Name the blockage as a decision
Replace “What happens next?” with a question that has pressure in it: “What can Leena do here that protects her brother but makes the investigation harder?” or “Which reveal changes the negotiation without ending the mystery?”
A constrained question gives you choices you can compare. It also keeps an assistant from filling the page with a generic continuation simply because no sharper problem was supplied.
Assemble a small context packet
Include the active passage, the immediate scene goal, and only the story facts that constrain the answer. For a character decision, that may be what the character wants, what they know, what they refuse to do, and the consequence already waiting outside the scene.
Separate manuscript facts from possibilities. Label an unresolved note as a possibility so it is not treated as established canon.
Ask for options and consequences, not prose
Request three materially different moves. For each, ask what becomes inevitable, who gains or loses agency, what information reaches the reader, and which later scene would need to change.
Discard options that depend on invented facts or make the character act against the evidence on the page. The point is to widen the decision space, not outsource the decision.
Choose the pressure, then write the line
Select the consequence that best serves the story and write the next beat yourself. This preserves your syntax, timing, and degree of explanation while still benefiting from the comparison.
If you request a continuation, treat it as a disposable probe. Read it beside the current passage, keep any useful discovery, and reject the prose if it sounds more finished than true.
Verify the choice against the story
After drafting, check the new beat against character knowledge, timeline order, and the release version of neighboring chapters. Record any follow-up as a task rather than interrupting the scene to solve every downstream detail.
You can practice this loop in the local demo: write a short passage, move to Think to frame the decision, then compare the result in Review. The demo is a bounded example; it does not promise a complete editorial assessment.