How to Use ChatGPT for Lesson Planning (And Where It Falls Short)
ChatGPT can draft a lesson plan in seconds. The question isn't whether it's fast — it's whether the plan it hands you would survive a conversation with your instructional coach. Used well, general AI is a genuine time-saver. Used carelessly, it produces a smooth-reading plan with no defensible reason behind a single choice. Here's how to get the good version.
Start with a structured prompt, not "make me a lesson"
The difference between a useless plan and a usable one is almost entirely in the prompt. Give ChatGPT the same things you'd give a student teacher:
- Grade and subject ("7th grade life science")
- The specific topic or standard ("cell organelles, aligned to NGSS MS-LS1-2")
- The objective, in the standard's own verb ("students will construct an explanation of how organelles function together")
- Time available and any constraints (no lab access, 45-minute period, mixed reading levels)
- The output you want ("a lesson plan with a warm-up, direct instruction, guided practice, and an exit ticket")
A prompt with those five things produces something you can actually edit. "Write a lesson on cells" produces filler.
Prompts that actually work
| Goal | Prompt pattern |
|---|---|
| Backward-design from a standard | "Given [standard], write a measurable objective, then activities and an assessment that all target that exact verb." |
| Differentiate | "Take this lesson and give me a scaffolded version for students reading two grade levels below, and an extension for early finishers." |
| Build a check for understanding | "Write three exit-ticket questions that would tell me whether students hit [objective] — not just whether they were present." |
| Pressure-test it | "What's the weakest part of this lesson plan, and what would a skeptical observer ask about it?" |
That last one is the most underused. AI is surprisingly good at critiquing its own output when you ask it to play the skeptic.
The mistakes that get plans sent back
- Objective-activity drift. ChatGPT will happily write an objective that says "analyze" and then give you activities that only ask students to "identify." Always re-read objective → activity → assessment in a row and make sure all three target the same cognitive level.
- Generic engagement. "Have a class discussion" is not a plan. Push for the actual question, the protocol, the timing.
- Invented authority. This is the big one. If you ask ChatGPT why a method works, it will often produce a confident-sounding citation — a researcher, a year, a study — that may not exist. General language models generate plausible text, not verified references. Never paste an AI-generated citation into a plan you'll hand to an evaluator without checking it yourself.
The one thing general AI can't do
The hard part of planning was never the writing — it's the "why." When an administrator, a parent, or a National Board portfolio asks you to justify a design choice, "ChatGPT suggested it" is not an answer. You need the actual learning-science reason: this is retrieval practice, here's the research, here's why it raises retention.
That's the gap. General AI writes the lesson; it can't reliably ground it. It doesn't know which of your moves is a real, named, evidence-backed method versus a nice-sounding idea — and it can't cite a source it's actually allowed to use.
A planning tool that closes the gap
EvidenceLesson was built for exactly this. Instead of free-generating text, it selects from 165 research-validated teaching methods, sequences them into a coherent workflow, and cites a real source for every step — pulled from a curated literature library, never invented. You enter grade, subject, topic, and standard; it returns a lesson, rubric, assessment, or slide outline you can defend, then exports to Word or PDF.
Use ChatGPT for fast first drafts. When the plan has to stand up to scrutiny, generate it with the evidence already attached.