OpenAI Files vector storage scales to thousands of episodes without hitting context limits. Smaller shows under about a hundred episodes can run inline context without files. The plugin can be configured to switch between modes automatically based on archive size, and you can run inline context for the most recent fifty episodes plus vector storage for the back catalog if you want both speed and depth.

Yes, transcripts unlock the magic. Most shows already use Whisper or a service like Descript, and SleekAI takes those text files directly. If you have not transcribed the back catalog, batch-transcribing with Whisper costs cents per episode and is a one-time investment that pays for itself the moment the chatbot starts surfacing old episodes in answers.

Yes. The system prompt can include a URL pattern so answers always link to /episode-{number} or your equivalent. If your transcripts contain timestamps, the bot can also produce time-anchored deep links, for example /episode-127#t=2400 for the 40-minute mark, which lands listeners exactly on the moment they were asking about.

Anything stored as a custom post type in WordPress is indexable. External-only feeds from Buzzsprout, Transistor, or Megaphone need their show notes pulled into WP first, usually via a small ingest script that creates a post per episode. Once the posts exist, the chatbot reads them like any other content type.

Multibot supports a chatbot per show or per category, with different prompts, contexts, and accent colors. A network with a tech show, a culture show, and a finance show can run three distinct chatbots on the same site, each scoped to its own archive. Subdomains and category-based display conditions both work for routing.

In your WordPress database. You control retention and access, and you can export logs to CSV for editorial review. For shows with strict listener-privacy promises, the fact that conversations never leave your infrastructure is often the deciding factor over hosted Q&A widgets that store data on a third-party server.

Most shows mark ad reads with timestamps or section headers in the transcript, and the system prompt can instruct the bot to skip those sections when answering. If you want the bot to acknowledge sponsorships when relevant, that is also configurable. Either way, the bot will not invent ad copy or mistake an ad read for a host's actual opinion.

If your episode pages include guest bios, social links, or company info as custom fields, the bot can quote them. A listener asking 'who was the guest who talked about pricing pages' gets the guest name, the episode, and a link to the guest's site if you publish one. Multibot can also run a guest-specific chatbot scoped to a single guest's appearances across the show.