From Idea to Canon: One Place, Three Decisions

An AI can hand you a hundred setting ideas before you have decided what one of them means. That abundance feels useful at first. Then your notes fill with beautiful possibilities, half-revised names, and details that contradict one another. You cannot tell which sentence is an option, which sentence is your correction, and which sentence the world is supposed to believe.

The difficult part of AI-assisted worldbuilding is not generating more. It is deciding what survives.

This article uses one fictional night harbor to make that decision visible. The harbor, its names, its customs, and every example in this article were invented for this exercise. They are not a user success story or a recovered setting. The method is the part to keep.

The three decisions

Treat each new piece of worldbuilding as three separate decisions rather than one block of prose:

The same paragraph can contain all three stages, but mixing them makes later work harder. A candidate should be allowed to be odd. A revision should be traceable to a question. Canon should be short enough that you can use it without rereading the whole conversation.

This is also why a strong worldbuilding note needs a decision attached to it. “The harbor has a strange light” is an idea. “The harbor light is a tax signal that changes color when a ship has paid” is a setting rule. The first invites exploration; the second creates consequences.

Start with a question narrow enough to answer

Do not begin by asking for an entire culture, history, or region. Begin with one place and one job it must perform. A narrow question gives you something to accept, adjust, or refuse.

For the example in this article, the starting question is:

> Propose one small harbor that stays busy after sunset. Give it a name, one practical service, and one custom that visitors notice. Keep the example fictional.

The request has a boundary. We are not asking for the harbor’s complete history, ten neighboring kingdoms, or an explanation of real maritime practice. We only need enough material to make the next decision.

A possible candidate is:

Nothing here is canon yet. The name may be too literal. The lamps may make the harbor too welcoming. The sentence left by each captain may be a useful social ritual or an unnecessary decoration. The candidate has done its job if it gives you something specific to react to.

One candidate is enough to begin

An easy mistake is to ask for five more versions because the first one is not perfect. That postpones the interesting work. Pick one candidate and find the part that does not fit your world.

Suppose you want the harbor to be valuable but suspicious. It must help ships arrive, yet it should not feel like a friendly tourist landmark. The custom is the part that can carry that tension. A captain’s sentence is not a greeting; it is a record that the harbor can read later.

Your next question can be precise:

> Keep the harbor’s name, location, and lamp service. Change it so the arrival sentence is an obligation that makes the harbor suspicious of strangers. Do not add a new faction yet.

The request protects three things and changes one. That is the point. If every revision replaces the whole idea, you cannot learn which decision improved it.

A revised candidate might be:

The harbor now has a pressure point. A visitor can comply, lie, refuse, or accidentally describe something that another captain already denied. The custom creates future scenes without requiring a complete history today.

The pronoun test: change it, do not restart it

When a candidate is already on the table, use a pronoun or a short reference to point at it. “Make it suspicious of strangers” says that the harbor remains the object under discussion. Repeating the whole name and premise is sometimes necessary, but it can also invite an accidental restart.

The useful shape is:

> Keep it recognizable. Change one property. State what must not move.

For the harbor, that means the name and practical service stay in place while the social meaning of the custom changes. You can do the same with a character, a law, a species, or a city:

• Keep the city’s trade route; make it dependent on a seasonal risk.
• Keep the character’s goal; make it costly for them to pursue.
• Keep the law’s wording; make the people who enforce it disagree about its boundary.

The pronoun is not a magic command. It is a reminder that the next turn is a revision of a candidate, not a request for a new setting. If the conversation loses the reference, restate the smallest necessary noun and continue.

Decide as the author

After the revision, stop generating and make three decisions. Write them down before moving on:

“Leave open” is not failure. It is a boundary around what the setting does not need yet. An unanswered origin can become useful later, while a forced answer can turn a small harbor into a pile of lore that has nowhere to go.

The Keep / Change / Leave open worksheet

Use this blank worksheet for one candidate at a time. Fill it from left to right only if that helps; starting with the decision you already know is also fine. The empty cells are intentional.

Now apply the worksheet to the Night Ledger. A completed row might say: keep the reef channel, change the harbor’s welcome into a test, and leave the lamp-maker open. You do not need to fill every row. The worksheet is successful when it gives you one usable rule and one honest unknown.

Put one decision into Worldseed

Worldseed is useful here because the candidate and the final setting do not have to be treated as the same thing. Start a world with the premise you want to examine, then use the chat to ask for one proposed element. Read the proposal as a candidate, not as an instruction.

When the conversation gives you an element card, make the author decision yourself:

1. Ask for one candidate with a clear function and one tension.
2. Point to the candidate and use “it” to change one property while keeping the rest recognizable.
3. Fill the Keep / Change / Leave open worksheet in your own notes.
4. Use the Add action only for the version you want to store as an element.

The chat can use recent successful exchanges to continue a thread while you stay in that chat. It is not a permanent notebook for an unchosen proposal: if you leave or reload, an unsaved conversation or proposal may no longer be available. The Add action is the author’s save decision. This separation lets you explore without turning every suggestion into a fact in your world.

When to leave a question open

Leave a question open when answering it would not change what you are trying to use right now. The harbor can guide ships and test their stories before you know whether its founder was a grieving navigator, a council, or nobody anyone remembers. The unknown is doing useful work: it gives you room to connect the harbor to another element later.

Do decide a question when it changes behavior. If the lamps are a public signal, decide who can read them. If the arrival sentence is an obligation, decide what happens when a captain refuses. If the harbor’s service depends on a tide, decide what the harbor does when the tide makes the route unsafe.

A good rule is: leave origins open, decide consequences. Origins can wait. Consequences are what make the setting move.

Three checks before canon

Before you store the final version, run three quick checks:

1. Recognition: Can you tell which candidate this revision came from?
2. Difference: Did the revision actually change one meaningful property?
3. Consequence: Can you name one person, place, or action that this rule affects?

If recognition fails, the conversation restarted instead of revising. If difference fails, you only paraphrased the candidate. If consequence fails, the detail may be decoration rather than worldbuilding. Any of the three results is useful information; it tells you what the next question should target.

A complete pass you can finish tonight

Choose one location in your world and give it one job. Ask for one candidate. Keep the name and job fixed, then change one social property with a short follow-up. Fill the worksheet. Store only the version whose consequence you can see, and leave one origin question open.

You now have more than a paragraph. You have a small piece of canon with a boundary, a pressure point, and a next question. Repeat the loop when you are ready, but keep each pass small enough that you can remember what you decided.

If you want to explore the same discipline through a different lens, the alternate-history guide follows one divergence into its consequences, while the map audit turns settlement placement into a political question.

Worldseed gives you a place to explore a possibility, revise it in conversation, and choose what belongs in your world. Start with one idea, keep one decision, and let the next question grow from what you actually kept.