Live notes from a meeting you are already in

Almost every first attempt fails the same way, and it is always the sharing dialog. The timer runs, the level meter sits flat, and no text appears. Work down this list.

Nothing you say is in the transcript, but everyone else is

This is the most common one, and it is not a fault — it is how shared audio works. Sharing a tab or a screen gives you what your computer is playing, which is the other participants. It never contains your own voice, because a meeting does not play your voice back to you. Your side of the conversation reaches the transcript only through your microphone.

So if your own words are missing, the microphone was refused, muted or set to the wrong input. Watch the second meter while you talk: it turns green when your microphone is genuinely being heard. The page now asks for the microphone before the sharing dialog so the prompt is not buried behind it.

You shared a screen or a window instead of a tab

Chrome only offers Also share tab audio when you pick Chrome Tab. A whole-screen share can still carry your system audio on some machines, so it half-works — the other participants come through and everything looks fine. Sharing the tab is cleaner: it captures the call and nothing else your computer happens to be playing.

The tab audio switch was left off

It sits in the bottom-left corner of the picker and is off by default. It is easy to click the meeting thumbnail and hit Share without ever seeing it.

You shared the wrong tab

The thumbnails in the picker are small and a Meet tab looks much like any other. Share the tab the meeting is actually running in, not this one.

Nobody has spoken for the first 30 seconds

Text arrives in 30-second segments, so the first one takes at least that long — plus a one-off wait the very first time while the speech model downloads. Silence in a segment produces no words, which is correct behaviour rather than a fault.

You are not in Chrome or Edge on a desktop

Firefox and Safari do not implement tab audio capture at all, and no mobile browser does. The page checks before you start and says so, rather than half-working. There is no workaround; it is a missing browser capability, not a setting. If you are on one of those, record the call and use the free transcriber on the file afterwards.

The meeting is muted on your machine

Tab capture takes the audio the tab is playing. If you muted the tab in Chrome, there is nothing to take. Your own microphone is separate and is off unless you switch it on before starting.

Every other way of transcribing a Google Meet call takes one of two routes: send a bot into the meeting, or make you pay for a Workspace tier. This takes a third — your own browser listens to a tab you are already watching.

The last two rows are the honest cost of the approach and we are not going to bury them. Speaker labels need a second model that separates voices, which is not running here, so you get continuous text. And tab audio only exists in Chrome and Edge on desktop. If either of those matters more to you than keeping the audio on your own machine, a bot-based tool is the better choice — our Otter alternatives and Fireflies alternatives guides set out what each free tier actually caps.

For why Google’s built-in transcript is missing from most accounts, and which Workspace tiers include it, see how to transcribe a Google Meet call free.

Press Stop and finish and the full transcript is on screen, ready to copy or download as a text file. It is also already saved on this device, so closing the tab does not lose it.

Enter an email address and you also get a brief: a short summary of what the meeting was for and what came of it, the topics covered, the action items with whoever was named against them, and the questions left open. The full transcript comes attached as a text file, because that is the part you will want to search in three months. A long meeting takes a few minutes to write, and the page tells you when it has actually been sent rather than only that it was queued.

The brief goes to the single address you type in — not to everyone who attended. Shared tab audio carries the sound of the meeting and no list of who is in the room, so there is no attendee list to send to. Forwarding it is a deliberate act on your part, which for a recording of other people is the right default.

The speech model is Whisper, running in your browser through WebGPU where your machine supports it and WebAssembly where it does not. The first meeting downloads about 73 MB of model and about 23 MB of speech engine; both are cached afterwards, so later meetings start immediately. Everything comes from this site rather than a third-party CDN, which is why it still works on networks that block one. On a graphics-capable machine it keeps up with the call comfortably; on WebAssembly it is slower but still finishes each segment well inside the next one.

Expect a good transcript of clear speech and an imperfect one of everything else. Crosstalk, heavy background noise, a bad connection on someone’s end and unusual proper nouns all cost accuracy — names of people and companies are the most common thing it gets wrong. Read the brief before acting on it, particularly the part that says who agreed to what.

It handles 12 languages including English, Hindi, Tamil, Telugu, Marathi, Spanish, French, German, Portuguese, Arabic, Chinese and Japanese. Leave it on automatic detection unless the meeting switches between languages, in which case naming the main one gives a steadier result.

On telling people: announce it at the start. Recording law varies by where each participant is sitting, and some places require every party to agree rather than just one — is it legal to record meetings goes through how the rules differ.