How We Build a Week of Lessons With AI

A lot of people hear "AI-assisted lessons" and picture a teacher typing "make me a worksheet." That's not what this is. What I do each week is closer to a production pipeline, where one set of materials feeds the next and every piece gets checked before it goes out. Here's an overview of how it works. Future posts will break down each step in detail.

It starts with my slides. Each week, I have two or three slide decks, one for each lesson day, and I've already built them. I load them into Claude, the AI assistant I work with. The first rule is that it has to look at every image on every slide, not just read the text. My image-only slides often carry the most important content. Early on, it skimmed a Three Branches deck and completely missed a recipe analogy I use, where the legislative branch writes the recipe, the executive branch cooks it, and the judicial branch tastes it. That analogy lived entirely in the pictures.

We talk before we build. Before anything gets written, we go over the week's topics together. I explain what to emphasize, what to go light on, and which analogies matter. Sometimes I walk through slides one image at a time, especially slides with animations that a static read can't capture.

One reading becomes the foundation. Claude writes a complete classwork reading that covers the whole week. It isn't really meant to be read start to finish. Its real job is to feed the next step, so it has to be thorough enough for Brisk, an AI quiz tool, to build 20 good questions from it. That's the actual quality bar.

The reading becomes the classwork and the quiz. Brisk turns the reading into a 20-question Google Form that's assigned as classwork during the week. For Friday's quiz, I duplicate that form and modify it, so the quiz covers the same material without being an exact copy of the classwork.

Notes are a separate study aid. Claude also writes a shorter set of notes from the same content. They're always complete notes, never fill-in-the-blank.

Bell work prepares for each lesson. Each lesson day gets a one-paragraph bell work reading that prepares students for that day's topic. Brisk then turns each bell work reading into a short multiple-choice Google Form to check understanding before the lesson starts. After the quiz is finalized, a Thursday bell work paragraph previews what to focus on before Friday.

The quiz becomes the review games. Once the quiz is done, we build two spreadsheets from it. The Gimkit version keeps the full questions and answers. The Kahoot version shortens everything to fit Kahoot's character limits, so nothing gets cut off on screen. We also build a one-slide review that covers every quiz question, which goes up right before Thursday's Kahoot game.

AI files everything in Google Drive. Claude is connected to my Google Drive, so it doesn't just write the materials. It creates the documents and saves them directly into the right folders. Bell work and review games each have a standing folder that we reuse every week, and lesson materials go into a folder for the current unit. New folders only come in when a new unit starts. I never copy and paste, rename files, or hunt for anything, because everything is already where it belongs.

Custom images fill the slides. For slides that need illustrations, Claude writes detailed prompts for Google's Nano Banana image tool, and I run them. We use the same style description at the start of every prompt so a whole deck looks like one artist drew it. We also learned to write rules into every prompt to prevent extra arms and legs, which was the most common glitch before we fixed it.

None of this replaces the teaching. My slides, my analogies, and my decisions about what matters are still the starting point. What changed is how fast that turns into a full week of materials that all line up with each other.

In upcoming posts, I'll walk through each step in detail, including the prompts and formats we use.