Email replies used to eat a chunk of my morning. Open the message, dig through my memory for what we actually agreed on in that call last week, then type the whole thing out from a blank box. Slow and easy to get wrong.
So when I came across this post from an AI professional on LinkedIn, I stopped scrolling. The author lays out a 5-step workflow that lets Claude write your email replies for you, and the whole loop takes about two minutes. The clever part is that Claude isn’t guessing. It reads your inbox and your actual meeting transcripts at the same time, so the draft reflects what was really said out loud.
I’ve broken down exactly what the creator shared, step by step, with the reasoning behind each move.
The two tools you need
Before the steps, here’s the setup the expert relies on. Two pieces, both doing one job each:
- Granola.ai for capturing what happens in your meetings.
- The Claude desktop app for reading your inbox and those meeting notes, then writing the reply.
Neither one writes emails on its own. It’s the combination that makes this work, and that’s the insight I think most people miss.
The 5-step workflow
- Install Granola.ai. It’s free, and it silently transcribes every Zoom, Teams, or Meet call you join. No bot slides into the meeting, so people talk normally and you capture the real conversation. The author’s habit: always open Granola when you’re on a call, so the context of every meeting gets gathered automatically.
- Open the Claude desktop app and go to Settings, then Connector, then connect both Gmail and Granola. This is the move that matters. Claude can now read your inbox and your meeting transcripts together. Most people connect only Gmail, and the original poster is blunt about why that falls short: without Granola, Claude drafts emails with zero memory of what was actually said in your meetings.
- Open the email you need to reply to. Note the sender and the meeting it ties to. The expert calls this the step that makes the prompt 10x sharper, because you’re telling Claude exactly where to look inside Granola instead of letting it guess.
- Prompt Claude with that context (the exact template is below).
- Ask Claude to “save as draft.” The reply lands straight in your Gmail with zero copy and paste. You just open Gmail, review it, and hit send. The whole thing happens without leaving Claude.
The prompt to copy
Here’s the exact prompt the creator uses in step 4. Swap the bracketed parts for your own details:
Pull [sender]’s email about [topic]. Search Granola for our meetings on [topic] in the last [timeframe]. Draft a reply that [confirms / answers / declines]. Tone: warm + direct. Length: short.
Notice how tight it is. It points Claude at a specific sender, a specific topic, and a specific timeframe, then tells it the job (confirm, answer, or decline), the tone, and the length. That’s why the output comes back usable instead of generic.
The safety tip worth stealing
This is the part I was most glad the author included. Don’t grant write access on day one. Start read-only. Watch how Claude uses your inbox for a week, then upgrade once you trust it.
The original poster’s warning: most accidents happen when you hand Claude full access and then ask it to “clean up my inbox.” Read-only first is a small friction that saves you from a big mess.
Why this clicked for me
The reason this workflow is smarter than the usual “AI writes my emails” pitch comes down to memory. A tool that only sees your inbox is working with half the picture. It knows what was written, but not what was agreed on the call. By wiring in meeting transcripts, this savvy professional gave Claude the missing context, and that’s the difference between a draft you rewrite and a draft you actually send.
A few practical ways to put it to work:
- Follow-ups after sales calls: Claude pulls the discovery conversation and confirms next steps in your voice.
- Declining politely: point it at the thread and the meeting, and it declines while referencing what you discussed.
- Answering detailed questions: it cross-checks the email against what was actually decided, so nothing contradicts your last chat.
It also fits a bigger shift I keep noticing. The value isn’t in the AI writing faster. It’s in connecting the AI to the context that already lives across your tools, so the output is grounded in reality instead of made up.
Give it a try
If you spend more time on email than you’d like, this two-minute loop is worth testing. Start read-only, use the exact prompt above, and see how close the first draft lands. For the full walkthrough and the creator’s own notes, check out the original LinkedIn post.