There’s a window after a great discovery call, (maybe 20 minutes, maybe an hour) where everything is still vivid. You know exactly what the other person needs. You remember the hesitation in their voice when they described their current situation. You caught the thing they said almost offhand that was actually the whole problem.
Then the next call starts. Or a Slack message comes in. Or you just need to eat lunch.
And by the time you sit down to write the follow-up email, half of it is gone. You write something professional and accurate, but it’s missing the texture. The thing that would have made them feel actually seen.
The transcript was sitting right there the whole time. You just weren’t treating it like a first draft.
The transcript is already a document. It just needs a translator.
Most people use transcripts as a record. Something to search if you need to confirm what was agreed, or find a specific quote. But a transcript is also a raw first draft of every follow-up asset you need from that meeting: the recap email, the discovery summary, the next steps document, the follow-up guide.
You already did the thinking in the meeting. The transcript caught it. The only question is whether you have a system to turn it into something usable.
Here’s mine.
The workflow: from transcript to done in under 20 minutes
Step 1: Get the transcript.
I use Otter.ai for most calls. It integrates cleanly with Zoom and runs in the background without anyone having to remember to hit record. If you’re on a lot of back-to-back calls and audio quality is a recurring issue, Krisp is worth adding to your setup. It’s not just a transcription tool; it handles noise cancellation in real time, which means cleaner audio going into whatever transcript tool you’re using. For calls where you want AI-generated summaries built right into the transcript, Fireflies.ai does this well and a lot of people I know have made it their default.
Honest caveat: speaker labeling across all of these gets shaky on calls with three or more people. I always do a quick scan before I use the transcript for anything.
Step 2: Feed it to Claude with a specific prompt.
Not a generic “summarize this.” Something like this:
“This is a transcript from a discovery call with a potential client. They’re a [one-line description of who they are and what they do]. Write me: (1) a follow-up email that recaps what they told me, names the three things they need most, and proposes a clear next step: warm, not salesy; (2) a one-page discovery summary I can keep on file, organized by: current state, desired state, blockers, what they’ve already tried.”
The output is almost always 80% of the way there. Usually more.
Step 3: Do the human edit.
This is the 10 minutes that matter most. Read the output as if you’re the client. Does it sound like you actually listened? Are the specifics right? The numbers they mentioned, the exact terminology they used for their role, the thing they said that made you both laugh a little? Add those back in. Delete anything that sounds like a form letter.
This is the step you can’t automate. And it’s also the step that makes the difference between a follow-up they forward to their boss and one they skim and file.
What this looks like for discovery call guides
If you run discovery calls regularly, (as part of a sales process, an onboarding workflow, or a coaching intake) you can take this one step further.
After about five to ten transcripts, you have enough material to build a reusable discovery call follow-up guide: a template that captures the most common questions your clients raise, the objections that come up most often, and the follow-up language that has resonated.
Feed those transcripts to Claude and ask: what patterns show up across these calls? What are the most common concerns? What language did I use that seemed to land?
You’re not replacing your process. You’re building institutional memory from conversations that would otherwise disappear the moment the call ended.
Your turn
Pull up the transcript from your last discovery call or client meeting. If you don’t have one, set up auto-transcription for your next one. Otter.ai has a free tier that’s genuinely good enough to start.
Then try this: feed the transcript to Claude with a specific prompt, not just “summarize.” Notice what it gets right. Notice what it misses. The edit you do after is where your voice comes back in.
Twenty minutes. One email that actually lands.
~ Sheila 💕
Coming up: how to write in your actual voice, even when you’re using AI to do the drafting.