130,000 Podcasts Are Now Searchable Text, Including the Ads

Particle launched a podcast search engine called Radar on August 26. It has transcribed more than 130,000 shows, including every podcast in the Apple Top 200 across 135 categories, and it adds roughly 20,000 episodes a day. The transcripts carry speaker labels. The system tags the people, companies, brands and products named inside them.

Sara Beykpour, Particle’s co-founder and CEO, gave TechCrunch the plainest description of the gap she’s filling:

“Agents are generally blind to audio; they can’t see it unless something or someone has transcribed it.”

That blindness was doing a lot of work for marketers, and most of them never noticed. Audio was the one large channel where a brand could be discussed at length and leave almost no machine-readable trace.

Two things a marketer just lost

The first is deniability about what you said. Every hour-long interview a founder gave, every offhand competitor comparison, every number quoted from memory on someone’s show: those were effectively unsearchable outside the people who listened. Now they’re indexed, attributed to a speaker, and reachable by a query.

The second is the privacy of podcast ad buys. Radar runs a dedicated search engine for podcast advertising that finds the episodes where a company advertises and tracks the pattern over time. Host-read sponsorships were one of the last ad formats with no public archive. There’s no Ad Library for a mid-roll. Now there’s a searchable one, and it works on your competitors and on you at the same rate.

Who is paying for it first

Not marketers. Beykpour named the customers pushing the most volume through the API:

“Hedge funds have been the highest-volume customers that are directly integrating with the API,”

Finance got there first because finance treats an executive’s offhand remark as a tradeable signal. The same remark is a brand asset or a brand liability, and marketing isn’t watching it.

Particle is not a podcast company by origin. It began as an AI news reader built by former Twitter engineers on $4.4M in seed funding, and Beykpour was a senior director of product management at Twitter before that. The podcast work grew out of a feature in that app. Beykpour frames the result as an infrastructure layer rather than an app:

“Our vision is really to have all new media intelligence and all audio intelligence in that API. One of the reasons why it’s an interesting space is that most API agents and services crawl the web and they’re focused on text. We are providing that layer with audio,”

Radar exposes that layer through an API and an MCP server, a plug that lets an AI assistant call an outside service directly rather than guessing from what it was trained on. Exa, a search service built for AI agents, is a launch partner. Individual access runs $29 a month per seat, a business plan is $399 a month for 20 seats, and API pricing is negotiated.

The mechanism, and why it changes discoverability

Assistants answer from what they can read. We’ve covered how narrow that set is: the median company gets named in 16% of relevant AI answers and linked in 6%, and the source mix moves fast enough that Reddit’s share of ChatGPT citations fell without any two trackers agreeing on the size. Transcribing 130,000 shows adds a large, freshly structured body of text to the pool those answers draw from, with named speakers and named brands already tagged.

Which upgrades podcast guesting from a reach play to a discoverability one. The show’s download count stops being the only measure of the placement.

Anuj Agarwal, founder of the podcast database MillionPodcasts, made the case for guesting on its own terms in PR News two days before Radar launched:

“Podcast listeners don’t trust guests. They trust the hosts who invite the guests.”

That borrowed trust is the reason the format works on people. Machines don’t extend it. An agent reading a transcript sees a claim attached to your name in a source it can cite, and the host’s endorsement doesn’t travel with the sentence.

The counter-case

Reach numbers argue against getting excited. A show with 400 listeners doesn’t become important because a machine can now read it, and the volume of transcribed audio entering the pool is enormous, so any single appearance is a smaller fraction of it than the same words on a well-linked page would be. Nobody has published a measured figure for how often assistants cite podcast transcripts, and this launched today. Treating it as a proven channel would be getting ahead of the evidence.

The part that isn’t speculative is the record itself. Whether an assistant cites your appearance or not, the appearance is now text, searchable by name, with your ad buys alongside it.

What to do

Search yourself. Pull the list of every episode where your company, your founders and your product names get spoken, then read what was said rather than what your PR recap claimed. Most teams have never seen that list because it didn’t exist in a form anyone could request. It exists now, your competitors can pull the same list about you, and the first useful version of this work is finding out what is already on the record.