Podcasts say a great deal that the machines never hear.
The premise. CorpusCast is a content layer for video podcasts. For every episode a show releases, it produces the written record the episode never had: a full transcript with speaker labels and timecodes, a companion article, chapter markers, timecoded key moments, quote cards, and outbound links to the people, companies, and products the guests actually named. It publishes into whatever site the client already runs, on the client’s own domain and under the client’s own branding.
Why now. People are putting their questions to assistants, and an assistant can only quote text it can retrieve. A show that exists solely as audio and video gives it nothing to quote. Only 3.9% of the episodes in the public podcast index carry a transcript in their feed. Captions may sit on a video platform, but they are not on the show’s own site and they are not a page anything can index. Whether filling that gap earns citations is a bet rather than a finding. Google says its AI retrieval runs on the same ranking systems as ordinary search, and no study yet shows that publishing a transcript gets a show quoted. The mechanism is plausible and unproven, and moving before it is proven is the bet.
The wedge. The branded episode page is already a commodity. Podpage ships it for $39 a month on annual billing, with guest pages and, in beta, automatic internal linking across the archive. What Podpage refuses to do is where the work is: it does not transcribe, it says so publicly, and it never translates a customer’s own words. So the product is transcription, translation into languages the show has never been heard in, and outbound entity resolution, meaning every company, book, and person mentioned resolved to its real address on the web, consistently across an entire back catalogue. Entity resolution is the part that compounds. Every episode processed makes the next one cheaper to link. The input is the video master rather than the audio track. Every writing tool in this category transcribes the audio and throws the pixels away, so none of them reads a slide, catches a name on screen, or pulls a still worth putting in the article.
The market. The buyer is agencies. Production agencies, networks, and the in-house teams behind branded and B2B shows, all of them running several shows at once and already carrying a budget line for written deliverables. That line is measurable: a published agency rate card puts the writing bundle at $210 to $230 an episode above editing alone. Hobbyists are out of scope on purpose. Of the 4.7 million feeds in the public index, only 6.9% published anything in the last thirty days.
Where it stands. Nothing is built. There is no pipeline, no page, no client, and no code. This is the idea written down before the work starts, on the record with a date. The first real test is one show’s entire back catalogue run end to end with the links checked by hand, and five agencies asked whether they would drop the writer they already pay.