AI suggestions feel finished because they arrive quickly and fluently. That is exactly why a deliberate approval loop matters. It creates a pause between possibility and canon.
The Five Steps
- Retrieve: select the relevant approved facts.
- Generate: ask for a limited set of ideas, questions or alternatives.
- Check: compare each result against rules, relationships and the current plot.
- Decide: approve, revise, defer or reject.
- Record: update the connected story bible only with the decision.
The workflow is simple enough to use after every important writing session. Its value comes from consistency.
Distinguish an Idea from a Fact
An AI may say that a character once served in a border war. Until you decide to use that history, it is not a fact. Mark it as a proposal. If you approve it, connect it to the character record, timeline and any locations or relationships it changes.
This protects continuity and gives you a clean audit trail of your creative reasoning. It also makes it easier to revisit a rejected path without accidentally treating it as established lore.
Review for More than Contradiction
Ask three review questions:
- Is this consistent with what is already true?
- Is it interesting enough to earn its complexity?
- Does it create a useful future consequence?
The last question prevents lore from becoming decorative accumulation. The strongest additions change a relationship, create a future choice, or deepen the meaning of an existing event.
Saganode’s public guidance similarly frames creators as the reviewers of extracted facts. The broader lesson is clear: AI output is raw material; authorial judgment makes it part of a story.
Define What Requires Approval
Not every generated line needs a formal review. Set a threshold. A private brainstorming note can remain temporary; a fact that changes a character, world rule, timeline, location, visual reference or future scene must be reviewed before it becomes canon.
This distinction keeps the creative process fast without making it careless. It also prevents a common continuity error: a detail appears in one generated scene, looks attractive, and is repeated later even though nobody consciously chose it.
Make Decisions Visible
Use four simple labels in your notes:
| Label | Use it when |
|---|---|
| Approved | The fact is now part of the world. |
| Revise | The idea is promising but needs a different formulation or consequence. |
| Deferred | The question is useful, but the story is not ready to decide. |
| Rejected | The idea should not quietly return in a later prompt. |
Add a short reason beside significant decisions. “Rejected: removes the cost of crossing the storm belt” is more useful than a silent deletion because it protects the dramatic purpose of the rule.
Review with the Right People
For a solo creator, the reviewer is simply the author at a deliberate distance from the prompt. For a team, decide who can approve each category: a writer may own character voice, a director may own visual canon, and a producer may own practical constraints. The workflow should make authority clear before a project has dozens of conflicting versions.
A Weekly Continuity Pass
Once a week, review the newly approved entries together. Check for duplicate facts, unresolved contradictions and decisions that require a scene or visual asset to be revised. This is faster than discovering a conflict after a whole sequence has been written or generated.
Example: Promote a Proposal to Canon
Imagine an AI session proposes that the city’s memory archive can erase a person’s public identity. Do not add it to the story bible immediately. First ask whether it fits existing rules, who has access, what it costs and which current storyline it changes. The author might approve a narrower decision instead: the archive can alter public records, but only through a ritual that leaves visible marks on the operator.
Record the final rule, the rejected alternative and the reason for the restriction. The next prompt can now explore scenes shaped by the approved cost, while collaborators can see why the broader version is not part of the world. This is what turns fast generation into durable creative knowledge.
The same pattern works for characters, locations and visual references: propose, compare against canon, approve a precise version, then record the consequence. The pause is brief, but it keeps the project legible as it grows.
Approval-Loop Checklist
- Is this output a suggestion or an approved fact?
- Which existing record does it alter?
- Does it create a consequence worth carrying forward?
- Has the decision and its reason been recorded?
- Can a collaborator tell which version is current?
The loop protects the human centre of the work. It lets AI remain fast and generative while meaning, continuity and responsibility remain with the creator.