Blog
What we learned building a meeting room, and what we found when we went and read what people actually say about the ones they already use. Every claim here carries its source, which is the same rule the product applies to a transcript.
You want the AI notes but not the "Notetaker" sitting in the participant list. The three ways to get meeting notes with no bot in the call, what each one costs and misses, and how to pick.
A one-minute microphone check before a video call, and the fixes for the four reasons a browser can't hear you — the site permission, the system permission, the wrong input, and the Bluetooth trap.
Zoom's free plan stops at 40 minutes, Google Meet and Teams at 60 for a group. The free limits side by side, checked against each vendor's own pages, and a free option with no time limit and AI notes built in.
An action item without an owner is a wish. A copy-paste template for meeting action items, the four things every item needs, examples of good and bad ones, and how to get them out of a call without anyone typing.
A decision log is the meeting document people actually go back to. What each entry needs, a template you can paste into Docs, Notion or Confluence, worked examples, and how to tie every decision to the words that made it.
A parking lot keeps a meeting on track without killing the idea that knocked it off course. A copy-paste parking lot template, when to park something, and the one rule that stops it becoming a graveyard.
Five people around one conference-room microphone show up in the transcript as one speaker, or as guesses. Why that happens, what voice enrollment does and doesn't solve, and the setups that give every line the right name.
AI meeting summaries sound certain even when they're wrong. A five-minute check before you forward one: find the words behind every decision, check who said them, and catch the classic mistakes before your stakeholders do.
Drive a browser yourself, rent one from Recall.ai or MeetingBaaS, or join a room that has an API. Real code for each, published prices, and the one fact that decides which to pick.
A Puppeteer script that joins a Google Meet as a guest, waits for admission and records the call with Xvfb, PulseAudio and FFmpeg — with the selectors production bots use today and the honest list of what stops working.
The two meeting-bot APIs side by side — per-hour prices, free tiers, platforms, transcription, what they have in common underneath — and when neither is the right answer. We run one of them, and say which.
An AI notetaker is in your Google Meet call. The host's three clicks that remove it, how to refuse it at the door, the chat commands some bots obey, the calendar step everyone skips, and what it already recorded.
Fireflies' notetaker Fred keeps joining Zoom, Google Meet and Teams calls. The four auto-join options and where they live, why fred@fireflies.ai on a calendar invite is the usual cause, how to remove it from a live call, and the 3-minute rule.
Otter Notetaker keeps showing up in Zoom, Google Meet and Teams calls. The exact Otter setting that stops auto-join, the 30-minute rule, how a host removes it mid-call, and what it keeps.
Read AI is in your Zoom, Google Meet or Teams call and nobody invited it. The two chat commands that make it leave, the exact settings that stop it coming back, and what happens to the recording either way.
A meeting bot is a program that joins a Zoom, Google Meet or Teams call as a participant to record and transcribe it. How it gets in, why it hears one mixed stream, what it costs to run, how to block one, and how to build one.
Per-platform steps to keep Otter, Fireflies and Read.ai out of your meetings, plus an honest account of where each vendor's controls still fall short.
A teardown of how meeting bots actually work — containers, headless Chrome, mixed audio — and why speaker attribution is structural, not a model problem.
Meet or Zoom transcript missing, empty, or never generated? Recovery paths in order, plus how to transcribe a recording you already have.
Speaker attribution errors are not a bug waiting for a patch. They are what happens when a tool hears one mixed stream and has to guess who spoke.