Uploaded Recordings That Know Who's Talking

Uploaded Recordings That Know Who's Talking

Fix speaker labels and proper nouns on an uploaded recording before generating anything, or every quote stays unusable.

fireflies-ai transcription uploads speaker-labels interviews archive

What it is for

An ingest checklist for recordings you upload rather than record live: name the speakers in the first thirty seconds, correct the labels and mangled proper nouns before generating anything, and name the file for how you will search later. Use it for interviews, calls and field recordings. It prevents an archive of unattributable quotes.

The skill

Use this when you upload an existing recording to Fireflies - an
interview, a site visit, a phone call, a webinar - instead of having
the notetaker join a live meeting.

Fix the speakers before you generate anything. Every summary, quote
and action item inherits the labels, and "Speaker 2" is worthless in
a document six months from now.

Before you record, when you can:
- Have each person say their name and role in the first thirty
  seconds. "Maryam, operations." Highest-value ten seconds in the file.
- One microphone per person if the room allows. Two people on one mic
  merge into one speaker and nothing separates them afterwards.
- Say the date and what the meeting is, out loud, at the start.

After the upload finishes:
- Correct the speaker labels first. Before you share, before you
  summarise, before you clip.
- Scan the first minute and the last two minutes. Openings are
  crosstalk; endings hold the real commitments.
- Fix people, company and product names by searching the transcript
  for the mangled spelling. Transcription gets proper nouns wrong
  consistently, so one search fixes many instances.

Only then generate the summary. If you generated one before fixing
labels, generate it again - it will not update itself.

Name the file the way you will search for it later: date, counterpart,
subject. "2026-03-04 Acme renewal call", not "New Recording 7".

Write one line into the notes stating that the other party consented
to being recorded. In many places you need it.

The failure this prevents: an archive of Speaker 1 and Speaker 2 where
every quote you find is unattributable, and therefore unusable.

How to install it

1. Open the tool and start a new conversation, project or template.
2. Paste the text below wherever the tool keeps standing instructions
   — usually called system prompt, instructions, or persona.
3. Save it so it applies to every run rather than a single message.

Did it work for you?

Where it came from

Skills app Original