How I Built an AI Planning Space That Actually Knows How I Teach

This may look similar to last Monday's post. I had several people reach out and ask for some additional steps or context on how I built the structure of my Claude project space. Also, the World History resource page is LIVE!!!!!!

Every August I tell myself this will be the year I stay ahead of my own lesson plans instead of writing them at 10pm the night before.

This year I tried something different.

Before I planned a single lesson for tenth grade Modern World History, I spent an afternoon building the planning space itself. Specifically, I set up a dedicated Claude project and loaded it with the documents that actually govern how I teach, so that every lesson plan, activity, and reading list I generate from here on out is automatically grounded in my district's curriculum, the right standards, and how I actually run a classroom.

Not a generic version of a world history teacher. Mine.

If you have ever asked a generic AI tool for a lesson plan and gotten back something that looked polished and was almost completely useless: wrong standards, wrong period length, rubrics that match nobody's gradebook, this post is for you. Here is exactly what I put in the space, why, and what it changed.

The Problem With Asking AI Cold

The first few times I asked a general-purpose AI for a lesson plan, I got something that looked great on the surface and fell apart the moment I tried to use it.

It invented standards that did not exist. It assumed a 45-minute period, when part of my week runs 90-minute blocks. It cited sources I had never assigned. It graded everything like a traditional points-off rubric, when my philosophy is mastery-based with open retakes.

None of that is the tool's fault. It was answering a generic question because I had only asked a generic question.

The fix was not a better prompt or switching tools. It was giving my tool of choice a permanent memory of the specific things that make my classroom mine, so unlike in the classroom, I never have to repeat the directions.

What Actually Went Into the Space

Six documents and one reference file. Every single one earns its place.

The district curriculum map.

This is the most important document in the whole project. My district updated its Modern World History scope and sequence in 2022, restructuring the curriculum into seven units. If I had not uploaded this and flagged which version is authoritative, any AI tool would happily generate lesson plans for a unit structure that it created all willy-nilly. Confidently and incorrectly. Uploading the actual document, not just describing it from memory, means the tool is working from grounded truth instead of my recollection of a curriculum meeting from three years ago.

The NCSS C3 Framework PDF.

While I am not explicitly required to align instruction to the College, Career, and Civic Life Framework (the NCSS C3 Framework), I like it a little bit better than our explicit standards in the state of Kansas. So, I uploaded the framework itself (the massive PDF file with EVERYTHING) rather than trusting Claude to already know it accurately. This matters especially for Dimension 3 and Dimension 4, which show up constantly in how I structure discussion and writing tasks.

Two lesson plan templates: one standard, one block.

My school runs 53-minute periods three days a week and 90-minute blocks the other two. Rather than explain that split every time, I uploaded both template structures once. Now the tool already knows a block period needs two full rounds of instruction, practice, and closure instead of one.

The state's scoring rubric for the assessment.

I uploaded the HGSS scoring rubric for the state of Kansas so that any assessment the tool drafts or I build starts from something my colleagues already recognize, and something the state requires. Not a rubric invented on the spot that I have to translate afterward.

A teaching philosophy and grading reference document.

This is the one I would tell any teacher to write first, because it is the one no curriculum document will ever contain.

Mine spells out the things that only live in my head and my gradebook. I use total-points grading, not weighted categories. Retakes are open by default. Missing work gets a 50% floor instead of a zero. Late work is not penalized except on two or three major deadlines per semester.

It also captures my actual activity library: Project Zero thinking routines like Philosophical Chairs and Save the Last Word for Me, plus the growing collection of disguised mastery constraint activities I have been developing all summer: Historical Whodunnit, Nextdoor Neighborhood Watch, the $5.00 Story, and more. And it includes the Binary + 1 rubric I built specifically for grading that kind of creative, non-traditional work.

None of this lives in any district document. It only exists because I sat down and wrote it out deliberately, once, as a reference the AI could return to instead of asking me to re-

teach my own philosophy in every conversation.

At the end, I asked Claude to ask me any other questions about my teaching philosophy it might need to know. This sparked 3-4 more items to write about, and I have been able to add more as I think of items throughout my planning process.

An instructions file that governs how the tool behaves in this space.

This piece makes the biggest difference.

Instead of hoping the tool infers the right format, I wrote explicit rules: always ask for topic, duration, and learning objectives before generating anything. Always ask whether it is a standard period or a block. Always structure lesson plans in the same order: warm-up, direct instruction, practice, closure, assessment, differentiation notes, teacher reflection space. Always name the specific standard and code at the top of the plan.

And the rule I care about most: never invent a source, author, page number, or publication date. If nothing relevant has been uploaded, say so directly instead of guessing.

That last rule alone eliminated the single biggest trust problem I have with AI-assisted planning. Fabricated citations that look completely legitimate are a real problem. Explicit instructions are the solution.

Why the Order Matters

I did not upload these in a random order, and I do not think the order is trivial.

The curriculum map and standards documents went in first, because everything else needs to be checked against them. The templates and rubric went in next, because they are structural, about how content gets shaped rather than what the content is. The teaching philosophy document went in last, because it is the most personal and specific, and I wanted it to be read in the context of everything more official that came before it.

The instructions file sits on top of all of it, acting less like a document and more like standing orders for how the tool should behave every single time I open the space.

What Changed Once the Space Was Built

The clearest test came when I asked for a first lesson plan.

Before generating anything, the tool asked me for the topic, duration, and prior knowledge to assume, exactly as instructed, and then asked whether it was a standard or block period. Small thing. Enormous difference. That is the gap between a tool that guesses and a tool that checks.

When the plan came back, it named the specific district standard code at the top. It used my seven-unit structure without me explaining it. It suggested one of the thinking routines from my own activity library instead of a generic class discussion.

When I later asked for a reading list, it did not invent a source. It told me directly that nothing in the project matched what I was asking for, which was true, and which is exactly what I had asked it to do.

None of that required a clever prompt in the moment. It required doing the unglamorous work upfront: gathering the documents that actually govern my classroom, writing down the parts of my practice that only exist in my head, and being explicit about the behaviors I most needed.

If You Want to Try This

You do not need seven documents to start.

The two that matter most are the ones few remember to write down: an explicit instructions file telling the tool how you want it to behave: ask before generating, never fabricate a source, use your period structure; and a personal teaching philosophy reference capturing the grading policies and activity types that live only in your head and your gradebook, not in any district PDF.

Everything else: curriculum maps, standards frameworks, templates, rubrics, you probably already have sitting in a folder somewhere. The work is not creating new material. It is finally handing over what you already have, once, so you stop re-explaining it every single time.

Grab the Documents!

Everything described in this post: the templates, the rubric, the teaching philosophy document, the instructions file, is now living on the World History resource page alongside the Unit Zero lessons and curriculum correlation spreadsheets. As I build new units, generate new activities, and refine the system across the semester, the page will be updated to reflect what is actually being used in a real classroom in real time. Not a finished product. A living document. That was the Year Zero promise, and it still is.

Play Around! What Are You Looking For!

Year Zero is not just a classroom rebuild. It is a systems rebuild. And this is one of the systems.