Worldbuilding Wednesdays Ep. 3: First Note and Geography

In Episode 1, I created The Slow Death of Magic from scratch. In Episode 2, I gave it structure: folders, reference material, and a campaign library loaded with the Mage "How do you DO that" guide so the AI would stop guessing about my world's magic system and start actually knowing it.

Now the pieces are in place. The AI has context. The folders are waiting to be filled. This week, we fill them.

What Happens in Episode 3?

This is the episode where I actually sit down with the AI assistant and build something together. I sit down with it and work through two big creative problems: what is the inciting incident of this campaign, and what does the world look like geographically?

The whole session runs about 19 minutes. By the end, I have a specific core conflict that goes way beyond "magic is dying," a naming language for the world, and five regions that each connect back to that central premise. Not bad for a Tuesday night.

From Vague Concept to Inciting Incident

"Magic is dying" is a mood. A vibe. It's the kind of idea that gets you excited in the shower and then stalls completely when you sit down to actually write something. I've been there with every campaign I've ever run. The concept is easy. The specifics are where the work lives.

So I asked the AI a direct question: why is magic dying?

What came back turned into an actual conversation. The AI pulled from the Mage sourcebook material in my campaign library and started proposing mechanisms that fit the world's metaphysics. Not generic fantasy explanations like "a dark wizard cast a spell" or "the gods withdrew their power." Answers rooted in how Mage actually works: consensual reality, the Spheres, the Tapestry (yes, that's an actual Mage term, not an AI buzzword) of magical energy that connects everything.

Through a few rounds of back-and-forth, the concept sharpened. "Magic is dying" became something much more specific: a broken bargain. An ancient agreement between factions that maintained the balance of magical power, and someone violated it. Now the Tapestry is unraveling, and the factions that once cooperated are pointing fingers at each other while scrambling to protect what's left of their power.

That's a campaign I can run. That gives me factions with competing motives, a mystery to uncover, an antagonist (or several), and a ticking clock. All from a 10-minute conversation with the AI.

Here's the part that matters for GMs running any system: the quality of the AI's suggestions scaled directly with the reference material I'd already uploaded. In Episode 2, I fed it the "How do you DO that" guide. That investment paid off here. The AI wasn't guessing about how magic works in my setting. It knew the vocabulary, the mechanics, the faction politics. When it proposed "a broken bargain," it could ground that in the existing lore about Tradition-Technocracy tensions, Paradox, and the way belief shapes reality in Mage.

If I'd skipped the library upload and just typed "magic is dying, help me brainstorm," I would have gotten serviceable fantasy ideas. Fine for a quick D&D one-shot. Not fine for a campaign that needs to feel specific and lived-in.

Building a World's Geography

With the inciting incident established, I moved to geography. This is where I wanted to push the AI a bit harder.

I didn't just ask for "five fantasy regions." I asked it to start with language, to create a thematic linguistic framework for the world and then use that framework to name the regions. The idea was that the geography shouldn't feel like five random biomes stapled together. It should feel like a world where the places grew out of the same cultural and magical soil.

I switched to the Powerful model for this part. More on model tiers in a minute, but the short version is that when you're asking the AI to do genuinely creative work with multiple constraints (maintain the linguistic consistency, tie each region to the core premise, keep it grounded in Mage-adjacent metaphysics), the more capable model earns its keep.

What I got back surprised me. The AI proposed a naming system derived from the world's relationship with magic. The way places were named reflected whether magic was strong, fading, or already gone in that area. Then it generated five regions, each with its own identity tied to the central conflict. One region where magic still flows freely and the old Traditions hold power. Another where the Technocratic paradigm has nearly erased ambient magic entirely. A borderland where the two worldviews collide. A dead zone where the broken bargain hit hardest. And a hidden place that might hold the key to understanding what went wrong.

Every region has built-in story hooks. Every region connects to the inciting incident. None of them exist in isolation. That's the difference between asking an AI for a list and having a conversation where each answer builds on the last.

For GMs not running Mage: this same approach works for any setting. Ask the AI to create a naming convention first, then generate locations using it. You'll get a world that feels coherent rather than assembled from random generators. Whether you're building a D&D continent, a Blades in the Dark city, or a Mothership space station, linguistic consistency is one of those subtle details that makes a setting feel real.

Teaching Notes Back to the AI

Here's where the workflow clicks into something bigger than a single brainstorming session.

After generating the inciting incident and the geography, I didn't just leave those ideas sitting in the chat. I wrote them up as proper notes in ScriptoriumGM, one for the core conflict and one for the geography overview. Then I added those notes to the campaign library so the AI could reference them in future conversations.

This is the cycle that makes the whole system work. You talk to the AI. You get ideas. You refine them into notes. You teach those notes back to the AI. Next time you ask it something, it has more context. The RAG-powered knowledge base indexes each new note, so when I sit down next week and say "Help me flesh out the dead zone region," the AI already knows what the dead zone is, why it exists, and how it connects to the broken bargain.

Think of it as a feedback loop. Every session of brainstorming makes the next session better, because the AI's understanding of your world gets deeper each time. After 10 or 15 notes, the AI knows your setting well enough to catch inconsistencies, suggest connections you hadn't considered, and generate content that sounds like it belongs in your world rather than a generic fantasy supplement.

I've run campaigns for over 12 years, and the closest analog I can think of is playing with a co-GM who's read all your notes. You don't have to re-explain the premise every time. You can just say "I need a new NPC for the borderlands region" and get someone who makes sense there, because the AI already knows what the borderlands are and what kind of people live in them.

The key habit to build: after every AI conversation that produces good material, take a few minutes to clean it up and add it as a note. Don't let good ideas live only in chat history. Promote them to the campaign library where they become permanent context.

AI Model Tiers and Session Approvals

This episode is a good time to talk about the AI model options, since I switched between them on camera.

ScriptoriumGM offers three model tiers:

Fast is the lightweight model. Quick responses, lower token cost. Good for simple questions, quick NPC names, short descriptions, and anything where you need speed more than depth. I use it for the kind of thing I'd otherwise Google: "give me six tavern names" or "what's a good quirk for a merchant NPC."

Balanced is the default. It handles most brainstorming, note generation, and creative work well. This is where I did the inciting incident conversation. It pulled from the campaign library accurately and made connections between the uploaded Mage material and my campaign concept. For 90% of worldbuilding conversations, this is the right choice.

Powerful is the heaviest model. I switched to this for the geography and linguistic framework work because I was asking the AI to juggle multiple constraints simultaneously: maintain a naming language, tie five regions to a central premise, keep everything consistent with Mage's metaphysics, and make each region feel like its own place. That's a lot to hold in working memory at once, and the more capable model handled it noticeably better. The tradeoff is it uses more of your monthly token budget.

One thing you'll see in the episode: session approvals. When the AI needs to use tools like searching the campaign library, it asks for permission. You'll see a small approval prompt pop up during the conversation. This is intentional. The AI won't silently search your files or perform actions without you knowing. You stay in control of what it accesses and when.

For longer brainstorming sessions like this one, you'll tap "approve" a few times as the AI pulls relevant context from your uploaded documents. It becomes second nature after a couple of conversations. Think of it like a collaborator saying "mind if I check your notes?" before flipping through your binder.

Why This Matters for Your Campaign

I want to zoom out from the Mage-specific details for a second, because the workflow in this episode applies to every game and every setting.

The pattern is:

1. Start with a vague concept ("magic is dying," "the empire is crumbling," "something is wrong on the station")
2. Use the AI to sharpen it into something specific, drawing from whatever reference material you've uploaded
3. Push for connected details. Don't brainstorm in isolation; ask for things that tie back to your core premise
4. Write up the good ideas as proper notes
5. Add those notes to the campaign library
6. Repeat, with each cycle building on the last

That's it. No special technique. No prompt engineering wizardry. Just a conversation where you treat the AI like a creative partner who's read your source material, and a habit of saving the best results back into the system.

After three episodes, The Slow Death of Magic has gone from a sentence to a campaign with a specific inciting incident, a world with five regions, a naming language, and an AI assistant that knows all of it. Total time invested across all three episodes: maybe 45 minutes of actual work, spread across three sessions. That's less time than I used to spend on session prep for a single session, and I haven't even started prepping sessions yet.

What's Coming in Episode 4?

Next week, I'm going narrow instead of wide. I'll pick one of the five regions and build it out in detail: locations within it, key NPCs, local conflicts, the kind of granular detail you actually need when your players walk into a new area.

I'll also show how to focus the AI's attention on individual notes rather than the entire campaign library. When you're developing a specific location, you don't need the AI thinking about every region at once. You want it locked in on the area you're building. That focus makes a real difference in the quality and relevance of what comes back.

Subscribe to the ScriptoriumGM YouTube channel so you don't miss it. New episodes every Wednesday.

Try It Yourself

If you've been following along, this is the episode where things get fun. You have your campaign. You have your folders. You have reference material in the library. Now go talk to the AI.

Start with your campaign's central question. Whatever the "why" is behind your setting. "Why did the gods go silent?" "Who killed the last king?" "What happened to the colony ship's original crew?" Ask the AI. Push back on the first answer. Ask follow-up questions. Let it surprise you.

Then take whatever you liked from that conversation and write it down as a note. Add it to the library. You just made your next brainstorming session better.

Fifteen minutes. One conversation. Your campaign will be sharper for it.

Watch Episode 3 on YouTube

---

Get Involved

The Slow Death of Magic is shaping up, but I want to see what you're building too. Share your brainstorming conversations, your geography notes, your AI-generated campaign concepts. Some of the best ideas come from seeing how other GMs tackle the same creative problems from completely different angles.

Join us in the ScriptoriumGM Discord and show us what you're working on. Subscribe to ScriptoriumGM on YouTube for weekly episodes. And if you haven't tried ScriptoriumGM yet, create a free account and start building. The getting started guide walks you through everything, or follow along with the videos.

See you next Wednesday.

---

Sources

1. Mage: The Ascension 20th Anniversary Edition - Onyx Path Publishing's definitive edition of the Mage the Ascension system, the core inspiration for The Slow Death of Magic campaign.

2. How Do You DO That? - Mage: The Ascension - The practical Mage supplement uploaded to the campaign library in Episode 2, covering Sphere-based magic and its applications.

3. ScriptoriumGM YouTube Channel - Home of Worldbuilding Wednesdays and other ScriptoriumGM tutorials.

4. How RAG-Powered Knowledge Base Saves Your Campaign Memory - Explanation of the RAG system that powers the campaign library's document indexing and retrieval.