AI Prompt: Write a Decision Memo From an Email Chain

The ai prompt to write a decision memo from an email chain solves a specific problem: a decision was made over email, but anyone who joins the project later cannot reconstruct it without reading every message. Decisions made in threads are distributed across a dozen replies, buried in hedged language, and split between the person who proposed the idea, the person who pushed back, and the person who ultimately said yes. A decision memo collapses that into a structured document that anyone can read in two minutes and forward to someone who was not on the thread.

A decision memo is not the same as a thread summary. A summary tells you what happened. A memo records a specific slice of it: what was decided, who made the call, what options were considered, who argued for each, what the stated reasoning was, and what remains unresolved. The distinction matters six months later when the decision gets questioned and the original thread is three hundred messages deep.

This guide gives you the full prompt, the preparation steps that make attribution reliable, how different platforms handle the task, what to do when the output is vague about who decided what, and a note on doing this at inbox scale rather than one thread at a time.

What a decision memo needs to capture#

Before running the prompt, it helps to know what the output should contain, because the prompt is a set of instructions and the model needs to know what success looks like. A decision memo is attribution-heavy by design. Unlike a summary, which is written for the people who were on the thread, a memo is written for the person who was not: the new hire, the executive who needs to approve the next step, the teammate who was on leave. That reader needs to understand the decision without reading the original chain.

Six fields make a memo usable as a permanent record.

• The question being decided — one sentence, what was actually on the table
• Options considered — each alternative that was seriously raised, not just mentioned in passing
• Who argued for what — each position tied to the named person who held it, with a quoted or closely paraphrased line from the thread
• The decision reached — what was chosen, in plain language
• Who made the call — the person or role with authority to decide, not just whoever spoke last
• Open items — unresolved questions, conditions mentioned but not confirmed, commitments made without an owner or date

Before you run the prompt#

Three things matter before you paste anything into an AI tool.

Get the full thread in chronological order. Expand every collapsed message in your email client, include the quoted history where it carries context, and copy from the opening message down to the last reply. A thread pasted newest-first can still work, but tell the model so in your prompt. Without that note, the model may treat the most recent reply as the opening context, which inverts the logic of who proposed what and when.

Keep sender names and dates intact. A decision memo depends entirely on attribution. Strip names from the paste and the model cannot tell you who held which position. Keep the sender names, timestamps, and any role identifiers that appear in the thread — a header that reads Sarah Chen (VP Engineering) tells the model something a bare first name does not.

How to write a decision memo from an email chain#

Five steps from raw thread to filed memo.

1. 1

   Copy the full thread, oldest message first

   Open the thread in your email client and expand all collapsed history. Select everything from the first message to the last reply and copy as plain text. Delete signatures, unsubscribe footers, and repeated quoted blocks — they add length without adding facts. If your client displays threads newest-first, add a note at the top of your paste identifying the order so the model reads the conversation correctly.

2. 2

   Open a new chat and paste the thread

   Start a fresh conversation in your AI tool of choice. Do not pre-summarize the thread before pasting. The prompt below depends on the model reading the original sender names and message sequence. A pre-summary strips the attribution signals the prompt needs to work reliably — the model cannot name who argued for what if the names have already been removed.

3. 3

   Run the decision memo prompt

   After the pasted thread, append this prompt: Convert this email thread into a decision memo using these sections in order. Section 1, QUESTION: one sentence, what was being decided. Section 2, OPTIONS CONSIDERED: each alternative that was seriously raised; for each, name who argued for it and quote or closely paraphrase the specific line from the thread that shows their position. Section 3, DECISION: what was chosen, in plain language. Section 4, DECIDED BY: who made the call, name and role if stated; if this is unclear, say so explicitly. Section 5, RATIONALE: the reasoning given, using only what was stated in the thread, do not infer. Section 6, REVERSIBILITY: whether the decision was described as easy or hard to reverse; quote any relevant line; write Not stated if none. Section 7, OPEN ITEMS: questions raised but not resolved, conditions mentioned but not confirmed, commitments made without an owner or date. Rules: attribute every position to a named person — if you cannot, write unnamed contributor. Do not invent reasoning the thread does not state — write Not stated. Use only facts in the thread.

4. 4

   Check attribution before circulating the memo

   Read the OPTIONS CONSIDERED section while the original thread is open next to it. Each option should be tied to a named sender with a quoted or closely paraphrased line. If the model attributed a position to the wrong person, or generalized to the team argued, correct it before sharing the memo. An attribution error in a decision record is more harmful than in a summary because the document will be forwarded to people who were not on the thread and will treat it as a reliable source.

5. 5

   File the memo where decisions are stored

   Copy the memo into your shared notes, project tracker, or decision log immediately. A decision record is only useful if it is findable when the question resurfaces. A memo that lives in a chat window is not a record — it is a longer message that disappears the next time someone clears their chat history. The decision and the memo need to be stored in the same place to be worth writing.

Does the prompt work the same on every platform?#

The core prompt works on any capable model. The differences are in how each platform handles long threads, how well it follows the attribution instruction without a follow-up, and what happens to your input after the session.

What to do when the output misses attribution or gets the decision wrong#

Missing attribution in the OPTIONS section is the most common failure. If the model writes several participants instead of naming a person, use a follow-up in the same chat: for each option listed, identify the person who argued for it by name and quote the specific line from the thread that shows their position. This targeted follow-up consistently produces better results than re-running the full prompt, because the model is now focused on one task rather than seven.

A vague DECISION field usually means the thread was genuinely ambiguous. If the output says the team leaned toward option B rather than stating a choice, it is probably correct that no explicit confirmation was recorded. Ask: was there an explicit statement in the thread confirming this as the final choice? Quote the line if yes, and write Not explicitly confirmed if no. A memo that says not formally decided is more useful than one that invents a clean verdict, because it tells the next reader exactly what still needs to happen.

For threads too long to process in one paste, split the conversation into chronological chunks, produce a decision memo from each, then ask the model to reconcile them into a single memo. Tell it: these are decision memos from consecutive chunks of the same thread — reconcile them into one, and treat an option raised in chunk one and rejected in chunk three as a considered and rejected option, not as an open item.

A faster way to document every decision#

The steps above work in any chatbot. What wears thin is the loop: open thread, expand history, copy, switch tabs, paste, run the prompt, check attribution, copy the memo out, find the right place to file it. For a single critical decision that is a reasonable investment. For the twelve decisions scattered across this week's inbox, it is a discipline most teams sustain for a few weeks and then quietly abandon.

AI Emaily is the AI-native email client we build. It reads your threads inside your own mailbox — no paste required — and produces a structured decision memo with attribution pulled from the real sender names and timestamps, attached to the original thread. Your email is processed within your account and never copied to a third-party server. The memo is linked to the source thread and findable by search when the decision gets questioned later. Start a 7-day free trial at aiemaily.com — no cost if you cancel before day 7. See what is included at aiemaily.com/pricing.

We build AI Emaily.