Can I trust it to catch every decision?

No — treat it as a first pass that is auditable, not as the record. The quotes exist so that checking it takes a minute instead of re-reading the whole thread.

Should I paste a thread with private details in it?

Strip names, phone numbers and anything financial first, or replace them with placeholders like [TEACHER]. Assume anything you paste into a hosted tool may be retained.

Writing & content

Decision memo from a thread

Turn a 40-message email or WhatsApp thread into a one-page memo: what was decided, who owns what by when, and what is still unresolved.

Use this when a long thread has gone quiet and nobody is sure what was actually agreed.

The pattern

Copy this and replace everything in [SQUARE BRACKETS] with your own specifics.

Prompt template

You are the person who has to make this thread actionable.

Context: below is a [EMAIL / WHATSAPP / SLACK] thread between [WHO, and their roles]. The decision that needs to come out of it is [WHAT DECISION].

[PASTE THE THREAD, oldest message first.]

Task: turn this into a decision memo.

Format:
- Decided: only things someone explicitly agreed to. Quote the message it came from.
- Owners: one line each — person, what they own, by when. Write "no date agreed" where no date was given.
- Open questions: what still blocks the decision, and who can answer it.
- Not decided: anything that was discussed and left hanging.

Rules: do not infer an owner from who talked most. Do not invent dates. If the thread never settled something, it belongs under Open questions or Not decided, never under Decided.

Why each part is there

The decision that needs to come out of it: A thread contains dozens of side topics. Naming the decision gives the model a filter, so the memo is about one thing instead of everything.
Quote the message it came from: The single most important line here. It makes every “Decided” item auditable against the thread in about five seconds, and it makes fabrication obvious.
The four fixed sections: Decided / Owners / Open / Not decided is a shape that forces the ambiguity into the open instead of letting it get smoothed into confident prose.
Do not infer an owner from who talked most: This is the specific failure mode of thread summaries: the loudest participant gets assigned everything. Naming the failure prevents it.

Before and after

The lazy version

Summarise this email thread.

What goes wrong: You get a chronological retelling — “Rahul then raised concerns about timing” — which is exactly as hard to act on as the thread was. Decisions, guesses and speculation all sound equally settled.

The pattern, filled in

Worked example

You are the person who has to make this thread actionable.

Context: below is an email thread between our science teacher (Ms Nair), the lab in-charge (Ravi) and the vice principal. The decision that needs to come out of it is whether the Class 10 science exhibition happens on 12 or 19 October.

[thread pasted, oldest first]

Task: turn this into a decision memo.

Format:
- Decided: only things someone explicitly agreed to. Quote the message it came from.
- Owners: person, what they own, by when. Write "no date agreed" where none was given.
- Open questions: what still blocks the decision, and who can answer it.
- Not decided: discussed and left hanging.

Rules: do not infer an owner from who talked most. Do not invent dates. If the thread never settled something, it belongs under Open questions or Not decided, never under Decided.

What that buys you: One page. Two real decisions with quotes attached, three owners with one honest “no date agreed”, and the actual blocker — nobody has confirmed lab availability on the 19th — sitting under Open questions where somebody can see it.

Now do it on your own work

Find a real thread of 15+ messages you were part of (a group project, a club, a family plan). Run the pattern. Then check every “Decided” line against the thread yourself and count how many you can verify from the quote alone.

Check yourself — every box should be true

Every Decided item carries a quote you can find in the original thread.
At least one item correctly landed under Open questions or Not decided rather than Decided.
No owner was assigned to somebody who never agreed to it.
You found at least one thing the memo got wrong, and you can say which rule would have prevented it.

Then explain it back

The memo separated what was decided from what was merely discussed. Why is that distinction so easy for a summary to lose, and what in the prompt kept it?

Write two or three sentences in your own words. If you cannot, you copied a prompt rather than learning a pattern — which is the whole difference this library is trying to make.

Does this depend on which AI tool I use?

The constraint here is context length, not tool choice: a very long thread may not fit in one prompt on a smaller-window tool. When that happens, memo each chunk separately with the same format, then merge the memos — do not ask for a summary of summaries, which loses the quotes.

Why it actually works

Notice what made this trustworthy: every claim was tied back to a source you could check. That is precisely the idea behind retrieval-augmented generation.

Why grounded answers beat fluent ones →

Prompting well is a skill you can pick up in a week. Understanding what is happening underneath is the 18-month curriculum — and it is what stays useful when the tools change.

FAQ

Can I trust it to catch every decision?: No — treat it as a first pass that is auditable, not as the record. The quotes exist so that checking it takes a minute instead of re-reading the whole thread.
Should I paste a thread with private details in it?: Strip names, phone numbers and anything financial first, or replace them with placeholders like [TEACHER]. Assume anything you paste into a hosted tool may be retained.

Next patterns

The Prompt Patterns Library is a separate, optional track — it does not change the 78-week curriculum or your roadmap. See what the curriculum actually covers →