Podcast guest pitches from episode transcripts

The problem

Visualping’s B2B marketing reaches many segments, from law firm research teams to insurers, and each audience listens to different podcasts. Podcast directories let you search show titles and descriptions, which tell you what a show says it covers. To learn whether recent episodes touched your topic, or whether the show books outside guests at all, you’d have to listen.

Exa has announced that it indexes more than 3 million podcasts, with transcripts updated daily, so episodes can be searched by what’s said in them. That announcement on X is what started this run.

An agent that found the shows, wrote the pitches and sent them would hold both keys. A host could hear from us before anyone at Visualping had read what we were proposing, and a wrong contact or a generic pitch would go out under our CEO’s name or mine.

The system

Grok Bot did the research and the drafting, and the run moved through six steps.

1. It read Visualping’s B2B use case pages and an internal segment document. For each segment it wrote down the buyer and their problem, a few talk angles phrased like episode titles, the podcast niches that audience listens to and phrases to search for. It also wrote a few angles that apply to the whole company.
2. It searched Exa’s podcast index segment by segment. The phrasing that worked best imitates the top of a transcript, as in Podcast episode transcript Hosts Guests: <topic>. Each result is a podcast library page with the hosts, the guests, the date and the full transcript, so it was quick to tell whether a show was still publishing and whether it takes guests.
3. Each show got a fit score, with extra weight if it served one of this year’s priority segments, and shows that charge guests to appear were scored lower. Shows that didn’t fit were filtered out, along with shows our guests had already appeared on, and the best fits were picked for pitches.
4. It looked for each show’s contact in its RSS feed, where the itunes:email and managingEditor fields often hold one, and then in show notes and on guest application forms. It wrote the source next to every address or form it found.
5. It wrote one pitch per show. Each pitch opens with a real recent episode, proposes an episode title written for that segment, gives a few talking points and one line of credibility built from public facts, and ends with a soft ask. Where a show takes guests through a form, Grok Bot wrote answers for the form’s fields.
6. The humanizer-vp skill went over every pitch to remove the patterns that make writing read as machine-made, with facts, figures and contacts left as they were.

The episode titles the topic map proposed were written for each audience. For law firm research teams it suggested “The $600-an-hour refresh button”, and for e-commerce brands it suggested “How brands catch MAP violations across resellers without manual checks”.

The two keys

Everything up to a finished draft is the agent’s work, from the topic map through the searches, the scores and the contact lookups to the pitch itself.

Every pitch is drafted and waiting for my review. I’ll read each one and send it myself, by email or through the show’s guest form, and none of the sending is automated. The pitches are signed with my name, including the ones that propose Visualping’s CEO as the guest, and each one asks a host to give one of us a spot on their show.

Because the agent noted where it found every address, I can see that each one comes from the show or its host. None of them were guessed or bought from an enrichment tool.

What it produced

The run produced a ranked shortlist of podcasts across the segments and the company-wide angles, with a drafted pitch for each of the best fits. Nearly every top show publishes a way to reach it, usually an email in its feed or show notes or a guest form, so the run didn’t need any paid enrichment.

The pitches are waiting for my review and nothing has been sent or booked, so this page reports no appearances. I’ll add any bookings here once shows reply.

What I’d change

I’d have the agent settle whether a show charges guests before it ranks the show. A few shows on the long list had sponsored segments or bonus episodes that might be paid slots, and the agent flagged those for me to check by hand.

For some shows Exa returned an episode page from its library with no link to the show’s own site, and the episode title or the contact had to be found another way. A step that looks up the show’s feed from the episode page before scoring would save that work.

Once the pitches are sent and replies come in, I’ll record which shows answered and which booked a guest, so the next run can score shows on how they responded as well as on fit.