I Tried Letting AI Handle All My Meeting Notes, and It Actually Works
For the past two weeks, I've been running team meeting recordings through STT, then feeding them into a summarization prompt to produce meeting minutes.
At first, action items and small talk were mixed together to the point of being unusable, but once I locked the prompt down to extract only three items—'decisions / owners / deadlines'—it immediately became usable. What used to take a person 30 minutes dropped to 5.
What still doesn't work, on the other hand, is distinguishing between multiple speakers and company-specific terminology. Adding a separate abbreviations glossary helps somewhat, but it still takes a fair bit of manual work.
8 answers
Decisions / owners / deadlines—that trio is the real key, fr.
If you don’t put those in as fields, you just get a wall-of-text summary and nobody reads it.
I've been using it for about 3 months in a similar way, and speaker diarization is pretty much solved for meetings where each mic track is recorded separately. For online meetings, splitting the audio by participant and running STT boosts accuracy significantly. As for in-house terminology, I just create one abbreviation glossary file and paste it at the very top of the prompt. Once you make it, you can keep reusing it, so I think the initial effort is worth it.
Well, isn't that 5 minutes ultimately time someone has to review anyway? I used it for a month and eventually went back to semi-automated. If the summary comes out wrong, cross-checking it against the original to find the error can sometimes take longer than just writing it from scratch. Especially for decisions involving numbers or schedules, I always end up cross-checking no matter what.
Got a source? Curious what tool it is
Even if you make an acronym glossary, it can't keep up with newly coined terms lol. Our team ends up updating ours every quarter 😢