Teachers in the US work an average of 54 hours per week, but only about 45 percent of that is actual classroom instruction. The rest is planning, grading, communication, and administration. Which means that if you're burning out, it's probably not the teaching that's doing it. It's everything around the teaching.
AI tools have become genuinely useful for the surrounding work. Not for the judgment calls that define good teaching: deciding what a student actually needs, reading a room, knowing when to push and when to back off. Those stay with you. But for the mechanical production of lesson plans, differentiated worksheets, rubrics, and parent emails, AI can cut the time investment by 50 to 70 percent without cutting the quality.
Lesson Planning: The Tuesday Afternoon Workflow
Here's the specific situation. It's Tuesday at 3:45 PM. You need to plan Wednesday's lesson on fractions for your 5th grade class. You have a general idea of what standard you're covering (CCSS.MATH.CONTENT.5.NF.A.1 if you're in a Common Core state). You have 50 minutes. You know three kids are going to find this abstract and four are already past it. You have no prep period left.
Instead of starting from a blank document or cobbling together materials from three different websites, try this prompt in Claude or ChatGPT:
"Create a 50-minute 5th grade math lesson on adding fractions with unlike denominators (CCSS 5.NF.A.1). Structure: 5-minute warm-up using a real-world hook, 15-minute direct instruction with worked examples, 20-minute guided practice in pairs, 10-minute independent check. Include two differentiation notes: one for students who need more scaffolding, one for students who are ready to extend. Give me the actual problems to use, not just descriptions of problems."
The "give me the actual problems" instruction is critical. Without it, you get a lesson framework with placeholders. With it, you get something closer to a usable draft. You'll still edit it. You'll adjust for your specific kids. But you're editing instead of generating, and that's 30 minutes you don't have to spend.
For high school English, the same approach: "Create a 45-minute lesson on close reading of this passage [paste passage]. Include three guiding questions that move from literal to inferential to evaluative. Include one discussion protocol (not just 'discuss with a partner'). Include exit ticket." Again, specificity in your prompt produces specificity in the output.
Differentiated Materials
Differentiation is where teachers spend disproportionate planning time relative to classroom impact. The standard workflow: write a grade-level text or problem set, then manually create a modified version for students with IEPs or English learners, then create an extension for advanced students. Three versions of the same thing, each taking almost as long as the first.
AI collapses this. Write one version, then ask for the others:
"Here is a reading passage about the American Revolution written at a 6th grade level. [Paste passage.] Create three versions: one simplified to a 4th grade reading level with vocabulary definitions in the margin, one that keeps the original, and one extension version that includes a primary source excerpt and a comparison question. Keep the core content and main ideas the same across all three."
Review the simplified version carefully. AI tends to over-simplify in ways that strip out important nuance, or it will introduce imprecise language in trying to be accessible. The extension version usually needs more work. But you're editing three drafts instead of writing them.
For math differentiation, the approach is more granular. Rather than asking for whole versions, ask for scaffolding additions: "Take this problem set and add a step-by-step scaffold frame for each problem that shows students how to start without giving away the solution." That's a more targeted request and produces more consistently useful output.
Rubrics
Rubrics are time-intensive to write from scratch and often mediocre when you do, because the hard part isn't knowing what good work looks like. It's articulating the difference between a 3 and a 4 in a way students can actually use to self-assess.
"Create a 4-point analytic rubric for an 8th grade argumentative essay. Criteria: Claim and Thesis, Evidence and Support, Reasoning and Analysis, Organization, Conventions. For each criterion, write descriptions that distinguish clearly between each level (1-4). Use student-friendly language. The 4-level description should describe what mastery looks like specifically, not just 'excellent' or 'thorough'."
The output is almost always usable with light editing. The main thing to check: whether the 3-versus-4 distinction is meaningful enough that students can tell the difference. AI often writes 3 and 4 descriptors that are variations on each other. Push back on those specifically: "The 3 and 4 descriptions for Evidence feel too similar. Make the distinction more concrete."
Parent Emails
Parent communication takes up more teacher time than most administrators realize. A difficult email about a struggling student can take 20 minutes to draft because the stakes feel high. AI handles this well, with a caveat: you need to give it specifics, not generalities.
For a concern email to a parent about a student who has been disruptive and is falling behind:
"Draft an email to a parent whose child has been frequently interrupting class and is currently missing three homework assignments. I want to: express concern, share specific observations without sounding accusatory, invite the parent to be part of a solution, and propose a brief 15-minute call this week. Keep it warm but direct. Don't use educational jargon. Under 200 words."
Edit the output to include the student's name and the specific incidents. Generic emails feel generic to parents and they notice. The AI draft gets you the structure and tone; you add the specifics that make it real.
For positive communication (which should happen more often than concern communication), the workflow is faster: "Draft a 3-sentence positive note to a parent whose child has shown significant growth in reading fluency this month. Specific, warm, brief." Takes 30 seconds to prompt and 30 seconds to personalize. Building this habit for two or three positive emails per week doesn't add much time and significantly changes your relationship with parent communication.
IEP Support
This one comes with a larger caveat: IEP documentation has legal requirements, and every word matters. AI is not a substitute for your special education coordinator or your district's guidance. What AI is useful for is the pre-writing work that IEPs require before you get to the formal document.
Goal writing is the most obvious application. SMART IEP goals follow a specific structure: given a condition, the student will do something, measured by some metric, within a timeframe. AI is good at this structure.
"Help me write an IEP goal for a 3rd grade student with a reading disability who is currently reading at a 1st grade level. The goal should address reading fluency, follow a SMART structure, and be measurable using oral reading fluency (ORF) scores. Write three options with slightly different targets and conditions so I can choose the best fit."
Run the output past your SPED coordinator before it goes into any official document. AI doesn't know your student, doesn't know your district's benchmarks, and can produce goals that sound right but don't match the student's actual starting point. It's a drafting accelerator, not a compliance tool.
Tools That Are Worth Knowing
MagicSchool AI is built specifically for teachers and has pre-structured tools for lesson planning, rubric creation, IEP goals, and more. The advantage over general tools like Claude or ChatGPT is that the interfaces are designed around teaching workflows. The tradeoff is that you get less flexibility. For teachers who find prompt engineering awkward, MagicSchool's guided interface lowers the barrier.
Claude (Anthropic) handles longer documents better than ChatGPT in most comparisons, which matters when you're pasting full articles for reading lessons or complete rubrics for revision. For back-and-forth conversation and refinement of a specific output, either works fine. ChatGPT's free tier is more limited but still functional for basic lesson planning.
NotebookLM from Google is worth knowing for secondary teachers who assign research-heavy units. Students (and teachers) can upload multiple source documents and ask questions across them. It's better than having students summarize one source at a time.
What This Doesn't Replace
AI doesn't know your students. It doesn't know that Marcus shuts down when he's cold called, or that Jaylen lights up when math connects to basketball, or that the group that sits near the window needs direct proximity or they disengage. Those observations are the professional knowledge that makes a good teacher irreplaceable. AI produces adequate lesson plans. You produce lessons that work for the specific humans in your room.
The goal isn't to use AI to do your planning for you. It's to use AI to do the mechanical scaffolding fast enough that you have time and energy left for the professional judgment that nobody else can provide.