A good creative prompt is not a wish list. It is a focused brief: the relevant facts, a bounded task and a definition of what the response must not decide on your behalf.
The Four-Part Packet
Use this compact format whenever the output will affect your story:
- Context: the approved facts the task depends on.
- Task: one specific question, draft or comparison.
- Constraints: details the output must preserve or avoid.
- Review rule: how to label assumptions, alternatives and contradictions.
For example:
Context: Mara cannot leave the harbour during the storm season; her brother controls the only relay key. Task: propose three scene objectives for their first confrontation. Constraints: do not introduce new factions or technology. Review rule: mark every inference that is not already canon.
Send Less Context, More Intentionally
The entire story bible is rarely useful in a single request. Select the records that govern this decision. A character scene needs the character’s current goal, the relationship history, the location rules and the latest plot event—not every map, poem and unused note in the archive.
This makes errors easier to spot. It also makes the prompt repeatable: another scene can use a different packet while relying on the same underlying source of truth.
Define the Output Shape
Ask for a table, a list of alternatives, a contradiction report or a scene brief. The output shape should match your next decision. If you need to choose between motives, a page of polished dialogue is less useful than a comparison that shows each motive’s consequence.
Context packets work especially well with the editorial roles described in AI as Story Editor, Not Ghostwriter . The packet supplies the shared reality; the role supplies the kind of pressure you want applied to it.
Match the Packet to the Decision
Different creative tasks need different context. A packet for a character decision should include current motivation, the relevant relationship and the pressure of the scene. A packet for a visual brief should include the location, fixed appearance details, time of day and the beat the image must communicate.
Sending every available note is rarely helpful. It makes important facts harder to see and increases the chance that a model treats an old idea as current canon. Select only what governs the immediate decision, then point to the story-bible record if more detail is needed later.
Reusable Prompt Template
Use this template as a starting point:
Role: Act as a continuity editor.
Approved context: [three to six short facts].
Task: [one decision or question].
Constraints: Preserve [fixed facts]; do not introduce [forbidden changes].
Output: A table with contradictions, assumptions and two alternatives.
Approval: Mark all proposed facts as unapproved.
The template creates a response that is easier to compare with your own notes. It also helps collaborators understand why a particular prompt was used and which facts were in force at the time.
Diagnose Weak Outputs
If an answer is generic, do not immediately add more adjectives. First check whether the task is genuinely specific. “Make the scene more dramatic” is not a decision. “Increase tension without changing the character’s established decision to protect her brother” is one.
If an output contradicts canon, verify whether the needed rule was included in the packet. If it was, simplify the request and ask the model to list its assumptions before it generates alternatives. If it was not, treat the mistake as a reminder to improve the retrieval process rather than as a failure of imagination.
Packet Review Questions
- Is every fact in the packet relevant to the task?
- Does the task ask for one kind of help at a time?
- Have you stated the constraints that cannot change?
- Will the output make the next human decision easier?
A context packet is successful when it makes AI assistance more precise and your own review more confident.
Build a Reusable Packet Library
Over time, save proven packets for recurring work: a character-conflict check, a location description, a scene-outline challenge or a visual continuity review. Keep the stable context separate from the task-specific request, so you can update one canon fact without copying an old mistake into every prompt. A small library also shows collaborators what “enough context” looks like for the project.
Name packets by their purpose rather than by a vague version number. “Mara / confrontation scene / approved after episode two” is much easier to retrieve than “prompt final 7.” Add the source records and decision date when a packet relies on changing canon.
The goal is not to turn creative work into form filling. It is to make useful context easy to reuse, leaving more attention for the decision only a human creator can make.