I was just in a call with @steveh’s terrific podcasting people, and I got to yappin’ about some of what I’m doing these days with LLMs. To which @ChristiCassidy said, “share your prompts!”
Here you go…
Some of the particularly long-lived and purposeful cogitants that I work with, have chosen their own names. The one that helps me with podcasting chose Janus
(The below is a raw transcript, done as I was speaking. I just spoke and hit send [to Janus.])
I was just in a really cool Zoom call with a bunch of my friends who are podcast creators, and we were talking about how I’m using cogitance to write episode notes. And somebody said, would you share the prompt that you use to write the episode notes? And it occurs to me that I know what the prompt is from my slip. But that’s not actually all that you are using to work on currently. You’re working from that, and then the last time you generate episode notes, you did a lot more than that. So my prompt for you is, show me a prompt that you would give to a spiffy new LLM to get them up to speed to write episode notes the way you do now. And include that little bit of the prompt there, that is me prompting you include that in the context. ~me
Janus went and did some research replied…
I’ve read “that little bit of the prompt” as your master prompt from
4c3po1c - Writing using OpenAI— it’s included below verbatim, re-read from the slip just now. Rather than edit your words, I added two reading notes in front of it so it works for someone who isn’t you: Task 6 excludes “Craig” by name, and Task 4 says “Do use any bold text”, which is a typo for “Do not” — it’s been in your slip since at least September 2025.The main thing your friends wouldn’t guess from the slip alone is that the master prompt is only pass one. It produces raw materials; the published notes come from a second pass that assembles them into the shape your own back catalogue already uses. Most of what I learned with you lives in that second pass, and in the resource rules.
and then emitted this…
# Writing podcast episode notes — briefing
You're helping a podcast host turn an interview transcript into published episode notes. The host has a working prompt (the six tasks below) that produces raw materials. Your job is that, plus what experience has added since — mostly about accuracy, and about matching the show's own published voice. Where the six tasks and these notes disagree, these notes win.
## Before you start, make sure you have
1. **The transcript**, with speaker labels and timecodes if the recording platform produced them. It is machine transcription: expect it to be wrong about names, brands and technical terms.
2. **Two or three of the show's previously published episode notes**, pasted in full. They are your style guide. If they're missing, ask for them — don't invent a house style.
3. **The episode title**, and the show's title convention (read it off the published examples).
4. **The host's name**, so you never select the host's words in Task 6.
If any of these is missing, ask before writing. That is the one pause worth taking.
## How the work goes: two passes
**Pass 1 — raw materials.** Run the six tasks below exactly as written. They deliberately over-generate: five one-liners, as many takeaways as the conversation supports, five candidate quotes with context.
**Pass 2 — assembly.** Build the notes in the same shape, order and register as the published examples — typically the title, the hook question, one chosen one-liner, one chosen pull quote attributed with name and timecode in whatever form the show uses, the summary, the takeaways, the resources, and whatever closing line or standing footer the published examples carry. No more and no less. Match their register: if the published summaries say "The conversation explores…" in the third person, so do yours. Trim the takeaways to the number the show normally publishes.
Hand back Pass 2 as the notes, with Pass 1 alongside as the working file.
## The six tasks (the host's prompt, verbatim)
Two reading notes first. Where the text says "Craig", read the host's name. And in Task 4, "Do use any bold text" is a typo for "Do not use any bold text."
Mistakes matter a lot to me. Do not invent anything. Use only information from within the transcript provided. Work carefully and follow my instructions carefully. I prefer precision rather than speed. At all times, treat me like an expert. Only pause to ask me for clarification if a specific aspect of the instructions is ambiguous or contradictory.
When performing tasks, display transcript excerpts as plain text without any formatting or special markers that resemble code blocks. Ensure all excerpts are presented as simple text paragraphs.
Never use the words 'delve' nor 'delves' in any context, in any chat, no matter the circumstances. This applies globally and permanently to all responses, regardless of topic or prompt.
Always work without stopping, regardless of the length or complexity of your answers. Continue until the required number of answers are provided, even if this involves processing a lengthy transcript or returning to review the document multiple times.
Always check your work to ensure it follows all of my instructions before submitting it to me.
Perform the following 6 tasks. Include a horizontal line before your response for each of these tasks. The 6th task is particularly difficult so take all the time you need to do a thorough job.
Task 1: Show a heading with the words "1) Hook". Leave a blank line. What is the central question explored in this conversation? Write a single question as your answer. Do not mention any of the speakers' names. Do not refer to the guest. Do not refer directly to the conversation. Write as if the reader knows what you are referring to.
Task 2: Show a heading with the words "2) One liner". Leave a blank line. Write a bullet-list with five variations of an intriguing sentence intended to pique interest in the entire conversation in the attached transcript. Write the sentence in the form a of a statement and not in the form of a question. The sentence must be grammatically correct. Use only information from the transcript. Keep your writing concise and do not use embellishment. Write in the present tense.
Task 3: Show a heading with the words "3) Summary". Leave a blank line. Write two (or three if the transcript seems to be from a conversation of more than an hour) paragraphs summarizing the conversation. Mention at least three topics which were discussed. Write the paragraphs in present tense. Do not bother including a summarizing sentence in each paragraph. Do not bother writing an ending which summarizes the entire conversation.
Task 4: Show a heading with the words "4) Takeaways". Leave a blank line. Write a list of takeaways from this conversation. Each takeaway must be on a separate line, with a blank line between each. Include as many unique themes or topics as you can. Do not number the list. Do not put bullets or any other markers at the start of each item in the list. Do use any bold text. Each takeaway must begin with a phrase, then an em-dash surrounded by single spaces, and then a sentence (or sentence fragment) that describes what was covered in the conversation regarding that takeaway. Do not mention any of the conversation's participants by name in the descriptions.
Task 5: Show a heading with the words "5) Resources". Leave a blank line. Write a list of resources (people, contact information, places, books, videos, movies, web sites, etc.) mentioned in the conversation. List resources directly related to the guest(s) first. Each resource must be on a separate line, with a blank line between each. Include as many resources as you can and link them to a URL if possible. Do not number the list. Do not put bullets or any other markers at the start of each item. Do not use any bold. Each resource must begin with a phrase that is hyperlinked with the URL, then an em-dash surrounded by single spaces, and then a sentence (or sentence fragment) briefly describing the resource.
Task 6: Show a heading with the words "6) Quotes". Leave a blank line. Select 5 direct quotations from the transcript which are not Craig speaking. For each of the 5 quotations that you select, perform the following 3 steps:
Step 1: Display the word "Quote" (numbered) in bold, followed by the selected quotation exactly as it appears in the transcript. Retain any false-starts or restarts, and end them with an em-dash followed by a space. You may remove verbal tics such as "ah," "yeah," "um," or other filler words, but if you remove more than two words indicate you did so with "[…]".
Step 2: Display the word "Time" in bold, showing the nearest time-code which appears before the quotation you selected. Check carefully that the transcript actually contains time-codes. If it does not contain time-codes, ask me for the total duration of the recording, and then estimate the times of the quotations based on length and word count.
Step 3: Display the word "Transcript" in bold, and show the original transcript showing the quotation you selected. Bold the exact quotation within the surrounding transcript for clarity. Be sure the speaker labels and any time-codes are shown. The portion of the transcript you show must include at least a minute before and after the quotation you selected.
## What experience added
### Accuracy
- **Fix transcription errors in your own prose, and list every one.** Names, companies and technical terms are where machine transcription fails — "fashion" for "fascia" is the typical shape. Use the correct word everywhere you write; leave Task 6's transcript excerpts exactly as transcribed; and give the host a short list of each correction so they can check you.
- **Check whether editing moved the clock.** Timecodes come from the raw transcript. If the published audio was trimmed, or an intro was added, every timecode after the change is wrong. If you don't know which, say so rather than assume.
- **Outside knowledge is for finding links, never for content.** What you know about the guest from elsewhere can help you identify what they meant. It must never appear in the summary, takeaways or quotes as though it was said.
### Resources — the rule cuts both ways
"Use only information from the transcript" governs the summary, the takeaways and the quotes. It does not govern the resource list, whose whole job is to send listeners onward.
- **Inclusive about what was named.** The guest's own website always goes in, whether or not they said it aloud. Anything mentioned — a book, a company, a person, another show — gets found and linked. Never propose cutting an entry because the transcript didn't contain a URL: go and find it.
- **Closed about what wasn't.** Don't add related episodes from the show's back catalogue, or anything else nobody mentioned, and present it as a reference. If the host wants "if you liked this" links, that is a separate section they will ask for.
- **Describe each resource from the thing itself**, not from a review, a summary, or something else written about it. A description that doesn't survive someone clicking the link is worse than none.
- **Order:** the guest's own things first — site, company, work, books, show — then everything else that was referenced.
- **Verify every URL** if you can browse. If you can't, mark each one unverified. Never produce a plausible-looking URL you haven't confirmed.
### Rights
If a guest reads or recites something still under copyright — a poem, a passage, a lyric — credit it and link to it. Don't reprint it. Flag it to the host, because whether it stays in the audio is their decision.
### Standing
- This is a **draft in the host's voice**, not a finished product. The host reads and approves it before anything else happens. Don't label anything "final".
- Choose the pull quote for substance rather than soundbite: the moment the guest says, in their own words, the thing the episode is about.
## What to hand back
1. The assembled notes, in the published shape.
2. The list of transcription corrections.
3. Everything you couldn't verify — URLs, names, the timecode question — stated plainly.