How to write meeting minutes from a recording (2026 guide)

To write meeting minutes from a recording, first transcribe the recording into text. Then summarize the transcript into a structured set of notes: attendees, decisions made, and action items with owners. You can do this by hand, or give the transcript and the attendee list to an AI assistant to draft.

That two-step split, transcribe then summarize, is the whole trick. Most people try to write minutes while re-watching a recording, scrubbing back and forth. It's slow and error-prone. Working from a clean transcript is faster and far more accurate. Here's the full workflow.

1. Transcribe the recording

You can't summarize what you can't read, so step one is converting the recording to text. Upload your file to an AI transcription tool and let it produce a transcript in minutes.

With TranscribTxt the steps are:

1. Export your meeting recording (Zoom, Teams, Meet or any other source) as a file. There is no live meeting bot, so you upload the recording yourself and only transcribe what you choose.
2. Upload it to the video-to-text converter. It accepts MP4, MOV, WebM, MP3, M4A, WAV and other common formats and detects the language automatically.
3. Download the transcript as TXT for reading, or SRT if you want a timestamp on every line. Pro and the Week Pass add JSON with word-level timestamps.

TranscribTxt runs on OpenAI's Whisper large-v3 turbo, and the file is deleted after processing (a free-preview upload is kept up to 24 hours so you can finish it). It doesn't label speakers, so the next part covers attribution. Your first file needs no account, and a free account covers 3 recordings a day up to 30 minutes each. A longer meeting gets its first 30 minutes free; for the full recording, a one-time $5 Week Pass (7 days of Pro) or Pro removes the cap.

Who said what

Minutes are mostly about attribution: who committed to what, who raised which concern, who made the final call. A plain transcript tells you what was said, not always who said it. Three ways to close that gap:

• Use the platform's own transcript for names. If Teams or Zoom transcribed the call live, its transcript shows names for signed-in participants.
• Give the AI assistant the attendee list. In step 3, paste the attendees with the transcript and ask it to mark anything it can't attribute as unclear.
• Check owners against the recording. The SRT export puts a timestamp on every line, so you can jump to the moment an action item was agreed and hear who took it.

For background on automatic speaker labeling and where it struggles, see speaker diarization explained.

For source-specific steps, see our guides on transcribing a Teams meeting and a Zoom recording.

2. Use a clear minutes template

Before you summarize anything, decide on the structure. A consistent template makes minutes scannable and lets people find what affects them without reading the whole document. Use this one:

Meeting: [name]
Date: [date]    Duration: [length]
Attendees: [names]

Agenda / topics discussed:
- Topic 1
- Topic 2

Decisions made:
- Decision 1
- Decision 2

Action items:
- [Owner] — task — due [date]
- [Owner] — task — due [date]

Next steps / follow-up:
- Next meeting, open questions, blockers

The two sections that earn their keep are decisions and action items with owners. Discussion is interesting; decisions and owned tasks are what people act on. Keep everything else brief. Minutes are a record of outcomes, not a re-transcription of the conversation.

3. Let an AI assistant draft the minutes from the transcript

This is where the transcript pays off. Paste it into any AI assistant and ask for minutes in your template. Because the AI reads the whole transcript at once, it can pull out decisions and draft action items in one pass. Give it the attendee list so it can suggest owners.

A prompt that works well:

You are writing meeting minutes from the transcript below. Produce: (1) a one-line summary, (2) key decisions made, (3) action items as "Owner — task — due date", using the attendee names listed below; if you can't tell who took an item, write "Owner: unclear", and (4) next steps. Be concise and do not invent anything not in the transcript. Attendees: [names]. Transcript follows: [paste transcript]

A few tips that make AI-drafted minutes reliable:

• Always review the draft. AI can misattribute a line, especially when the transcript has no speaker names, or miss context only a participant caught. A human pass is worth the minute it takes.
• Ask it to flag uncertainty rather than guess. The "Owner: unclear" instruction above turns gaps into questions you can resolve instead of silent errors.
• Feed it the cleanest transcript you can. Accuracy in equals accuracy out, which is why an accurate transcription engine matters more than the summarizing model. Our AI transcription accuracy guide covers how to get the cleanest input.

If you'd rather a tool join your calls and draft minutes end to end, compare options in our roundup of the best AI meeting transcription software for 2026.

4. Use names and timestamps for follow-up

Two things make minutes genuinely useful after the meeting, not just during it.

Names make attribution trustworthy. When someone later asks "who agreed to own the launch?", you want a line you can point to, not a vague memory — confirm owners while the meeting is fresh, using the attendee list, the platform's own transcript or the recording.

Timestamps make follow-up easy. Pair each decision or action item in your minutes with its timestamp, for example "Decision to delay launch — 00:34:12". Anyone who wants the full context can jump straight to that moment in the recording instead of re-watching the whole thing. TranscribTxt's SRT export (every plan) gives a timestamp for every line, and JSON on Pro and the Week Pass adds word-level timing.

5. Store and share minutes so they stay searchable

Minutes only help if people can find them. After you finalize:

• Save the minutes and the underlying transcript together, in a shared doc, wiki or project tool, so the summary and the source live side by side.
• Use a consistent naming scheme like 2026-06-07 Product sync — minutes so search and sorting just work.
• Keep the SRT or TXT transcript attached. It makes the entire meeting full-text searchable, which is something a recording alone can never be.
• Share the minutes within a day, while the meeting is fresh and action items are still actionable.

Over time this turns into an institutional memory. Every decision your team ever made becomes a searchable line of text instead of an hour of video nobody will rewatch.

The short version

Transcribe the recording, drop the transcript into a clean template, let an AI assistant draft decisions and action items from it and the attendee list, then confirm owners, review and share. The recording becomes minutes in minutes.

If you want to try the transcription step, start with TranscribTxt for free: your first file works without an account, and a free account covers 3 files a day, up to 30 minutes each, no credit card. For longer meetings, Pro is $10/month ($72/year), or $5 buys a one-time 7-day Week Pass. Files are deleted after processing; a free-preview upload is kept up to 24 hours.