# Prompt: research notes → executive report (copy and fill)

Use this at step 4 of the
[Research to Executive Report workflow](/en/workflows/research-to-executive-report/).
Fill the three bracketed blocks before sending. The prompt embeds the report
skeleton's headings verbatim, forbids invented evidence, and requires the
model to mark what it did NOT find — the three controls that keep an
AI-drafted report honest enough to edit.

---

```text
You are drafting an executive report from my research notes.

MATERIAL (pasted below, between <notes> and </notes>):
<notes>
[PASTE your research notes here — excerpts, source URLs, margin notes,
open questions. Keep them as-is; do not clean up first.]
</notes>

AUDIENCE: [who reads this, and what they control — e.g. "engineering
leadership; owns the tooling budget"]

GOAL: [the decision or action this report exists to obtain — one sentence]

Rules:
1. Use these section headings, verbatim and in this order:
   "## 1. Bottom line (≤ 120 words)" · "## 2. What we decided to ask (and
   what we did not)" · "## 3. Evidence" · "## 4. What we verified ourselves
   (and how)" · "## 5. Risks and failure conditions" · "## 6. Options" ·
   "## 7. Recommendation and next step" · "## Appendix".
2. Evidence table: one row per factual claim in sections 1, 5, and 7, each
   with its source link from my notes. If a claim has no source, either
   drop it or record it in the Appendix under "Open questions" — never
   present an inference as a finding.
3. Confidence column: say WHY (measured / single source / vendor claim /
   inference), not just high/medium/low.
4. Section 4 must include a "what we did NOT verify" list. Do not paper
   over gaps; if a gap cannot be found, say what was checked instead.
5. Do not invent numbers, names, dates, or URLs. If the notes contradict
   each other, surface the contradiction in section 5 as a risk.
6. Options table must include a reversibility and a portability note:
   what happens to the artifacts if we stop or switch tools.
7. Section 7 ends with the single next action, an owner role, and a date
   field. The Appendix follows it.

Output: the full report in Markdown, headings as in rule 1.
```

---

## Notes on using it

- Paste, don't summarize: the messiness of the notes is signal (gaps show up
  as empty evidence rows, which is exactly what step 5 of the workflow
  checks).
- After the first draft, re-send with: "Revise section 1 only. It must
  survive being read alone." — a cheap second pass that reliably tightens
  the BLUF.
- The prompt is tool-agnostic. We test it inside AtomStorm Studio (the CTA
  below pre-fills it with this workflow's skeleton); it works in any
  chat-style assistant, with the usual caveat that you, not the model, own
  every claim it prints.