Skip to content

Turn a numbers dump into a one-page decision memo

A crisp decision memo from a pile of analysis — the question, what the data says, what it pointedly does NOT say, a recommendation with a confidence level, and the line that would change your mind.

Works in any frontier model — ChatGPT, Claude, Gemini

The prompt

Turn the analysis below into a one-page decision memo for a busy reader who
won't see the underlying data.

Sections:
1. QUESTION — the single decision this analysis is meant to inform, in one line.
2. WHAT THE DATA SAYS — the findings that are actually supported, each as a
   bullet with the specific number(s) behind it.
3. WHAT THE DATA DOES NOT SAY — the limits: what's out of scope, confounded,
   too small a sample, or correlation-not-causation. Be honest about the gaps.
4. RECOMMENDATION — the call you'd make, in one or two sentences.
5. CONFIDENCE — High / Medium / Low, with a one-line reason.
6. WHAT WOULD CHANGE THIS — the specific result or new data point that would
   flip the recommendation. This line is mandatory.

Hard rules:
- Every number you cite must appear VERBATIM in the input. Do not round,
  recompute, extrapolate, or invent figures. If a number you'd want isn't in
  the input, say "not in the data" — don't estimate it.
- Do NOT overstate causation. If the data is correlational, say so in section 3.
- One page. No preamble.

Analysis / numbers:
"""
<paste your analysis, query results, chart figures, or data notes>
"""
Built-in guardrail: Every cited number must appear verbatim in the input (no rounding, recomputing, or estimating), missing figures are marked "not in the data," and a mandatory "what would change this" line keeps the recommendation falsifiable.

Then

Confirm the confidence level and the "what would change this" line match your own read — if the memo says High but the data can't support it, push it down and say why before this goes to a decision-maker.

Field-tested Jul 25, 2026. If it stops working, tell us.