How do I get meeting notes without a note taker joining?

This page may contain affiliate links from which Blur Studio earns commissions.

Meetings end, decisions get made, and three days later nobody agrees on what was actually decided. A written record with names attached settles that fast. The friction is usually social, not technical: a visible note-taking bot joining the call can feel intrusive, especially with clients or on sensitive calls, which is why a lot of teams end up wanting the text without the obvious bot.

The options, from least to most effort

• Your platform's own transcript: Zoom and Google Meet both offer built-in live captions and, on paid plans, a saved transcript, with no separate app joining. Quality is decent for straightforward speech and gets worse with accents, overlapping speakers or background noise. Check what's included on your specific plan, since this varies by tier.
• Otter: free for up to 300 minutes a month (30-minute cap per session), $8.33 a month on annual billing for 1,200 minutes on Pro. Its bot can join visibly or record from your own device audio, and it labels speakers automatically in most cases.
• ElevenLabs Speech to Text (Scribe v2): record the meeting with your platform's own recording feature (no separate bot at all), then run the saved audio file through Scribe v2 afterward. It costs $0.22 an hour through the API, labels up to 32 separate speakers, and handles recordings up to 10 hours long. This is the cleanest way to get a transcript with zero visible third party on the actual call, at the cost of a manual step after the meeting ends.
• Fathom: worth naming because its free plan includes unlimited recordings, transcriptions and basic AI summaries at no cost, which is the one option here that solves the "transcript isn't a summary" gap without paying anything. It currently offers a bot-free joining mode in beta for Mac users; on other setups it still joins as a visible participant, so check that specifically if avoiding a visible bot is your main reason for reading this.

If avoiding a visible bot is the actual point, Scribe v2 run on your own recording afterward, through ElevenLabs' plans, is the most direct way to get there, since nothing joins the call that didn't already belong there.

A worked example: a weekly leadership meeting

Say a five-person leadership team meets for an hour every week, 52 meetings a year, about 52 hours of audio total. On Scribe v2's API rate that's roughly $11.44 for the whole year, genuinely negligible. On Otter's free tier, 300 minutes a month (5 hours) comfortably covers one weekly hour-long meeting with room to spare, so this specific cadence actually fits inside Otter's free plan without paying anything. The math only changes once you're recording more than about 5 hours of meetings a month in total, at which point Otter's Pro tier ($8.33 to $16.99 a month) and the Scribe v2 API both become worth comparing, and Scribe v2 wins on pure cost past a few hours a week.

What actually makes a meeting transcript accurate, or not

Speaker labeling on a one-on-one call is easy; a meeting with five people talking over each other is where accuracy drops. The single biggest factor isn't the tool, it's how people talk: overlapping speech, where two people start a sentence at the same time, is where every transcription tool, Scribe v2 included, is most likely to misattribute a line to the wrong speaker or drop a few words entirely. If accurate attribution matters, who actually said "I'll own this", a habit as simple as not talking over each other during the meeting does more for transcript quality than any tool choice.

On the output itself: Scribe v2 returns speaker-labeled, timestamped segments rather than a flat block of text, which is what lets you or a summarizing tool later pull "what did Sarah commit to" out of an hour of discussion. And as of a change ElevenLabs made on 28 September 2026, you can now edit a transcript after it's generated, for both batch and real-time transcription, so a misattributed line or a misheard decision doesn't have to sit wrong in the record permanently.

If your team meets in more than one language, or someone regularly switches languages mid-sentence, that's worth planning around too. Scribe v2 supports more than 90 languages, but accuracy on a meeting that mixes languages within the same conversation tends to run lower than a meeting conducted entirely in one language. If that's a regular pattern for your team, test it with a real recording of how your meetings actually sound before assuming the tool will handle the switching as cleanly as a single-language call.

What I'd tell someone who just wants this to work

Don't build a custom pipeline for this unless you're doing it dozens of times a week. For most people, the honest answer is: turn on your platform's own recording, and if the built-in transcript isn't good enough, drop the recording into Scribe v2 afterward and read the output. That's maybe five minutes of manual work per meeting, and five minutes is nothing compared to the time lost to disagreeing about what was decided. The API-based automation (auto-transcribe every recording, auto-file it somewhere) only starts paying for itself once you're doing this often enough that the manual step becomes the bottleneck, which for most small teams is further out than it feels.

Step by step

1. Turn on your meeting platform's own recording (audio or video, whichever your platform supports) for calls you want notes from.
2. If more than one person is sharing a single device in the same room, use a dedicated conference microphone or have each person join on their own device audio rather than relying on one laptop mic to pick up the whole room. Distance from the microphone is one of the most common, and most avoidable, causes of a garbled transcript.
3. After the meeting, export or download the recording as an audio file. Noisy recordings can be fixed first: see how toclean up audio before transcribing it.
4. Run it through Scribe v2 (directly, or through a tool/automation that calls the API for you if you don't want to touch code yourself) or upload it to Otter if you'd rather have a ready-made app handle it.
5. Check the transcript against what you remember being said, especially names and any numbers (dates, prices, quantities) that got discussed. Recurring errors can be reduced, as covered instopping transcripts getting technical terms wrong.
6. Have someone write a two- or three-line summary from the transcript, or run it through a summarizing tool. The transcript alone is not meeting notes; it's the raw material for meeting notes. For phone calls, see how toget a call summary without relistening.

The detail people miss: a transcript with 20 minutes of back-and-forth is not the same as "what we decided." Budget for that extra condensing step every time, whether it's a person or a tool, or the transcript just becomes another unread document.

One more thing worth deciding up front, not after the fact: who can actually see a meeting transcript once it exists. A written record of a sensitive discussion, a performance conversation, a client negotiation, a legal matter, is more exposed than the conversation itself ever was, since it's searchable, forwardable and easy to screenshot out of context. Decide where transcripts are stored, who has access, and how long you keep them before you start generating them routinely, rather than discovering after the fact that a sensitive meeting got transcribed into a folder half the company can open.

When to skip this entirely

If your meetings are short, low-stakes, and everyone already sends a one-line follow-up email with action items, you likely don't need a transcription step at all. Add the transcript step specifically for meetings where disagreement about what was said is common or costly, not as a blanket policy for every call on the calendar. It's also worth skipping a dedicated transcription tool if your meeting platform's own built-in transcript, even an imperfect one, is already good enough to jog people's memory. The bar for "good enough" here is lower than for a customer-facing transcript you might need to stand behind in a dispute; for an internal team meeting, a transcript that's 90% accurate usually settles a "what did we agree" disagreement just fine, so don't assume you need the most accurate, and most effortful, option by default. The same method works forgetting written records of customer calls.

How I'd approach this

The point where this becomes worth building something custom is when "record, then transcribe, then summarize" turns into a real weekly chore across many meetings, and someone's manually doing all three steps every time. I've built this as a small automated pipeline before: recording lands in cloud storage, gets transcribed automatically, and a short summary gets sent to whoever needs it, with no bot ever visible on the call. For one client, that meant every client call auto-transcribing to a shared drive with a same-day summary flagging any price or discount mentioned, which let the owner catch an under-quoted deal without listening to a single call personally. If you're at the point where the manual version is eating real hours each week, that's worth a conversation rather than more manual workarounds.