::play 01 · Lesson Plan Draft (60-minute period, one prep)
Claude.ai or ChatGPT (Free is fine)
Get a teachable lesson plan in under 5 minutes — objectives, hook, instruction, practice, exit ticket — that you can edit, not write from scratch.
You are an experienced [SUBJECT] teacher in [GRADE LEVEL]. Draft a 60-minute lesson plan on [TOPIC] aligned to [STATE STANDARD CODE if known, otherwise 'Common Core' or 'NGSS']. Structure: (1) Learning objective in student-friendly 'I can' language, (2) 5-minute hook or warm-up that activates prior knowledge, (3) 15-minute direct instruction with one anchor example, (4) 20-minute guided/independent practice with built-in checks for understanding, (5) 10-minute closure activity, (6) 10-minute exit ticket — 3 questions, one at recall, one at application, one at synthesis. Assume a class of 28 with 4 students reading 2 grades below level and 2 students needing extension. Note where I should differentiate. No student names — this is a generic class.
::what to notice · Whether the hook actually engages a 7th grader or sounds like a textbook. Whether the exit ticket questions are answerable from what the lesson actually taught.
::trap · AI loves to overwrite. If the plan is 4 pages, it's too dense for a 60-minute period. Cut anything that doesn't earn its minute.
::play 02 · Three-Level Differentiated Worksheet (one assignment, three difficulties)
Claude.ai or MagicSchool
Generate three versions of the same worksheet — below-level, on-level, above-level — that all hit the same learning objective so the class can do parallel work without anyone feeling singled out.
Create three versions of a worksheet on [TOPIC] for [GRADE]. All three must assess the same learning objective: [OBJECTIVE]. Version A (below-level): more scaffolding, sentence starters, vocabulary support, fewer items, easier numbers/text. Version B (on-level): standard practice, mix of recall and application. Version C (above-level): same core but adds an extension question requiring synthesis or transfer to a new context. Same look/format/font feel for all three so they're visually indistinguishable when handed out. Include answer keys for each. 10 items per version. Do not label them A/B/C on the student copy — use unobtrusive codes like a small dot pattern in the footer.
::what to notice · Whether Version A actually scaffolds (sentence starters, visuals, reduced cognitive load) or just gives easier problems. Easier ≠ scaffolded.
::trap · Three versions of the same dull worksheet is still a dull worksheet. If the on-level version is boring, all three are boring.
::play 03 · Parent Email — Concern Without Conflict
Claude.ai
Draft a parent email about an academic or behavioral concern that opens conversation rather than putting the family on defense — calm, specific, partnership-framed.
Draft a parent email from a [SUBJECT/GRADE] teacher. The concern: [DESCRIBE BEHAVIOR OR ACADEMIC PATTERN IN GENERIC TERMS — e.g., 'a student has missed 3 of last 5 homework assignments and seems disengaged in class']. Do not use the student's name — write 'your child' throughout, I will swap in the name. Tone: warm, professional, partnership-oriented. Lead with one specific strength I've noticed. State the concern factually with one or two concrete observations. Avoid diagnostic language. Invite the parent to share context I might not have. Offer two times for a brief call or in-person meeting. Close with confidence in the student. 150-200 words. No exclamation points. No 'I'm sure' or other minimizers.
::what to notice · Whether the opening strength feels real or tacked on. Whether the concern is described factually (what you saw) vs. interpretively (what you think it means).
::trap · AI defaults to corporate-warm tone that reads as fake to families. Rewrite the first and last sentences in your own voice every time.
::play 04 · Formative Check Generator (3-question exit ticket)
ChatGPT or Claude.ai
Generate a quick, well-calibrated exit ticket that tells you who got it, who's close, and who needs reteaching — in three questions, not ten.
Generate a 3-question exit ticket for [GRADE] [SUBJECT] on the objective: [OBJECTIVE]. Question 1 = recall (can the student state the concept). Question 2 = application (can they apply it to a routine example). Question 3 = transfer (can they apply it to an unfamiliar context or recognize when it doesn't apply). All three should be answerable in under 4 minutes total. Include an answer key with notes on common misconceptions revealed by likely wrong answers — what does it tell me if a student misses Q2 but gets Q3, or vice versa. No multiple choice unless I ask — prefer short open response so I can see thinking.
::what to notice · The misconception notes are the gold. If a wrong answer pattern actually maps to a specific gap, you know what to reteach tomorrow.
::trap · Three questions is the point. If the AI gives you five, cut two. Long exit tickets eat instruction time and don't tell you more.
::play 05 · Unit Retro (Sunday-night, 15 minutes)
Claude.ai
Reflect on a unit you just finished — what worked, what didn't, what to change next year — without staring at a blank doc.
I just finished a [LENGTH]-week unit on [TOPIC] in [GRADE] [SUBJECT]. I'll dump my unstructured notes below. Organize my reflection into: (1) What worked — instructional moves, materials, sequencing decisions that landed. (2) What didn't — places where most of the class struggled, lessons that needed reteaching, time misallocations. (3) Surprises — anything I didn't expect (good or bad). (4) Three specific changes for next year. (5) One question I still don't have a good answer to. Don't soften my critique of my own teaching — I want honest, not affirming. Don't add fake insights I didn't put in my notes. Notes: [PASTE NOTES HERE — no student names, use 'Student A' or 'the class' or 'about 6 kids']
::what to notice · Whether the three changes are actually actionable next year, or generic 'engage students more' filler.
::trap · If you don't write notes during the unit, the retro will be hollow. Keep a sticky note on your desk: 'unit retro line of the day.'
::play 06 · Family Newsletter Blurb
Claude.ai or MagicSchool
Write the weekly classroom-update paragraph that families actually read — specific, warm, useful, under 150 words.
Write a 120-150 word classroom newsletter blurb for families of [GRADE] [SUBJECT] students. This week we covered: [TOPIC]. Highlight: (1) The big idea in plain language a family member with no background in [SUBJECT] can understand, (2) One thing students did this week that I'm proud of — phrased as a class accomplishment, not naming individuals, (3) One thing families can ask their student about at dinner to extend learning (a real conversation prompt, not 'ask about their day'), (4) Heads-up on next week. No emoji. No exclamation points beyond one. Tone: confident, warm, slightly wry — like a teacher who knows their stuff and likes the kids.
::what to notice · Whether the dinner-table question is actually one a parent could ask without already knowing the answer.
::trap · Newsletter blurbs that brag about the teacher get ignored. Blurbs about what families can do get read.
::play 07 · IEP Goal Draft Scaffold (case manager workflow)
MagicSchool, SchoolAI, or a district-vetted tool — NOT public ChatGPT/Claude.ai
Get a draft of measurable, observable IEP goal language to edit — never to submit as-is. The AI scaffolds; you write the actual goal with your knowledge of the student.
Draft three variations of a measurable annual IEP goal for a student in [GRADE] working on [SKILL DOMAIN — e.g., 'reading comprehension of grade-level informational text']. Present level (no identifying info): [BRIEF DESCRIPTION — e.g., 'currently reading 2 grade levels below, decodes accurately but loses meaning across paragraphs']. Format each goal: condition + behavior + criterion + measurement schedule. Use observable verbs (identifies, states, writes, demonstrates) — not internal verbs (understands, knows, appreciates). Specify criterion as a percentage AND a frequency (e.g., '80 percent accuracy on 4 of 5 trials'). Include progress monitoring method appropriate for the goal. Three variations should differ in ambition level — one conservative, one stretch, one moderate.
::what to notice · Whether the verbs are actually observable. 'Understands fractions' is not a goal; 'solves fraction-addition problems with unlike denominators with 80 percent accuracy on 4 of 5 trials' is.
::trap · An AI-drafted IEP goal is a starting point, not a finish line. You sign it. You're accountable. Edit until it sounds like you wrote it about this specific student.
::play 08 · Rubric Calibration Pass (before you grade essays)
Claude.ai
Before you grade a stack of essays, use AI to help you sharpen the rubric so your grading is consistent across the stack — not to grade for you.
I'm about to grade a stack of [GRADE] [SUBJECT] essays on [PROMPT]. My rubric has these dimensions: [LIST DIMENSIONS, e.g., 'thesis clarity, evidence use, organization, mechanics']. For each dimension, write a sharp 1-sentence description of what a 4 looks like, a 3, a 2, a 1. Then give me one example sentence or paragraph that exemplifies each level — short, realistic for the grade level. Then list 3 common rater-bias traps I should watch for when grading (e.g., halo effect from clean handwriting, fatigue in the back half of the stack, bias against non-standard dialect). After I grade, I'll spot-check 5 random essays against the rubric to verify I stayed calibrated.
::what to notice · Whether the exemplar sentences feel like real student writing or generic AI prose. If they're too polished, they won't calibrate you to actual kid work.
::trap · Do not paste student essays into the AI to grade. The rubric is the artifact; your eyes do the grading. Always.