The Only Lesson Planning AI I Actually Keep Using

Somewhere on my laptop there is a folder called "AI Tools Tried" with seventeen bookmarks in it.

Eleven of those bookmarks go to platforms I used once, found underwhelming, and never opened again. Two go to tools I genuinely liked but stopped using because the free tier ran out mid-unit and I was not ready to pay. One goes to something that generated a lesson plan so generic it described a classroom that does not exist anywhere. The remaining three I actually use.

I am telling you this because most articles about AI lesson planning software are written as if every tool deserves equal consideration and your job is to compare them neutrally and pick the best one for your needs. That is not how this works in practice. In practice you try things, most of them disappoint you in specific ways, and you eventually settle on a small number that fit your actual workflow. The useful question is not which tool scores highest on a feature comparison. It is what makes the tools that actually get used different from the ones that get abandoned.

That question has a specific answer for me and I want to give it to you directly.

The Problem With Lesson Planning That AI Does and Does Not Solve

Before I tell you what I use, I want to be honest about what the problem actually is.

Lesson planning is slow for two distinct reasons and AI only helps with one of them.

The first reason is blank page paralysis. You know roughly what needs to happen in a lesson. You have a sense of the sequence. But turning that sense into a structured document with timing and objectives and transitions takes a kind of focused writing effort that is genuinely draining when you are doing it for the fifth lesson of the week on a Wednesday evening. This is the part AI addresses well. A good prompt in a good tool produces a usable structure in under two minutes and removes the activation cost of starting from nothing.

The second reason lesson planning is slow is that you have to think carefully about your specific students. What they already know, where they typically get stuck, which misconceptions are likely to surface, how much time a particular activity will actually take with this group. No AI tool knows any of that. The tools that claim to personalise to your students are, at best, adjusting reading level based on grade inputs. They are not accounting for the fact that your third period class loses focus after twelve minutes of direct instruction or that your sixth period asks better questions when they have had two minutes to write first.

This distinction matters because it sets realistic expectations. AI lesson planning tools save the blank page time. They do not save the thinking time. If your planning is slow because you are a careful thinker who takes instructional decisions seriously, AI will help you somewhat but not dramatically. If your planning is slow because the blank page drains you before you have even started thinking, AI will help you a lot.

For me it is mostly the second. The blank page is my enemy. The thinking I actually enjoy.

What I Tried and What I Abandoned

I used a dedicated lesson planning platform for about six weeks earlier this year. I will not name it because my experience might not be representative and the point is the pattern rather than the product.

The platform had a clean interface, good standard alignment features, and a generation speed that was genuinely impressive. It produced complete lesson plans in under a minute. The problem was that the plans it produced were structurally identical regardless of what I put in. Every lesson had the same warm up format, the same three part main activity structure, the same exit ticket style. After three weeks I could predict the shape of the output before I had finished typing the prompt.

That is the core failure mode of purpose built lesson planning software. It is optimised for consistency and speed, which produces reliable output that all looks the same. After a while the lessons start to feel the same too, because they are built on the same skeleton.

I went back to using Claude for anything that needed to be genuinely different and MagicSchool for the lessons that just needed to exist quickly without much creative investment. That combination has held for several months now and I want to explain specifically why each one earns its place.

MagicSchool for the Lessons That Just Need to Happen

About sixty percent of my weekly planning falls into a category I think of as necessary but not precious. A practice lesson on a skill students have been introduced to. A review session before an assessment. A structured reading day. These lessons matter and they should be well designed but they do not need to be inventive. They need to be functional, appropriately timed, and aligned to what we are doing in the unit.

MagicSchool handles this reliably. I enter the grade level, the subject, the standard I am addressing, the duration, and any constraints that matter for that lesson. The output comes back in about ninety seconds with a warm up, an instructional sequence with time allocations, a formative check, and a closure. The differentiation options are built into the generation rather than added as a note at the bottom, which is a specific design choice that reflects actual classroom thinking.

I edit the output. Usually about twenty percent of it. The time allocations are the thing I check most carefully because AI generated timing is consistently optimistic. An activity allocated fifteen minutes often needs twenty with a real class. I adjust those before anything else.

Total time for a functional lesson plan using MagicSchool: about twelve minutes including my editing pass. That is the real number, not the ninety second generation time the tool would like you to focus on.

Claude for the Lessons That Need to Be Good

The other forty percent of my planning involves lessons where the design actually matters. Unit launches. The lesson where a difficult concept is being introduced for the first time. A lesson I know will determine whether the unit lands or loses the class. These are the lessons where I used to spend an hour and still felt uncertain at the end.

Claude is not a lesson planning tool. It does not have a form to fill in. It has a conversation, which means the quality of what you get is directly proportional to the quality of what you put in.

The prompt I use for lessons that need to be good: I describe what my students already know, the specific gap I am trying to close, what I want them to be able to do by the end of the lesson, and any constraints on the design. Then I ask for a lesson that creates genuine curiosity or cognitive surprise before any content is delivered, and I ask Claude to explain in one sentence why it chose each activity.

That last instruction, explain the reasoning, is the thing that separates useful AI lesson planning from template generation. When Claude explains why it chose a particular opening activity or why the sequence moves in a specific direction, I can evaluate whether that reasoning makes sense for my students. If it does not, I push back and the plan improves. If it does, I have a lesson I understand well enough to adapt on the fly when the room needs something different.

The lesson that I am most pleased with from this semester came from a Claude conversation that took about twenty five minutes. It was an introduction to unreliable narrators for a ninth grade English class. The opening Claude suggested was showing students two eyewitness descriptions of the same event, both confident, both contradictory, and asking them to figure out what actually happened before I had explained anything about narration or reliability. The class argued for nine minutes before I said a word. The concept landed in a way it had not when I taught it with a more conventional approach the year before.

I would not have designed that opening myself on a Wednesday evening with four other lessons still to plan. The blank page would have beaten me and I would have started with a definition.

The Honest Recommendation

Use MagicSchool for the majority of your planning. It is fast, the output is reliable, and the standard alignment is automatic. Expect to edit the timing and expect that all the plans will have a similar structure. That is a reasonable trade for the time it returns.

Use Claude for the lessons that matter most. Invest fifteen to twenty minutes in the conversation. Ask it to explain its reasoning. Push back when the reasoning does not fit your students. The output will be better than what MagicSchool produces and better than what most of us produce alone on a tired evening.

Do not use a dedicated lesson planning platform unless you have tried both of the above and found them insufficient. Most dedicated platforms are fast and consistent in ways that produce uniformity over time, which is a cost that does not show up immediately but accumulates across a semester.

The seventeen bookmarks in my folder represent real time I spent on tools that did not earn a place in my workflow. I am not telling you what to use. I am telling you what happened when I used everything and then had to decide what to keep.

Three tools survived. Two of them are free. One of them is a conversation, not a form.

That is enough.