AI follow-up emails after sales calls: how I draft 50 to 100 a day from meeting notes
Yes, AI can write your follow-up emails after sales calls, and it's the single biggest time saver in my week. My setup: every call is captured in Granola, Claude is connected to my Gmail, and at the end of the day it drafts follow-ups from the notes of each conversation. That's 50 to 100 drafts a day. I review them in batches and send them myself.
The rule that makes it work is the last part. The AI drafts. I send. When you take dozens of calls a day, the follow-up is the job, and handing the writing to an agent is what lets me do that job at all. Handing it the send button is how you end up apologizing to someone important.
Follow-up is the bottleneck
Most founders think their bottleneck is getting meetings. Once a show-led pipeline is running, it flips. You have more conversations than you can follow up on properly. Pre-interviews, recordings, debriefs, investor calls, client check-ins. Each one ends with something you promised: an intro, a deck, a link, a next time.
If the follow-up is late or generic, the conversation was wasted. The other person remembers that you said you'd send something. They notice when it arrives three days later with no reference to what they said.
I used to write these at night. With a few calls a day that's fine. With dozens, it's impossible, and the quality drops exactly when the stakes rise.
Notes in, drafts out
I described the setup to the team on our October 9 call: Claude connected to Gmail, generating 50 to 100 follow-up emails daily from Granola notes. The flow is simple.
- Every conversation gets captured. Granola records and summarizes each call, so I'm not typing notes while someone's talking.
- At the end of the day, the agent reads the notes. For each call it pulls what was discussed, what I committed to, and what the other person needs next.
- It drafts a follow-up in my voice. A short recap of their specific problem, what we'll do, and one clear next step. The draft lands in Gmail as a draft, not in anyone's inbox.
- I review in batches and send.
The draft is grounded in the real conversation. That's the difference between this and a template. A template says "great speaking with you today". A draft from the notes says what they told me about their raise, the investor they're worried about, or the hire they can't make.
Batch review, human send
On October 13, I explained this to a biotech founder who was weighing whether to work with us. He was raising and had a real concern: how do you keep context across dozens of back-to-back investor conversations? By the twentieth call, they all blur.
My answer was this exact system. Granola captures every conversation. At the end of the day, AI drafts the follow-ups. I review and send in a batch, about 25 emails drafted at once. The context lives in the notes, not in my memory, and every draft starts from that context.
Batching matters for a reason people underrate. Reviewing 25 drafts in one sitting is fast, because you're in editing mode. You fix a name, cut a line that sounds off, add a personal touch where you have one, and move on. Writing 25 emails from scratch in between calls would take all afternoon and you'd do it badly.
What we never let it do
The agent doesn't send. Ever. Every follow-up goes out because a person read it and pressed send.
I learned this rule the hard way on the reply side, not the follow-up side. Earlier in October, an AI auto-replier on a client campaign asked a top crypto CEO "which podcast?" on a thread that named the show. The model wasn't stupid. It was missing context, and nobody was checking before it sent. A follow-up after a real conversation is even more personal than a campaign reply. If it's wrong, it's wrong to someone who just spent half an hour with you.
Other things we keep human:
- Anything with a price in it. The draft can mention a next step. The number gets written or checked by me.
- Anything sensitive the other person shared. If someone talked about a co-founder issue or a personal situation, it stays out of the email unless I put it there.
- Who gets copied. The agent doesn't add people to threads.
What a good draft looks like
A follow-up that works has four parts, and the agent is briefed to write all four. First, one line that proves you were listening: their problem in their words. Second, what you're going to do, or what you already did, like an intro you promised. Third, one next step, with a time if it's a meeting. Fourth, nothing else.
The temptation with AI is to let it write more because more is free. Don't. A long follow-up reads like a newsletter. A short one reads like a person who remembered. When I edit drafts, most of my edits are cuts.
The other thing I check for is tone drift. Agents slide toward polite corporate phrasing over time. "I hope this finds you well" creeps back in. When I see it, I fix the brief, not just the email, so it doesn't come back tomorrow.
Where it goes wrong
Notes are summaries, and summaries can be wrong. Sometimes a note assigns a next step to the wrong person, or flattens "maybe next quarter" into "interested". That's another reason the send stays human. When a draft says something I don't remember agreeing to, I check the call before it goes out.
Context across dozens of conversations
The less obvious benefit is memory. Because every call is captured, the agent can look back across past conversations with the same person when it drafts. If an investor said in September that they'd look again after our next milestone, the October follow-up can mention it.
I've taken this further on the show side. I built a searchable database of every word spoken across my interviews. When a guest's name comes up again, I can find what they said, what they cared about, and who they mentioned. That's the kind of continuity that used to belong only to people with great memories and assistants.
For a founder doing 30 investor calls in a raise, this changes the experience on the other side of the table. Every investor gets a follow-up that sounds like you were listening, because the system was.
How to start
You don't need my full setup to get most of the value. Start with three pieces: a note-taker that captures your calls, an AI assistant that can read those notes and write Gmail drafts, and 20 minutes at the end of each day to review and send.
Give the agent a short brief on how you write. Short sentences, one next step per email, no "just following up". Then read its first week of drafts carefully and correct it. By week two you'll be fixing a word here and there.
The free start on gtm showrunner works the same way on the outbound side. Connect Gmail and you get about 10 invites a day, written for you and saved as Gmail drafts you send yourself. Nothing goes out without you. If you want to try a drafts-you-approve workflow without committing to anything, start free.
Frequently asked questions
Can AI write my follow-up emails?
Yes. I have Claude connected to Gmail drafting 50 to 100 follow-ups a day from my Granola meeting notes. Each draft starts from what was said on that call, then I review and send.
Should AI send emails automatically?
No. Let it draft, and have a person send. Follow-ups after real conversations are personal, and a wrong message reaches someone who just spent time with you. We keep the send button human.
How do I follow up after 30 calls a day?
Capture every call with a note-taker, have an AI draft follow-ups from the notes at the end of the day, and review them in one batch. Reviewing about 25 drafts at once is far faster than writing them one by one.
How do I keep context across many investor conversations?
Keep the context in captured notes rather than memory. The drafting agent reads each call's notes, and can look back at earlier conversations with the same person.