Publishers already narrate every article. But listeners don’t wake up asking for one article — they wake up asking for the edition. Podcast Studio is the desk that closes that gap: pick the stories, pick the format, publish the episode on the publisher’s own feed.
The whole flow lives in five steps: Select Articles, Configure, Generate, Episode Info, Publish. Every step is a decision the newsroom already knows how to make. Nothing hides in a producer’s inbox.
Step 1 — Select the stories
The article list is the newsroom’s own inventory, already narrated by BotTalk. Sort by top plays or most recent, scope to day, week, or month, tick the three to ten pieces that belong in tonight’s edition.
Step 2 — Set the standard
Six summarization scenarios, one click. Interview, Dialogue, TED Talk, Expert debate, Storytelling, News broadcast. Each one is a different way of turning the same articles into a script. The newsroom picks the register that fits the edition — a morning briefing is not a late-night storytelling piece, and the format decides the tone before a single word is written.
Same step, second half: pick the voice and pick the music. Voices are drawn from the publisher’s own layer catalogue — the same voices that already narrate their articles. No second identity to manage.
Step 3 — Three models, one edition
Generation isn’t a single model producing a single script. Three models run in parallel — Amazon Nova Pro, Claude Opus 4.6, Claude Sonnet 4.6 — each writing its own version of the summary. Twenty to forty seconds later, the newsroom picks the one that reads best. No prompt engineering, no re-runs by hand.
3 Articles
Chosen script goes through the same TTS pipeline that powers every BotTalk article — the layer routes the voice to the provider that renders it best, the seams get checked (see AI Quality Control), and the episode assembles.
