Every episode's edit screen has a Transcript and AI box with three buttons. They use the AI keys you already saved on AI Connect, and nothing runs until you press one.
Button 1, make the transcript:
This needs an OpenAI or a Google Gemini key, because Claude cannot listen to audio.
- Google Gemini costs about 7 cents for an hour of audio.
- OpenAI (whisper-1) costs about 36 cents for an hour of audio, 0.6 cents a minute.
- With both keys saved, Gemini is used unless you choose otherwise on the Podcasts tab.
A long episode is sent in parts of about 8 MB, roughly 8 minutes of a normal MP3, each its own short request, so no host or Cloudflare time limit can cut it off. A popup counts the parts: keep the tab open until it says Done. The times in the transcript run from the start of the whole episode. An episode longer than one part needs to be MP3, because MP3 can be cut cleanly between sounds. A shorter one can be MP3, M4A, WAV, OGG or WEBM. Only have a Claude key? Paste the transcript in yourself, and the next two buttons still work.
Button 2, write show notes, chapters and SEO:
Writes the show notes, the chapters and the search details from the transcript, with whichever writing AI you use (Claude, OpenAI or Gemini), in the voice you set on Train Your AI. It costs a few cents.
Button 3, make a blog post from this episode:
Writes a full draft post that teaches what the episode teaches, with the episode's player at the top. It is saved as a draft: nothing is published until you publish it.