How a Microsoft Teams Recording Bot Works

A recording bot is a participant, not a plugin

When people picture call recording, they often imagine software silently capturing audio in the background. A Microsoft Teams recording bot works differently — and understanding that difference is the key to understanding why it's reliable. The bot is a Microsoft Graph Communications applicationthat actually joins the call as a participant, receives the same real-time media streams every other attendee receives, and records them. Because it participates through Microsoft's own communications platform rather than scraping a screen or tapping a device, it captures clean, native audio for every call it joins.

In our service, that bot runs on a server dedicated to your organization, which we host and manage. You never install anything on user devices. Instead, the bot is authorized in your Microsoft 365 tenant and joins calls through one of four methods described below.

Four ways the bot joins a call

A single trigger rarely covers every scenario, so the bot supports four join methods that together capture scheduled meetings, ad-hoc calls, and direct conversations:

• Compliance policy. The most important method for regulated teams. When an administrator assigns a Teams compliance recording policy to a user, the bot is invited to every call that user makes or receives — automatically, with the standard Teams recording banner shown to participants.
• Manual join. The bot can be added to a specific meeting or call on demand through its API when you want a one-off recorded session.
• Calendar auto-join. The bot monitors Exchange Online calendars for a defined group of users and automatically joins their scheduled meetings — no per-user policy required.
• Direct calls. Direct one-to-one calls to and from in-scope users are captured too, closing the gap between formal meetings.

The recording bot page covers each method in more depth, and the how it works page shows the full flow.

Capturing every speaker separately

Once the bot is in the call, it records two things at once: a combined recording of the whole conversation and per-participant (unmixed) audio, where each speaker is captured on their own track. That second part matters more than it sounds. On a busy multi-party call, a single mixed recording makes it hard to attribute who said what when people talk over each other. Separate per-speaker tracks remove that ambiguity — which is exactly what you need when a recording has to stand up to an audit or a dispute.

From audio to searchable text

Audio alone is slow to review. So each recording is transcribed automatically using Azure AI Speech-to-Text, with speaker diarization that labels the transcript by speaker. Because the bot already captured each participant on a separate track, that speaker attribution is far cleaner than trying to separate voices out of one mixed file after the fact. The result is a searchable, speaker-labelled transcript alongside every recording. See the transcription page for detail.

Storing the record where you control it

Finally, the recording and its transcript upload automatically to a SharePoint document library in your own Microsoft 365 tenant. The evidence lives inside your environment, under your access controls and retention rules — not on a platform you can't govern. For a fuller picture of Teams storage, see where Teams recordings are stored.

Why this beats pressing record

The native Teams Record button depends on a person remembering to start it, produces a single mixed track, and lands in default storage. A recording bot governed by policy removes the human step entirely, captures every speaker cleanly, transcribes automatically, and archives to a place you control. For a regulated organization, that's the difference between hoping you have the call and knowing you do. Read our comparison of native versus compliance recording for the full breakdown.