Talk Type · Episode 18 · 3 min ·
Notes Without a Bot in the Call
You don't need a notetaker bot silently joining your meeting. Record it yourself, upload the file after, and turn it into clean minutes, with consent you can stand behind and audio that stays yours.
Transcript
This is Talk Type, from the team at Pepys, where we turn talk into text.
Here's a small thing that quietly changes a meeting. A bot on the participant list. Once people see a named notetaker sitting in the room, they talk differently. More guarded. More careful. And the strange part is, you never needed it there in the first place.
Think about what an auto-joining bot actually does. It dials into your call as a participant. It streams the audio off to a vendor in real time. Then it posts a summary the moment you hang up. Convenient, sure. But it seats a third party inside every conversation it joins, and it quietly makes a decision for you. Everything gets captured, live, whether you meant it to or not.
There's another way, and it splits from the first at one moment. When the recording happens. Your meeting platform already records. Zoom, Teams, Google Meet, each one saves a file you control, with nobody extra in the list. Turn it on. Say it out loud. Then, after the call, upload that file and let it come back as a clean, speaker-labeled transcript. Nothing joined the meeting. You just recorded your own.
That gap matters most for consent. Under Europe's privacy rules, a bot that announces itself, or a "by joining, you agree" banner, isn't actually consent. Consent needs a clear yes, and silence doesn't count. In the US, federal law leans one-party, but about eleven states want everyone on the call to agree. A bot joining a mixed-state call inherits that whole tangle. Recording yourself lets you just ask. A clear yes, captured in the audio, beats a banner nobody reads.
Now the worry. Is a file you upload afterward as good as the bot's live notes? On accuracy, basically yes. Researchers testing conversational speech found machine transcription and professional human transcribers both landing around five or six percent word error. The recording model isn't where quality lives. What you give up is the instant summary. And you can get that back too, minutes later, by running the transcript through a summary afterward. A little slower. Built from text you've actually seen.
Which raises the part nobody likes to say out loud. These systems can invent words. One peer-reviewed study found that roughly one percent of transcriptions contained entire fabricated phrases, and more than a third of those carried real harm. A bot's summary you never open is riskier than a file you can read against your own recording. So don't ship the auto-notes unread. Check the names, the numbers, the decisions.
And if you're writing formal minutes, this fits perfectly. Robert's Rules says minutes record what was done, not what was said. You're logging decisions, not debate. A transcript you own, with timestamps, lets every decision trace back to a line you can replay.
The bot was never the point. The notes were. You can keep the file, keep the control, and keep the room honest, all without anyone silent sitting in the call.
That's this episode of Talk Type. The full write up, with the links and sources, is in the show notes. Pepys transcribes any file or link, any length, pay once, and we never train on your audio. Your first sixty minutes are free at pepys dot co. Thanks for listening, and we'll see you next time.