § 1232g), it is "personally identifiable information" and disclosure outside the team that needs it requires written parental consent. Most state special-education statutes layer additional protections on top of FERPA. The Individuals with Disabilities Education Act (IDEA) procedural-safeguards rules also restrict who can see IEP content.
Practical consequence: do not paste the IEP into a consumer AI tool. Do not paste the disability category. Do not paste the student's name with the accommodation. Do not paste the present-levels statement. Do not paste the goals. Do not paste the assistive-technology assessment.
What you can paste is generic accommodation language ("reading at a fifth-grade level," "chunked instructions," "sentence frames"). That language is descriptive of a category of scaffold; it is not the IEP itself. The model produces a scaffold; you bring it back to the IEP-protected document and check it offline.
Take one assignment from your current unit and produce three adapted versions: one with reading-level adjustment only, one with chunking and sentence frames only, and one with all five accommodation categories applied. Print all three. Compare them to your existing adaptations for one of your IEP-served students (without pasting the IEP itself). Where does the AI version match? Where does it miss? The pattern of misses tells you what your students need that AI alone cannot deliver.
Day 4 turns to family communications — arguably the most FERPA-sensitive writing in your week.