This “smart chain” turns a messy topic into crisp content by linking seven steps:

agent research → voice dictation → synthesis → examples → metaphor → polish → micro-test & iterate.

It includes ready prompts, QA gates, and deliverables so you can run it end-to-end.

0) Setup (one-time)

Tools

• AI research agent (e.g., Manus or similar) for live web research with citations
• Transcription: built-in OS dictation, Voice Memos → transcript,ChatGPT Voice, Otter, or Whisper
• Writing models: GPT or Claude
• Editor: Google Docs or Notion with comments/track changes

Reference files (optional)

• Voice_Guide.md — tone rules (short, active, plain; banned words)
• Gold_Samples.md — 3–5 short excerpts (150–300 words each) of your best writing

Step 1 — Let the agent do the heavy research

Goal: A compact research brief (facts, stats, debates) with sources and confidence.

Agent prompt

You are a research agent. Scan the web on: [TOPIC] for [AUDIENCE].
Return a brief in this JSON-like order:
{
 "key_questions": [5–8 real questions the audience asks],
 "statements_with_citations": [
   {"claim":"...", "number":"...", "unit":"...", "source_title":"...", "url":"...", "date":"YYYY-MM-DD", "confidence":0-1}
 ],
 "contradictions_or_debates": ["..."],
 "unknowns_or_data_gaps": ["..."],
 "quote_snippets": [{"speaker":"...", "quote":"...", "link":"..."}],
 "further_reading": [{"title":"...", "url":"..."}]
}
Rules:
- Prefer primary/official sources; 2022+ unless older is the best available.
- No paywalled PDFs unless necessary; summarize if paywalled.