Prompts to summarize a transcript: meeting minutes, notes, show notes

What each prompt asks for

• Meeting minutes asks for the topics in order, the decisions and the open questions. The date and attendees are included only if the transcript states them.
• Action items and decisions lists each one with an owner and a deadline only where the transcript gives them, and otherwise says “owner not stated”. That wording is deliberate: an assistant left to fill the gap will often invent a name.
• Lecture notes asks for an outline, the key terms as the speaker defined them and review questions the transcript can answer.
• Interview summary asks for a paragraph and the themes, each with short verbatim quotes and their timestamps when the transcript has them.
• Podcast show notes asks for a description, topics with timestamps and the names or books mentioned, exactly as said.
• Clean up is the odd one out: it asks only for punctuation, capital letters, paragraph breaks and the removal of filler words, and forbids any other change, so the result is still the speaker's words.

Write minutes of this meeting: ...
Keep the answer to about 400 words.
Use headings and bullet points.
Answer in the language of the transcript.
Use only what is in the transcript. Do not add facts, names, numbers or quotes that are not there.
If a passage is unclear or looks like a transcription error, mark it [unclear] instead of guessing.

Transcript:
"""
[your transcript]
"""

Why the guard lines are in the prompts

A language model asked to summarise will, if it can, write something fluent, and a fluent summary can contain a number or a name that nobody said. The two guard lines tell it to stay inside the transcript and to flag what it cannot read. They reduce the problem; they do not remove it. (The clean-up prompt has its own guard, because it must not add anything, not even an [unclear] mark.) The check that matters is yours: compare the summary with the transcript.

Getting a better transcript first

A summary is only as good as the text it starts from. Names and technical terms that the speech recogniser got wrong will be wrong in the summary too. Correct them in the transcript first. If your transcript has a time on every line, you can keep them for the quotes option, or strip them with remove timestamps. To trim “um” and repeated words before summarising, use remove filler words.

What this page does not do

It does not run a model, store your text or send it anywhere, and it makes no promise about what an assistant will answer. It does not know which assistant you use or what it can handle. If you want the transcript itself, start with the audio to text page; the transcript is the part TryScriptorium produces.