Prompt PlaybookAUG 9, 20263 min read
What the thread does not say
One prompt that takes a long email chain and returns the facts it contains, who said them, what quietly changed, and the questions nobody has answered yet.
Keenan ChiassonAUG 9, 2026
By the time a request has been forwarded four times, the facts are spread across a dozen messages and two of them have been quietly overwritten. Re-reading the chain is not the hard part. The hard part is noticing what nobody ever said, because a missing number does not announce itself.
Below is an email thread. Do not summarize it. Give me four lists instead. One: every number, date, quantity and location stated in the thread, and for each one, who stated it and in which message. Two: anything that changed between the first message and the last, showing both versions. Three: decisions that were actually made, and who made them. Four: a list called Not Stated, for anything someone would need in order to act on this that nobody has said yet. If a value appears only as a guess or an estimate, it belongs in Not Stated. Do not fill in a value that is not in the text.
Why each line is there
- Do not summarize comes first because a summary is the one output that hides the problem. It reads just as smoothly whether the thread contains the facts or not.
- Attribution on every figure, because the same number carried by the customer and by someone's internal guess are not the same fact, and a paraphrase flattens them into one confident line.
- The change list catches the quiet revision: the date moved in message nine, the schedule was never re-sent, and everyone downstream is still working off message two.
- Not Stated is the whole point of the prompt. A model that helpfully fills the gap hands you a document that looks ready to act on, which is worse than an obviously incomplete one.
The Not Stated list is usually your next message. Send it back to whoever owns the request and you have asked every open question in one round instead of four.
