I came out of a day of back-to-back meetings to a 43-message email thread I’d been copied into.
Somewhere in there was a decision. Probably a question for me too. I had no idea where. My old approach was to read the whole thing from the top, lose the plot around message 20, and still miss the one line that actually needed me.
Now I give it ten minutes. AI reads the thread first and tells me what was decided, what’s still open and what’s waiting on me. Then I check it against the real emails and reply myself. Here’s how.
Step 1: Get the thread in front of AI, read only
The simplest way is to copy the whole thread into your AI tool. If your tool can connect to your email, give it read-only access to that one thread. It should never be able to send, delete or reply.
For work email, only use the AI tool your company has approved, and leave out anything confidential you wouldn’t want stored outside your company’s systems.
Step 2: Ask three questions, not for a summary
A generic summary tends to give you a nice paragraph and bury the bit you need. You want answers you can act on.
Here’s the prompt I use:
“Read this email thread. Tell me three things. 1) What has been decided, and who agreed to it. 2) What is still open or being debated, with the latest position on each. 3) Anything someone has asked me for or is waiting on me for, quoting the line and saying who sent it and when. Keep it short. Don’t guess. If something is unclear or contradicted later in the thread, say so.”
Asking it to quote the line matters. It makes the answer easy to check, and it makes it harder for the tool to invent a request that was never made.
Step 3: Check it against the actual emails
This is the bit you don’t skip. For anything in the “waiting on you” list, find the original message and read it properly. For decisions, check the latest email on that point, because threads often change direction near the end.
A quick rule I use:
- Trust, then verify: decisions and open points get a skim of the last few emails.
- Always verify: anything that says you were asked to do something.
- Ignore: anything the AI marked as unclear, until you’ve read it yourself.
Step 4: Reply yourself
Now you know which two or three emails actually matter. Read those properly and reply in your own words. If you want a first draft, ask for one, then rewrite it so it sounds like you. You hit send, never the tool.
If the thread ends with a request you can’t do yet, a short “Seen this, I’ll come back by Thursday” is often the most useful reply you can send.
Step 5: Use it for chat channels too
The same three questions work on a busy work chat channel you’ve been away from. Copy in the messages since you were last there, ask the same questions, check anything aimed at you.
Where AI stops
AI is quick at finding the shape of a long conversation. But it doesn’t know the history between the people, the tone that’s been building, or what was agreed in a hallway and never written down. It can also miss a request phrased politely. Use it to find which emails to read, keep it read only, and do the reading and replying yourself.
Quick questions
Should I let AI reply to the thread for me?
No. Keep it read only. Use it to find what matters, then read those emails yourself and reply in your own words.
What if the AI says I was asked for something I wasn't?
That is why you ask it to quote the line. Check every request against the original email before acting on it.
Is this okay on a work account?
Only use the AI tool your company has approved, and leave out anything confidential you would not store outside company systems.