You typed a solid prompt. "Write me a 7th grade lesson on central idea and supporting evidence, aligned to the standard, with a warm-up and an exit ticket." Ten seconds later you had a lesson.
And it was… fine. Technically correct. Completely generic. The warm-up was a bland "turn and talk," the "activity" was a worksheet you'd never actually hand out, and there was nothing in it that fit your seventh graders — not the eight ELLs in Period 3, not the IEP accommodations you're legally required to honor, not the honors kids who'd eat it for breakfast. So you spent the next forty minutes rewriting it. Which raises the obvious question: what did the AI actually save you?
If that's been your experience with AI lesson planning, you're not doing it wrong. The tools are built in a way that almost guarantees generic output. Here's why — and what to look for instead.
Reason 1: They're trained on the average of the internet
Most general AI tools learned to "plan lessons" by ingesting an enormous pile of whatever was online — blog posts, random worksheets, forum answers, curriculum fragments of wildly varying quality. When you ask for a lesson, you get back the statistical average of all that.
The average of the internet is, by definition, generic. It has no point of view. It can't tell a strong instructional move from a weak one because it was trained on both, equally. That's why the output feels like it was written by someone who read about teaching but never taught.
Reason 2: They have no pedagogical framework
Ask a generic AI for a lesson and it will happily produce a warm-up, an activity, and an assessment — but it has no underlying theory of why those pieces go together or what makes each one effective. There's no coherent method underneath, just a plausible-looking shape.
Good teaching isn't a shape. It's a sequence of deliberate decisions: how you surface prior knowledge, how you make thinking visible, how you check for understanding before it's too late, how you build in the flexibility to adjust mid-lesson. A tool with no framework can imitate the format of a lesson without any of the reasoning that makes one work.
Reason 3: They optimize for "an answer," not "your class"
A generic generator's job is done the moment it hands you a lesson. It has no concept of the five different classes you actually teach that lesson to. So it gives you one version — the middle-of-the-road one — and leaves the differentiation to you.
But differentiation isn't a finishing touch. It's most of the work. The reading level, the language supports, the scaffolding, the extension — that's the part that eats your Sunday, and it's exactly the part a one-shot generator ignores.
What to do instead
The problem isn't "AI." The problem is AI with no method and no knowledge of your students. So when you evaluate an AI lesson planning tool, stop asking "can it write a lesson?" (they all can) and start asking three better questions:
1. What is it actually built on? Is it the average of the internet, or a real, coherent methodology? A tool built on an actual method makes deliberate instructional choices instead of plausible-looking ones. That's the difference between output you rewrite and output you trust.
2. Does it know who's in my class? Can you tell it about your ELLs, your IEP accommodations, your reading levels, your MTSS tiers — and does it actually change the lesson accordingly? If "differentiation" means you still do all the adapting by hand, it hasn't solved your real problem.
3. Will it hold up when I'm observed? A lesson isn't done if it falls apart the moment an evaluator with a Danielson or Marzano rubric walks in. The questioning, the checks for understanding, the engagement — those need to be built in, not bolted on.
This is exactly why we built Tangram differently
Tangram isn't trained on the internet. Tan — our AI planning partner — is powered by an original methodology our founders developed over 50+ years, refined across hundreds of districts and thousands of teachers coached. We didn't curate other people's best practices. We built something new, tested it, and taught it.
That means you build one strong lesson on a real method, then tell Tan who's actually in each class. It adapts every section for your ELLs, your IEP accommodations, your reading levels — and because the personalization is rule-based, there's no extra cost per adapted copy. Every version is observation-ready out of the box, aligned to Danielson, Marzano, and Understanding by Design.
No more generating a bland lesson and rewriting it five times. You direct; Tangram adapts.
See what a real method feels like
Build one strong lesson and watch it adapt for every class you teach. 3 free lessons — no credit card.
Try 3 lessons free →The next time an AI hands you a generic lesson, you'll know exactly why. And you'll know what to use instead.