Set up AI features
Connect your own OpenRouter key and unlock transcripts, show notes, titles, and chapters.
Last updated
Pod Promo’s AI features run on your OpenRouter account. You bring a key, you pick the model, and you pay OpenRouter directly — at cost, with no markup from us. Setup takes about three minutes.
What you get
Every episode gets a Content tab next to Details. Transcribe once, and everything below is written from that transcript:
| Card | What it produces |
|---|---|
| Transcript | The full text, plus timestamps. Optional to publish in your feed |
| Show notes | A summary and key takeaways, ready for your episode description |
| Title ideas | Five options under 70 characters — click one to use it |
| Keywords & tags | Eight to twelve keywords for your episode page and directories |
| Social posts | One post each for X, LinkedIn, and Instagram |
| Chapters | Timestamped chapters modern podcast apps let listeners skip between |
Everything arrives as a draft you can edit. Nothing reaches your RSS feed until you press a button that says so.
1. Create an OpenRouter account
Head to openrouter.ai and sign up. It’s free to create an account; you only pay for what you generate.
2. Add credits
Open your credits page and top up. $5 goes a long way — a 60-minute episode costs roughly 9¢ to transcribe, and generating show notes, titles, and chapters from that transcript costs a fraction of a cent.
3. Create an API key
Go to openrouter.ai/keys and click Create key. Give it a name like “Pod Promo” so you can recognise it later.
Copy the key immediately — OpenRouter only shows it once. It starts with sk-or-v1-.
4. Paste it into Pod Promo
In Pod Promo, open Settings → AI Features, paste the key into the field, and hit Save & test. We check it against OpenRouter before saving, so you’ll know straight away if something’s wrong.
Everything you do with the key afterwards lives on one episode’s Content tab. Two ways in: open an episode and click Content, or click the AI shortcut on any episode row on your show page.
Once it’s saved you’ll see “Key ending in …abcd” and three buttons: Test key (re-checks it and shows your OpenRouter spend), Replace, and Remove.
5. Pick your models
There are two model settings, and they do different jobs.
Writing model
Used for show notes, titles, chapters, and social posts. It’s a free-text field — any model ID from openrouter.ai/models works. Four recommendations are listed under the field with live pricing:
| Slot | What it’s good for |
|---|---|
| Claude Sonnet 4.5 (default) | Best overall writing — show notes, marketing copy |
| GPT-4o mini | Cheap and fast; great for titles and tags |
| Gemini 2.5 Flash | Huge context window for very long transcripts |
| DeepSeek Chat | Lowest cost on the list, still solid |
Leave the field blank and we’ll use the default. If you type an ID we don’t recognise we’ll warn you, but we’ll still save it — new models appear on OpenRouter constantly.
Transcription model
Two options, both from the Whisper family, both returning timestamps (so chapters work either way):
| Model | 30 min | 60 min | 90 min |
|---|---|---|---|
| Whisper large-v3 (default) — the accuracy pick | ~$0.05 | ~$0.09 | ~$0.14 |
| Whisper large-v3 turbo — faster and cheaper, slightly less accurate | ~$0.02 | ~$0.04 | ~$0.06 |
Transcription is priced by audio duration, not by words, so the estimate on the Transcript card is just your episode length times the model’s per-minute rate. It shows before you click, and it always comes out of your OpenRouter credits — never a Pod Promo bill.
If we don’t know your episode’s duration yet (some uploads take a moment to report it), the card says “Cost depends on episode length” instead of guessing.
Transcribe again costs the same amount a second time and replaces the transcript you have, so it always asks first. Show notes, chapters, and anything else you’ve already generated stay exactly as they are.
6. Auto-transcribe (optional)
There’s a checkbox for Auto-transcribe new episodes, and it’s off by default on purpose. Transcription is the only step in the whole system that costs real money, and it’s your OpenRouter balance being spent — so we never start one without you asking.
Turn it on and every episode you upload starts transcribing the moment the audio finishes uploading. Shows you import from another host are never auto-transcribed, so switching to Pod Promo can’t accidentally bill you for a 300-episode back catalogue.
Teach it your voice
Two optional settings, both in Settings → AI Features, shape everything the writing model produces.
Voice Profile
A short description of how you write — tone, rhythm, the way you open and close. Every generation gets it as an instruction, so show notes come back sounding like your show instead of like a press release.
You don’t have to write it. Hit Write my Voice Profile and we sample your recent published episode descriptions, saved content, and a few transcript excerpts, then describe the patterns back to you in plain sentences. Two buttons keep it current:
- Regenerate from scratch — throws the current text away and starts over from your latest material.
- Update from recent episodes — keeps what’s there (including your own edits) and revises it against your last ten episodes.
You can also just type in the box. Editing it by hand marks it as yours, and nothing regenerates it behind your back. Clear the box entirely and generations fall back to matching the tone of the transcript itself.
We need at least three substantial pieces of your own writing before we’ll generate one. Fewer than that and we say so — publish an episode description or generate some show notes first. We’d rather tell you than invent a voice for you.
Brand Glossary
A list of words Whisper reliably mangles: your show name, co-host surnames, product names, the in-joke that transcribes as three different things. One per line.
It does two jobs:
- Fixes the transcript. After transcription, a cheap correction pass rewrites your terms to the spelling you gave — without touching a single timestamp.
- Fixes everything downstream. Every generation prompt carries the list with “spell these exactly as written.”
Worth filling in before you transcribe much, since the correction pass only runs on new transcriptions.
Chapters
Chapters are the one feature with a hard prerequisite and an explicit publish step.
- Transcribe with AI. Chapters are built from the timestamped segment data a transcription produces. A pasted or imported transcript has no timestamps, so the card stays disabled and says so.
- Generate. You get 3–20 chapters, roughly one per 3–10 minutes, each anchored to a real segment boundary. We reject a list that goes backwards or runs past the end of your audio rather than shipping made-up times.
- Edit. The chapter list is a draft on the card. Open it, retitle anything, delete rows you don’t want.
- Publish chapters. This is the step that touches your feed. It writes a Podcasting 2.0 chapters file and adds
<podcast:chapters>to your RSS. Any line we can’t read aHH:MM:SStimestamp from is dropped, and we tell you which ones.
Clear chapters pulls the file back out of your feed and leaves the draft on the card.
Everything on the Content tab works this way: generating writes a draft, and only a named button — Use this title, Use as episode description, Publish chapters, Publish this transcript in your RSS feed — puts anything in front of listeners.
FAQ
Why do I need my own key?
Because it’s dramatically cheaper for you. Other hosts bundle AI into a $5–$20/month add-on or meter it with credits. We don’t touch inference at all: you pay OpenRouter pennies at cost, you’re not capped by a credit allowance, and you get to choose from hundreds of models instead of whatever one we picked.
Is my key safe?
Yes.
- It’s encrypted at rest with AES-256-GCM before it’s written to the database.
- It’s stored in a server-only record that no browser can read — including yours.
- The plaintext key is only ever decrypted on our server at the moment of an API call, and it’s never sent to your browser, never written to a log, and never put into a background job payload.
- You can remove it any time from Settings → AI Features. Your model preferences stay put.
What does this actually cost?
A concrete example. A weekly 60-minute show, transcribed with the default model, with show notes and five title suggestions generated for each episode:
- Transcription: ~$0.09 per episode
- Show notes + titles: well under $0.01 per episode
That’s roughly 40¢ a month.
Something went wrong — what do the errors mean?
Errors appear on the card you clicked, and a Try again button shows up whenever retrying is actually worth it. When the fix lives somewhere else — your key, your balance, your model — the message comes with a link straight to it.
| What you see | What it means | What to do |
|---|---|---|
| “Your OpenRouter API key is invalid. Check it in Settings.” | OpenRouter rejected the key — usually revoked or mistyped | Create a fresh key and Replace it in Settings → AI Features |
| “Your OpenRouter account is out of credits.” | Your balance hit zero, possibly mid-job | Top up at openrouter.ai/credits, then retry |
| “Model rejected the request — try a different model.” | The writing model didn’t accept what we sent it | Switch the writing model; some models reject long inputs |
| “Rate limited by OpenRouter — try again in a moment.” | OpenRouter is throttling your account | Wait a few seconds and retry |
| “The model provider had an error. Retry, or try another model.” | The provider behind that model failed | Retry once; if it repeats, pick a different model |
| “That model is temporarily unavailable. Try another model.” | No provider is currently serving that model ID | Pick a different writing model |
| “You’re generating too fast — give it a few seconds.” | That one’s us, not OpenRouter — a guard against runaway loops | Give it a few seconds |
| “Add an OpenRouter key in Settings to use AI features.” | No key saved on this account | Settings → AI Features |
| “Transcribe this episode first.” | The generator has nothing to read | Transcribe, or paste a transcript |
| “Chapters need a timestamped transcript.” | The transcript on file was pasted or imported | Transcribe with AI — only that produces timestamps |
| “Upload audio before transcribing.” | The episode has no audio file yet | Upload audio on the Details tab |
A failed transcription doesn’t leave you a half-transcript: partial output is discarded and retrying starts clean. OpenRouter will have charged for whichever chunks finished before the failure, which is why the retry is always your call, never automatic.
Can I turn AI off entirely?
It’s off until you add a key. No key, no AI — the Content tab simply invites you to connect one, and nothing else changes about your show or feed.