We're happy to announce a new AI-powered search for podcast transcripts.

Until now, search on Podgist has been keyword-based: you typed a phrase, and we found the transcripts that contain those exact words. That works well when you know the specific terms you're looking for, but it's less helpful when you have an actual question in mind.

The new AI search lets you ask in plain language. For example, what do guests say about the risks of AI regulation?, and returns a short, synthesized answer drawn from across our transcript library, with citations linking back to the specific episodes and passages that informed it. You can follow those links straight to the source and read or listen to the full context.

Under the hood, your question is embedded and matched semantically against our transcript index, so it surfaces passages that are close in meaning rather than in wording. The most relevant passages are then handed to a language model that reads them and writes the answer, citing its sources as it goes. Because each query runs an embedding and a model call, we offer a small free daily allotment for every visitor, with paid credit tiers for heavier use to help cover the cost.

You can try it now from the AI Search page. As with everything here, it's a work in progress, and we'd love to hear what works and what doesn't.