Research signals → qualify the problem → define the point of view → assemble evidence/examples → outline the narrative job → draft → verify → edit for spoken delivery.
1. Research for a question, not for “ideas”
Collect candidate questions and examples from your audience, peers, comments and current content signals. The goal is to identify a problem worth resolving, not accumulate an infinite swipe file. For the upstream method, see how to find trending content ideas.
2. Qualify the problem
Keep the topic when you can answer three questions: who has the problem, what changes after consuming the content, and what evidence or example can make the answer credible?
3. Decide the angle before AI drafts
Write a one-sentence point of view. This stops the model from defaulting to a generic consensus summary. If your point of view is not clear enough to state, you are not ready to script.
4. Build an evidence packet
Give the drafting system only the facts, quotes, examples and constraints you are willing to stand behind. Mark uncertain claims. Do not ask the model to fill factual gaps with plausible-sounding material.
5. Outline the one job
A useful short script usually needs a tension, explanation, proof/example and completed next thought. It does not need seven artificial sections. The structure should be as long as the audience's question requires.
6. Draft for speech
Generate a first draft using short clauses, spoken transitions and concrete nouns. Then read it aloud. Sentences that look polished on a page can still be awkward to say.
7. Verify, then edit for voice
Check every factual statement against the evidence packet. Remove generic filler, repeated premises and invented specificity. Finally, restore your natural rhythm and examples.
8. Route the finished script into the larger system
The script is an output, not the end of the operating model. The broader AI content workflow for creators handles what happens before and after this chain.