By Thursday the work chat looks like a junk drawer.
Slack. Teams. WhatsApp. Somewhere in there, a decision got made. Someone asked "are we going with option B?" Six people reacted. Two more replied "sounds good." Then the conversation moved on. A week later nobody can find the bit where you actually agreed.
I used to scroll back and reconstruct the whole thing from memory. Half the time I missed a caveat someone added in a side reply. The other half I made something sound more decided than it was.
These days I copy the thread into AI and ask for a short decision note. What we agreed. Who owns what. What’s still open. Then I edit it and send it myself.
Same rule as the rest of this series. AI drafts. You decide. It never posts into the chat for you.
Here’s how I do it.
Step 1: Grab the bit of the thread that matters
You don’t need the whole channel history. Copy from the message that started the topic through to the latest reply. If people jumped threads, paste the related bits in order and note where they came from.
Keep names and timestamps if your chat app includes them. That helps the AI match owners to actions. Strip anything that shouldn’t leave the channel, like personal details or stuff outside the decision.
Step 2: Ask for a decision note, not a summary
A summary retells the chat. You want something people can act on. Paste the thread and use a prompt like this:
Turn this chat thread into a short decision note I can send back to the group.
Include only:
1. Decisions (what we agreed, in plain language)
2. Owners and next steps (who does what, and by when if it was said)
3. Open questions (things raised but not resolved)
4. Assumptions I'm making that the thread didn't state clearly
Keep it under 180 words. Use the people's names from the thread. Don't invent agreement. If something only got a thumbs-up emoji and no clear yes, put it under open questions or assumptions, not decisions.
Flag anything that needs my judgement before I send it.No long persona. No project-management jargon. Just a draft you can tighten in two minutes.
Step 3: Check it like you’re about to put your name on it
Read the draft against the thread. Watch for three traps.
First, false certainty. The AI will sometimes turn "maybe we should" into "we decided." Soften or move those lines.
Second, missing owners. If the thread never named who would do something, don’t let the draft invent a name. Put it under open questions.
Third, side comments that weren’t decisions. A joke reply or a "we could also…" should not become a commitment.
Edit until you’d be happy to send it under your own name. Then send it yourself, in the same chat or as a short email, whatever your team actually reads.
What you get back
Usually four useful things.
A short list of decisions pulled out of the noise. The "we’re going with option B" that lived in a reaction and never got written down.
Names next to next steps, with dates where someone gave one. Where nobody did, the draft should say so instead of inventing Friday.
Open questions. These are the bits that would otherwise turn into quiet assumptions. Check them before you hit send.
And a small assumptions section. That’s your warning light. If the list is long, the thread wasn’t as clear as it felt. Ask the group before you treat those as fact.
The five-minute Thursday version
If you’ve only got five minutes between meetings, skip the polish. Paste the thread, run the prompt, fix anything that wasn’t actually agreed, and send. A plain decision note today beats a perfect one that sits in your drafts until Monday.
When to skip it
Skip it when the thread was a quick yes on something tiny and everyone already knows. Forcing a decision note onto "sure, book the room" just creates noise.
Skip it when the conversation is still live and people are mid-debate. Wait until the back-and-forth has settled, or you’ll freeze a draft that changes five minutes later.
And think twice when the chat is sensitive, private, or outside tools your company allows. Don’t paste confidential threads into a tool that isn’t approved. If in doubt, write the note yourself from memory and keep the source in the original chat.
Try it once
Pick one messy thread from this week. Paste it in, run the prompt, edit for two minutes, and send the note yourself.
If the group stops asking "wait, what did we decide?", keep the prompt. If the draft keeps inventing certainty, tighten the instructions. The goal is fewer decisions lost in replies, not another process to look after.