Building Community Worlds: Tips from EchoQuest Creators
EchoQuest's community library exists for one reason: people want to share the worlds they dream up. I built the World Builder Wizard and the publishing flow, and I wrote the prebuilt worlds that sit in the library myself, so I've spent a lot of hours watching what the AI Game Master does with different kinds of input. Some worlds come alive the moment the narrator opens its mouth. Others sound like a brochure. This post is my own playbook for landing in the first group, based on how the tools work under the hood and on what I learned writing those nine worlds.
A quick map of the flow first. On the Storyteller plan and up, you can build a private world in a few ways. Quick Build asks you four questions and lets Claude fill in the rest. The World Builder Wizard walks you through ten spoken questions, with Claude suggesting ideas at each step. You can also paste old campaign notes, or upload a Game Bible as a PDF, DOCX, TXT, MD or JSON file. On the Creator plan, a Publish button on your My Worlds page puts the world into the library's Community tab, and you can unpublish it whenever you like.
None of the advice below is mandatory, by the way. Some of the best worlds I can picture would break every rule here. They'd break them deliberately, though, for a reason, and that's the gap between an intentional choice and an accidental rough edge. So treat these as defaults to depart from on purpose.
Start With the Opening Scene, Not the Lore
When you build a world, the instinct is to start with history: the ancient wars, the founding myths, a timeline of great events, the cosmology of the gods. Resist it. Players don't experience your world through its history. They meet it through one specific moment, the opening scene, and that's the only paragraph guaranteed to land. Everything else is optional.
So build the opening first. Who is the player character? Where are they standing, and what's happening right now? What's the first decision they have to make? A grounded, specific opening does more for a player's engagement than pages of backstory. The wizard is built around this idea, actually. Of its ten questions, only four are required, and "Describe the opening scene in one or two sentences" is one of them. You can always add lore later. Honestly, lore hits harder when it arrives as an answer to something the player asked, in pieces shaped by what they wanted to know.
Here's a rule I lean on. If the first paragraph of your opening could appear in some completely different world, it isn't specific enough. The opening should announce this world and no other. When I wrote Saltbound, the scene starts with the first mate offering the newly elected captain "a battered tricorne that fit the previous captain badly and will fit the player no better." That hat couldn't wander into a cyberpunk city. Then read your opening as a stranger would, someone who's never heard of your setting. Do they know where they are? Do they want to know what happens next? If yes, ship it. If not, rewrite.
Write for Listening, Not Reading
Every word of a community world gets narrated aloud, either by the browser's built-in voice on the free plan or by ElevenLabs voices on Storyteller and up. That changes how you should write. Prose that reads beautifully on the page often goes flat when spoken, and rhythms that sing in audio can look odd in print. NPR's training team, who know a thing or two about this, put it bluntly: for the ear, sentences need to be far shorter and information needs to unfurl more slowly, and our ears can only handle about one fact or idea per sentence. So play to the medium:
- Use shorter sentences than you would in prose fiction. A narrating voice can't skip ahead the way your eye does, so a long sentence has to be held in memory until it finally resolves
- Avoid nested clauses that are hard to untangle by ear. NPR's own example swaps "My dad, who spent 40 years teaching physics, is a smart guy" for "My dad taught physics for 40 years. He's a smart guy." Apply the same surgery to any sentence with three sub-clauses and you'll get two clean sentences
- Favour concrete sensory detail over abstract description. "The smell of woodsmoke and roasted onions" beats "the homey smell" every single time. EchoQuest's GM is told to open each scene on one concrete sensory detail, so give it good raw material
- Read your opening aloud before you publish. If it sounds awkward spoken, rewrite it. This is the single highest-leverage editing pass a creator has. Better still, play your world privately first: when the wizard finishes, it drops you straight into character creation, so you'll hear your own words in the narrator's mouth within a minute
- Watch out for names that are easy to read and hard to say. Aeryndel-Tael-Khorin will trip up every text-to-speech engine ever built, so consider a shorter alternative or add a note on how it's pronounced
I've spent more hours than I'd like listening to synthetic voices read EchoQuest text. Chrome's built-in voice used to cut off after roughly 15 seconds, and premium voices pitched up like chipmunks whenever playback sped up, until I fixed both. If there's one thing I'm sure of after all that, it's this: text that sounds good spoken survives every engine. The best world descriptions have a rhythm when read aloud, and that's no accident.
Give the AI Specific Constraints
The AI Game Master is powerful, but it needs guardrails to stay faithful to your world. The clearer and more specific your constraints, the better it plays. Anthropic's own prompting guide suggests thinking of Claude as "a brilliant but new employee" who lacks context on your norms, and it notes that explaining why a rule exists helps Claude apply it to cases you never mentioned. Your world is exactly that kind of context.
Vague: "Magic is limited in this world."
Specific: "Magic requires spoken incantations and physical components. It is rare, feared and tied to the heretical old religion. No character casts magic publicly. The Church executes practitioners. Petitioners sometimes claim to have witnessed miracles; these are usually frauds, sometimes mass hysteria, and rarely something more disturbing."
Specific constraints let the GM make confident, consistent calls whenever something magic-adjacent comes up in play. They also shut out the generic-fantasy defaults the model would otherwise drift into. Left alone, it's perfectly happy to produce "a wizard with a staff casts a spell". Tell it magic is illegal and feared, and you get something far more interesting.
The wizard gives you one slot for this: "Name one hard rule of this world the game master must respect." Its examples are no firearms, magic always costs blood, and the dead never speak. Whatever you type there reaches the GM labelled as a hard rule it must never break, so spend it on the rule generic defaults are most likely to trample. Need more than one? A Game Bible has room for every mechanical rule you care to write, and the GM's core instructions say plainly that the world's rules come first. In Saltbound, for instance, I wrote that sea combat is gun-and-board and that "Most engagements end in chase or surrender, not slaughter." I also wrote that the hedge-witches "Take payment in fresh fish, not coin." Tiny lines like those steer whole scenes.
The same principle covers technology and social structure, religion and economics. The further your world drifts from the generic defaults, the more specific you have to be. A world with no kings needs to say so outright. A world with universal literacy needs to say so too. And if commerce isn't conducted in coins, spell out what people trade instead.
Design for Replayability
A community world gets played by many different people with many different approaches. Some players are aggressive and others cautious. Some chase politics while others want action, and you'll get suspicious players and trusting ones, rule-followers and chaos merchants. Your opening should welcome any of those starting energies and let each player play to their preference.
The strongest campaigns have a central tension that creates interesting choices whatever the approach. The approach changes which choices are available and what they cost, but the core tension holds. Take a campaign about a city on the verge of revolution. It plays differently for a sympathetic player than for a counter-revolutionary, and differently again for someone who just wants to keep their head down. Yet every one of them finds something worth doing, because the tension doesn't depend on which side they pick. Saltbound does something similar on a smaller scale. The new captain's navigator is a week overdue. Meanwhile a rival captain is waiting in the cabin to talk, and somebody has spotted a Crown frigate three islands east. A bold player chases one thread and a careful one chases another, and nobody's left standing around.
If your campaign only works when the player pursues one specific goal in one specific way, it isn't built for replayability. It's built for a single play-through. That's a perfectly legitimate choice (some short, focused campaigns make great one-shots), but say so in your description. When you publish, the library listing uses your world's title and description, with its genre and tone as tags, so that description is your one chance to tell players what they're getting into.
Let the AI Improvise
Some creators try to script every outcome. They pre-write branches, anticipate every possible choice and attempt to steer the GM's replies in detail. With an AI GM, that simply doesn't work. The model improvises by nature, and forcing it into a pre-written script produces stilted, generic output. So stop trying to control what happens and focus on the furniture: who the characters are and what they want, how the world feels and what's at stake. The GM will fill in the rest. It's even written into its instructions: "Follow the player's creativity instead of forcing them back onto a path."
Tabletop GMs worked this out long before AI came along. Justin Alexander's well-known essay Don't Prep Plots draws the line neatly: a plot is a sequence of events, while a situation is merely a set of circumstances whose outcome depends on what the player characters do. He also points out that when a plotted outcome doesn't happen, the GM is left to railroad the players back onto the tracks. That's exactly the trap.
Look inside any of the nine prebuilt worlds and you won't find a plot. You'll find locations, people with wants and grudges, factions, a few rules, an opening scenario and a short list of true real-world details for the GM to weave in when they fit. That's the furniture. A world with detailed factions, rich NPCs, vivid sensory texture and clear constraints produces wildly different sessions for different players, and each one feels like a real story in that world. A world with a pre-scripted plot feels like a railroaded video game, even with an AI doing the narration.
Iterate After Publication
Your first version is a draft. Publish it, watch how people actually play it, and revise. On the Creator plan, each of your worlds on the My Worlds page shows its session count, total turns and unique players. You can also open a 30-day engagement trend of sessions started and player turns per day (it's a little chart, with a hidden data table underneath for screen reader users). Those numbers won't tell you which scene bores people, so read them with some imagination. Lots of sessions with very few turns each, for example, makes me suspect the opening isn't pulling people in. Then revise: tighten the opening, sharpen a rule, rename the city nobody can pronounce. I firmly believe a creator who treats a world as a living document will end up with a better one than a creator who publishes once and walks away.
Don't use any of this as a reason to delay publishing, though. Eric Raymond's old open-source maxim from The Cathedral and the Bazaar applies here almost word for word: "Release early. Release often. And listen to your customers." Put a rough version out early and revise based on what you learn. Over six months, that produces a stronger world than polishing a perfect version in private for a year. I'd know, because EchoQuest itself got better precisely that way.
Be Generous With Other Creators
The library thrives on a culture of mutual help. Creators who play other people's worlds, leave thoughtful feedback and credit their influences openly tend to get the same in return. I even built a tool for it. Every community world can be forked: you play the original, then build your own version with Quick Build, the wizard or a Game Bible, and publish the remix. The wizard keeps a note at the top saying which world you're forking from, and I'd encourage you to name the original in your description too. Nothing would please me more than watching a whole little genre grow out of one creator's idea, with other creators building on it. So what's the opening line of yours going to be?
The library gets richer when everyone treats it as a shared garden rather than a competition.