Last Tuesday I opened an episode page I had published three weeks earlier. The audio was fine. The title was fine. The description said: "Great chat with Alex about shipping." That was it. No chapters. No links. No quote. No transcript. A listener who wanted the bit about pricing had to scrub through 48 minutes.
I had "finished" the episode when I hit Publish. I had not finished the work that makes an episode usable after it leaves my headphones.
This post is about that missing half. Not how to record. Not how to host. How to turn one cleaned transcript into a small kit you can reuse every week: show notes, chapter markers, pull quotes, and a short written recap.
What an "episode kit" is
Think of five files (or five sections on one page) that all come from the same master text:
- Master transcript — clean read, speaker labels, checked names.
- Show notes — short description + 3–5 topic bullets + links.
- Chapters — timestamps with plain titles.
- Quotes — 2–4 lines checked against the audio.
- Recap — a short article for people who prefer reading.
You edit the master once. Everything else is cut from it. That is the whole trick. If you rewrite show notes from memory, you invent details. If you cut them from a checked transcript, you stay honest.
Start with the painful minute, not with "Step 1"
Most guides begin with "save a raw copy." Useful. Boring. Here is a better starting question:
What will a new listener do in the first 30 seconds on your episode page?
They skim. They look for a reason to press play. They look for a way to jump. Empty descriptions fail that test. So write the public-facing pieces first in your head, then build the transcript so those pieces are easy to cut.
A concrete example. Suppose your guest is "Siobhan" from a bakery in Lisbon, and she talks about menu pricing around minute 12. Your kit should make that findable:
- Show notes bullet: "How Siobhan prices a new menu without guessing"
- Chapter:
12:05 Pricing the new menu - Quote: one sentence she actually said about margins
- Recap heading: "Pricing without undercutting yourself"
None of that exists until the transcript is clean enough to search.
Gate A — freeze the raw export
Before you touch a word, export two things:
- A raw TXT (or JSON) dump.
- A timestamped copy if you have one (TXT with times, or SRT / VTT).
Name them clearly. Example: ep42-siobhan-raw.txt and ep42-siobhan-timed.srt. Then duplicate the raw file to ep42-siobhan-master.txt and only edit the master. If you mess up filler removal, you still have the original.
If you do not have a transcript yet, get one from the episode file first. A practical option is the browser podcast transcription page on whisperweb.dev. Official site: whisperweb.dev (not similarly named .app / .tech sites). Free local mode runs in the browser, so the audio stays on your device for that path; each free file can be up to 20 minutes and 200 MB, with no sign-up and no daily file cap. It supports 100+ languages. Longer episodes can be split, or you can use the optional Unlimited cloud path (up to 10 hours / 5 GB per file) when you choose upload. Either way: keep a timestamped export if you want chapters later.
Gate B — speakers before polish
Speech-to-text often dumps one long block. For a two-person show, listen to the first few minutes and mark turns:
HOST:
GUEST (Siobhan):
Tips that actually save time:
- Questions are usually the host. Long answers are usually the guest.
- New paragraph on every speaker change, even for "Right."
- For three or more voices, jot a one-line voice note ("Alex = softer, interrupts") and keep it beside the doc.
- Overlap? Write
[crosstalk]instead of guessing who won the fight.
Do not delete filler yet. Wrong speaker labels poison every quote you pull later.
Gate C — names, products, numbers
This is where published episodes get embarrassing. "Siobhan" becomes "Shivon." "Kubernetes" becomes three nonsense words. A price becomes the wrong year.
Make a tiny word list before you search-and-replace:
- Guest name and spelling
- Company / product names
- Places
- Jargon from the episode outline
- Any number you plan to quote (years, prices, percentages)
Then find near-miss spellings and fix them in one pass. A wrong number in a pull quote is worse than leaving the quote out.
Gate D — choose verbatim or clean read
Two honest styles:
- Verbatim — keep ums, restarts, repeats. Good for research archives.
- Clean read — remove filler and false starts, keep the speaker's meaning and voice. What most podcasts publish.
For a clean read, cut "um," "uh," filler "like," and stutter repeats ("I, I, I think"). Do not "improve" the guest into a TED talk. If they said "it was kind of a mess," leave it. Pick one style per show and stick to it so episode 40 matches episode 4.
Gate E — timestamps you can live with
A time code on every line is unreadable on a web page. For the published transcript, keep one stamp at the start of each speaker turn, or one every few minutes, in one format like [12:40]. Keep the full timed export for yourself. That private timed file is what feeds chapters.
Cut the kit (in this order)
1. Show notes (listener decision layer)
Write like a store window, not a diary.
- 1–2 sentences: who + what this episode is about
- 3–5 bullets in episode order
- Links mentioned on air
- Chapters (next)
Bad: "An inspiring conversation about business."
Better: "Siobhan runs a small bakery in Lisbon. We talk about pricing a new menu, hiring a first employee, and what she would redo."
2. Chapters (jump layer)
Open the timed export. Mark topic changes. Aim for roughly one chapter every 5–10 minutes. Titles should be specific, not cute.
00:00 Intro
03:15 Finding the first location
11:40 Pricing the new menu
24:05 Hiring employee one
37:30 What she would redo
Check your host's help docs for the exact description format. Many apps expect the first chapter at 00:00.
3. Quotes (share layer)
Hunt for lines that stand alone in one or two sentences. Copy from the master, then scrub the audio at that timestamp. Ask: does this still mean what she meant out of context? Add speaker + time so future-you can find it.
4. Recap (reader layer)
A recap is not the transcript pasted into a CMS. It is a short original piece:
- Its own title (what the reader learns)
- Three or four best ideas, each with a heading
- One or two checked quotes
- Links to the episode and the full transcript
If you draft the recap with an AI writing tool, read every sentence against the master. Delete anything the guest did not say.
A 45-minute ritual you can reuse
When the episode is already transcribed:
| Block | Minutes | Output |
|---|---|---|
| Freeze raw + timed exports | 2 | Two named files |
| Speaker labels (first pass) | 10 | Labeled master |
| Names / jargon / numbers | 8 | Corrected master |
| Filler pass (clean read) | 8 | Publishable transcript |
| Show notes + chapters | 10 | Description-ready text |
| Quotes + recap outline | 7 | Social + blog seeds |
Longer interviews need more time on speakers and names. Panels need more. The point is a repeatable budget, not a hero night once a quarter.
Checklist (print this)
- [ ] Raw + timed copies saved with clear filenames
- [ ] Speakers labeled; crosstalk marked
- [ ] Names, products, jargon, numbers checked
- [ ] Filler policy chosen and applied consistently
- [ ] Published timestamps simplified; full timed file kept
- [ ] Show notes written with real bullets and links
- [ ] Chapters in the host's required format
- [ ] Quotes checked against audio
- [ ] Recap drafted and fact-checked against the master
FAQ
Do I need both a transcript and a recap?
If you can, yes — and keep them separate. The full transcript helps people who are deaf or hard of hearing, and anyone who wants the exact wording. The recap is easier to share, but it leaves things out, so it cannot replace the transcript. A practical layout: show notes and chapters at the top of the episode page, a short recap in the middle, full clean-read transcript at the bottom or on a linked page. Short on time? Ship the transcript first. Add the recap later.
What if my episode is longer than 20 minutes?
Split the audio at quiet pauses into chunks that fit your tool's free local limit (20 minutes / 200 MB per file on whisperweb.dev's free path), or use a cloud workflow that accepts longer files when you are comfortable uploading. Overlap a few seconds between chunks so you do not lose a sentence on the cut. Name files with the time range (ep42-20-40.mp3) so timed chapters still map back to the full episode.
Should quotes be polished?
No. Fix obvious ASR errors in names and numbers. Do not rewrite the guest's voice. If a line only works with five sentences of context, it is not a pull quote — put the idea in the recap instead.
Where should the transcript live?
Wherever listeners already look: the episode page, a /transcript URL, or your newsletter archive. Link it from show notes. Search engines and humans both benefit from a stable URL.
Bottom line
Empty episode pages waste good audio. One cleaned master transcript is enough raw material for show notes, chapters, quotes, and a recap — if you edit in gates and cut in order. Build the kit once, then reuse the ritual every week.
Written for independent podcasters and newsletter writers. Tools mentioned: whisperweb.dev — browser-first speech-to-text at the official whisperweb.dev site.
Top comments (0)