Recording a game you run online is easy to start and easy to get wrong in ways you only notice afterward — a single muddy track, no one's sure who agreed to it, and three hours of audio you never turn into anything. Here's how to do it so the recording is actually useful.
One track per player beats one track for the room
The single most important choice is whether you capture one mixed track for the whole call or a separate track per speaker. Per-speaker is far better, and not just for audio quality. When each person has their own track, you know who said what without guessing — which means a transcript can attribute lines correctly, and you can drop a single player's audio if they'd rather not be recorded, without losing everyone else's.
A mixed room track throws that away. Everyone is on top of each other, cross-talk is unrecoverable, and any transcription has to guess at speakers and usually guesses wrong. Discord's architecture gives every participant their own audio stream, so per-speaker capture is available if the tool you use takes advantage of it.
Get consent once, out loud, and make it easy to withdraw
Recording people's voices is not a thing to do quietly. Before the first recorded session, say it plainly to the whole table: this is being recorded, here's what it's used for, here's how long the audio is kept, and anyone can opt out. Then make opting out real — a player who changes their mind should be able to have their track excluded without a negotiation.
Treat it as a room-level agreement that any individual can step out of, not a box buried in a settings page. It's the difference between "we record the table" and "we record you," and only the first one is something a group actually consents to.
Turn the audio into something you'll use
Raw audio is a liability, not an asset. Nobody re-listens to three hours of a session, so the recording only earns its keep once it becomes something you'll actually open: a transcript you can search, a recap you can send, notes that feed your campaign wiki. Plan for that step before you hit record, or you'll accumulate a folder of files you never touch.
The usual pipeline is: capture per-speaker audio, transcribe it (speaker-labelled, because you kept the tracks separate), then summarize the transcript into a recap. Doing each step by hand works but rarely survives contact with a busy week.
The in-person case
If your table is around a physical table, the same principles hold but the mechanics differ: you're usually working with one microphone in the room, so per-speaker separation isn't available and consent is genuinely room-level. The trade-off is that an in-person recording can hear the thing an online one often can't — the numbers people say out loud as they roll.
Where Six Axes fits
Six Axes is built around exactly this: it records per-speaker over Discord, enforces consent at the track level (an un-consented player is never even uploaded), transcribes with speakers attributed, and writes the recap and campaign wiki from the result. On a supported virtual tabletop it also captures what was rolled, not just what was said. See what it does, read how we handle consent and your data, or join the pilot.