Meeting recordings → minutes your model can quote

MakeItMarkdown · July 2026 · 5 min read

Every recorded meeting already has a transcript — the caption track your video tool generated automatically. Almost nobody uses it, because raw .srt is two-thirds numbering and timing apparatus. Ten minutes of workflow turns that dead file into minutes with receipts and a searchable meeting archive.

1 · Get the captions out

Meeting and video platforms export captions as .srt or .vtt (usually under the recording's download or accessibility options). That file is the whole meeting, timestamped to the second.

2 · Convert

Drop it here. The subtitle mapping strips sequence numbers and styling tags, flows caption fragments into readable lines, and keeps one compact [23:14] marker per block — the part that makes everything downstream quotable. A one-hour meeting typically lands well under a chat-paste budget once the apparatus is gone.

147
00:23:14,200 --> 00:23:17,900
so the vendor shortlist is
approved by everyone? great

148
00:23:18,100 --> 00:23:21,400
then Sam owns the contract
draft by Friday

[23:14] so the vendor shortlist is
approved by everyone? great then Sam
owns the contract draft by Friday

3 · Minutes with receipts

Paste the converted transcript and ask for exactly the minutes you wish someone had written:

From this transcript, produce:
1. Decisions made — with the [mm:ss] marker where each was agreed
2. Action items — owner, deadline if stated, marker
3. Open questions that were raised but not resolved
Quote the transcript for anything ambiguous.

The time markers do two jobs: they let anyone jump into the recording to verify ("did we really agree to that?"), and they keep the model honest — an answer that must cite [41:30] can't drift far from what was said at 41:30.

4 · The meeting archive

Converted transcripts are ideal workspace citizens: name them 2026-07-03-vendor-sync.md, upload to your Project or Space, and "what did we decide about the shortlist?" becomes answerable across months of meetings — with markers back into the recordings. (Why Markdown retrieves so much better than raw files: the workspace guide.)

5 · Honest limits

• Auto-captions mishear names and jargon — minutes inherit the captioner's errors; the markers make spot-checking cheap.
• Speaker attribution exists only if your captioner writes names into the text (many do; subtitle formats carry no speaker field).
• A malformed block here and there is normal — they're skipped and counted in the fidelity report, not silently absorbed.

Also see: the .srt format guide