When I tested AI note-taking tools, I paid particular attention to what happens after the summary: can I tell who made a commitment, find the decision again, and carry that context into the next conversation?
For an in-person meeting, I want an AI note taker that captures the people in the room, lets me review speaker labels, and keeps decisions connected to the conversation. My choice is MeetZen when I also want browser recording without an installation and a memory I can revisit through ChatGPT or Claude.
I’m building MeetZen, so that is the workflow I’m designing it around. Here is how I choose a tool for the room, then turn the recording into notes I can use later.
How I choose an AI note taker for in-person meetings
I start with five questions:
- Can it capture the room? I need microphone recording, not a workflow that depends on a video-call link.
- What needs installing? MeetZen runs in the browser. That matters when I want to record without installing a separate app.
- Can I check who said what? Several people sharing a microphone is different from separate participants on a call.
- Do the notes preserve commitments? I look for the decision, the person responsible, the agreed deadline, and anything still unresolved.
- Can I find the conversation again? A summary becomes more useful when I can connect it to the right project, people, and follow-up questions.
These are the differences I found most relevant in the workflows I tested:
| Tool | Fit for a conversation in the room | What matters to my choice |
|---|---|---|
| In-person microphone recording in the browser, with speaker identification | My choice for connecting the conversation to tasks, folder-specific vocabulary, related sessions, and later AI retrieval | |
| In-person recording with speaker identification | A strong option for summaries and tasks; I prefer MeetZen when I want folder-specific dictionaries and a visual view of related conversations | |
| A useful option for capturing in-person conversations | In my tested workflow, speaker labeling was limited to online meetings; I would check attribution carefully when several people share a microphone | |
| An online-meeting-first workflow in my evaluation | I would not choose it for a room conversation where identifying individual speakers is essential; my tested workflow did not identify speakers in person |
For the broader trade-offs, see my comparison of six AI note-taking apps. Here, I’ll focus on the room-recording workflow and the questions I want the notes to answer.
1. Decide what you want to recover
I start with the question I expect to ask after the meeting. For a project discussion, it might be: “What did we agree to change?” For a lesson: “Which exercise should I practice before next time?”
That question gives me a useful check when I review the result. A readable summary may still miss the particular commitment you care about.
Before recording other people, explain that you want to use AI notes and agree on what will be recorded. A meeting bot does not need to join an in-person conversation, so make the recording visible through the conversation itself.
A simple opening is enough to establish the purpose: “I’d like to record this so I can check the decisions and follow-ups afterward. Are you comfortable with that?”
2. Set up the microphone for the room
For this workflow, I use MeetZen’s in-person recording mode in the browser. Select the microphone you intend to use and allow the required access. You do not need to install a separate MeetZen app.
For an unfamiliar setup, make a short test recording before the discussion. Check that voices from the seats people will use are understandable. Move the device away from obvious noise sources, and avoid placing it so one speaker is much closer than everyone else.
A laptop is a straightforward starting point. If you intend to use a phone browser, test that specific device and browser first. Room recording and capturing audio from a phone call are different situations; this guide covers people speaking in the room.
Keep the recording page available and follow its status. This workflow assumes an internet connection; recording without internet is not a current MeetZen capability promised here.
3. Give the conversation its own context
I put the conversation in its project folder and add useful vocabulary before recording. MeetZen lets each folder have its own dictionary, so terms from a product project can stay separate from terms used in a language lesson.
For the example below, I’ll use a fictional design review for a product called Northstar. Its folder vocabulary might include “Northstar” and the feature names discussed in the review. A tutoring folder would contain a different set of words.
Keep the list purposeful. The value comes from supplying relevant context, rather than collecting every term you might ever say. Review important names in the resulting transcript even when you have added them to the dictionary.
During the meeting, make decisions and ownership explicit when you can. “I’ll handle it” is harder to interpret later than “I’ll send the revised proposal on Thursday.” If no deadline was agreed, leave it unresolved rather than inventing one in your notes.
4. Check who said what after processing
When the session is ready, I review the transcript and speaker labels. MeetZen supports speaker identification for conversations in person as well as online. I can correct the result when a name or attribution needs attention.
I start with the moments that affect action: a promise, a changed decision, a deadline, or a disagreement. Attribution matters because it changes who I should follow up with.
Here is the fictional Northstar exchange I’ll use to illustrate that check:
Alex: “Let’s test the shorter onboarding flow first. I’ll send the revised proposal on Thursday.” Sam: “Agreed. I’ll review your proposal on Friday. Who will recruit the participants?” Mia: “We haven’t assigned that yet. Let’s decide at the next meeting.”
If those lines are assigned to one speaker, the notes lose the distinction between sending the proposal and reviewing it. I want to verify Alex’s and Sam’s labels before treating either commitment as an assigned task.
5. Review decisions and turn commitments into usable tasks
I read the summary alongside the transcript, checking the agreed outcome, any conditions attached to it, and what remains open.
For that exchange, here is what I would check in the notes:
| Item | What the notes should preserve |
|---|---|
| Decision | Test the shorter onboarding flow first |
| Commitment | Alex will send the revised proposal on Thursday |
| Commitment | Sam will review the proposal on Friday |
| Open question | Who will recruit participants? No owner agreed yet |
These are different kinds of information. An open question should not quietly become an assigned task, and a suggestion should not become a decision.
MeetZen extracts tasks with owners and deadlines when they are present in the conversation. Review those details and adjust them as needed. You can also choose one of three task priorities manually.
Before the next meeting, I want the task statuses in MeetZen to reflect what I have actually checked. A saved task marked open tells you the state of that record; it does not establish whether someone completed the work elsewhere.
6. Make the next conversation easier to prepare for
I give the session a useful title and retain its folder, people, and relevant tags. A title such as “Northstar — onboarding review” tells you more than “Thursday meeting.”
Later, I can return through the project folder, a participant, or a topic shared with another session. Graph View offers a visual way to explore related conversations.
If you connect ChatGPT or Claude and authorize access, you can also ask the assistant to search and read the saved material. For example:
Use MeetZen to find my latest Northstar onboarding review. What did we decide to test, who agreed to send and review the proposal, and who is responsible for recruiting participants? Include agreed deadlines and the saved status of any related tasks. Cite the source session. If an owner or deadline was not agreed, say so.
For the fictional exchange above, I would check that the answer keeps Alex’s and Sam’s commitments separate and leaves recruitment unassigned. I open the source session before using an answer to prepare an agenda or follow-up.
What if the notes are incomplete?
I work back from the problem. If a speaker is wrong, correct the attribution. If a specialized term is wrong, correct the transcript and review the folder vocabulary. If the recording is difficult to understand, improve the microphone placement before the next session.
If the original conversation did not settle a question, follow up with the people involved. AI notes are useful partly because they make those gaps easier to see.
Why I recommend MeetZen for this workflow
MeetZen combines browser capture with in-person speaker identification, tasks, folder-specific vocabulary, and ways to revisit related conversations. A discussion can stay connected to the next meeting and to the voice note you record afterward.
That makes it my preferred choice when I want to preserve this meeting and use its context again. I would start with one conversation, review its decisions and speaker labels, then try to retrieve it before the next meeting.
. Access is currently by invitation.
Once you have several recordings, use my guide to building a second brain from conversations to organize them around the questions you want to answer.

