Three people sit around a conference table. Two more join on Zoom. Everyone attends the same meeting, but they do not get the same meeting record.
The remote attendees show up clearly in the transcript, each tied to their own microphone and account. The people in the room are captured through one shared device, flattened into a single speaker label, or misattributed entirely.
That is the problem with most AI notetakers for in-person meetings. They were built for a world where every participant is individually logged into a video call. But as teams return to offices, the most important conversations are happening around tables again, and the meeting stack has not followed them back.
For the past few years, the meeting tools market has moved quickly. Transcription has improved. Summaries are faster. Integrations with Slack, Notion, CRMs, and project management tools are easier to set up. AI notetakers have become genuinely useful for remote work.
The gap appears when the meeting leaves the screen.
The tools that work beautifully when everyone is individually logged into Zoom start to break down when three people sit around a conference table. The breakdowns go right to the core of what makes a meeting record useful: accurate transcription, reliable speaker attribution, and a clear record of who said what.
Most AI notetakers were built for video calls
The best AI notetakers work well on video calls for a simple reason: the audio infrastructure is already solved before the AI gets involved.
On a Zoom or Teams call, every participant usually joins from their own device with their own microphone. Each person's voice is captured separately, associated with their account, and passed into the meeting platform as an individual audio source.
That makes the job much easier for transcription and note-taking tools. The platform already knows who is speaking because each participant is connected separately. Speaker labels can be tied to accounts, microphones, and meeting participants.
When that same meeting moves into a conference room, the structure changes completely.
Now there are multiple people sitting around a table, one device in the middle, and a single mixed audio stream. The AI has to work out both what was said and who said it from a recording that was never designed to make attribution easy.
That is a fundamentally different capture problem. Most meeting tools were not designed around it.
Why one microphone fails in a real room
The most common workaround is a single high-quality microphone in the center of the table. That might be a Jabra speakerphone, a Rode mic, a conference room system, or a phone running a voice memo app.
This can improve audio quality. It does not solve speaker attribution.
Even with strong audio, a single device capturing everyone in the room produces one mixed stream. Speaker diarization, the process of automatically separating and labeling different voices, can work reasonably well in controlled conditions. It becomes less reliable with background noise, crosstalk, similar voice profiles, side conversations, or people sitting at different distances from the microphone.
That describes most real conference rooms.
The result is often a transcript where the words are mostly right, but the speaker labels are wrong often enough that someone has to go through and fix them manually. For many teams, that manual cleanup is the reason AI meeting notes stop being useful.
A meeting record that needs to be repaired after every meeting quickly becomes another administrative burden.
The speaker attribution problem is the hard part
For many business meetings, transcription accuracy has improved enough that the words themselves are no longer the main failure point. The harder problem is attribution.
A note that says "the team agreed to push the launch date" gives you an outcome.
A note that says "Sarah proposed pushing the launch date, Marcus pushed back because of the sales pipeline, and the team agreed to revisit the timing on Friday" gives you usable business context.
The useful knowledge sits in the attribution.
It tells you who raised the issue, who objected, who committed to the follow-up, and how the decision evolved. Without it, the transcript captures words but loses the structure of the decision.
When speaker labels are unreliable or missing, meeting notes collapse into a flat record. They may capture the words, but they are much less useful for accountability, search, follow-up, or understanding the reasoning behind a decision months later.
In real businesses, the person behind the comment often matters as much as the comment itself.
Why hybrid meetings are even harder
Hybrid meetings make the problem worse.
The standard setup is familiar: some people sit together in a conference room, while others join remotely through Zoom or Teams. The remote attendees each have their own microphone and usually show up as individual speakers in the transcript. The people in the room are captured through one shared device.
That often means one speaker tag for multiple people.
The transcript may clearly distinguish between the remote attendees, while treating everyone in the room as a single participant. There is no reliable way to know who said what among the people who physically showed up.
This creates a documentation inequality that cuts against the purpose of returning to the office.
Remote attendees get clean, attributed transcripts. In-room attendees get lumped together. The people who came into the office often end up with worse meeting records than the people who stayed home.
Better microphones can improve the sound. They cannot fully solve the structural problem of multiple people sharing one audio stream.
Single room mic vs meeting bot vs distributed capture
Different meeting setups solve different parts of the problem.
| Setup | Works well for | Breaks down when |
|---|---|---|
| Video meeting bot | Fully remote Zoom or Teams meetings where each person joins separately | Multiple people are together in one room and appear as one shared audio source |
| Single room microphone | Capturing general room audio for playback or basic transcription | You need reliable speaker attribution across several in-room participants |
| Dedicated conference hardware | Improving room audio quality for remote attendees | The transcript still needs to know which in-room person said each line |
| Distributed audio capture | In-person and hybrid meetings where speaker attribution matters | Participants need to join from their own devices, which adds a small setup step |
The key question is whether you only need a rough record of what happened, or whether you need an accurate record of who said what.
For many teams, that difference determines whether meeting notes become searchable company knowledge or just another transcript nobody trusts.
What actually works for in-person meeting notes
The approach that addresses speaker attribution most directly is distributed audio capture.
Instead of relying on one shared microphone in the middle of the room, each participant contributes their own audio stream from their own device. That changes the problem. The system no longer has to guess who is speaking from a single mixed recording. Each stream is associated with a known participant from the start.
The transcript can then be created per stream and merged by timestamp into a single meeting record.
This is the approach Roundtable uses. Each person in the room opens the app on their own device. Their microphone captures their voice as a separate source. The final transcript can show who said what with far more reliable attribution than a single shared room recording.
The practical difference is less manual cleanup, clearer speaker labels, and a meeting record that is much more useful for search, follow-up, and decision tracking.
The setup that works
If you run in-person or hybrid meetings and want accurate notes, there are a few things to consider.
For fully in-person meetings, distributed capture gives you much stronger attribution than a single microphone setup. A central room mic can capture the room, but it still has to infer speakers from a mixed stream. Per-person capture starts with each speaker already separated.
For hybrid meetings, in-room participants need the same documentation fidelity as remote participants. That means separate audio streams for people in the room, rather than one shared conference room feed. The dominant source for each voice should win, whether that person is sitting at the table or joining remotely.
For any meeting where accountability matters, speaker attribution should be treated as core infrastructure. If you want to go back later and understand who raised a concern, who made a commitment, or who shaped the decision, a single shared audio stream will always be a bottleneck.
The issue is architectural. Better summaries cannot fix unreliable capture.
In-person meetings need in-person meeting tools
In-person meetings are back. The decisions that matter to your business are happening in conference rooms again, in leadership offsites, in workshops, and in conversations that do not always fit neatly into a video-call workflow.
The tools most teams use today were designed for a world where everyone was on a screen, logged in separately, and speaking through their own microphone.
That world still exists. It just no longer describes every important meeting.
As teams return to the office, the meeting stack needs to return with them. In-person and hybrid meetings need tools built around how office conversations actually happen: multiple people, shared rooms, overlapping discussion, fast decisions, and a real need to know who said what afterwards.
Roundtable is an AI meeting notetaker built for in-person and hybrid meetings. Instead of relying on one room microphone or a bot built for video calls, Roundtable uses distributed audio capture. Each participant contributes their own audio stream, so the final transcript can show who said what with far more accuracy.
No meeting bot. No single shared microphone. Far less manual cleanup on speaker labels.