If transcript search cannot find your podcast episode, generating text from accessible audio may still work. You do not have to wait for an existing transcript to appear in a search index.
We saw this with three Spotify episodes: Podtyper had generated text for them, while our Exa searches returned no exact transcript match. Those Podtyper results covered the first 30 minutes of each recording, so they demonstrate that generation worked, rather than establish full-episode delivery.
The useful distinction is between finding text that already exists and creating text from a recording. Here is how to choose a route, what our checks found, and what to try when neither works.

When to search and when to generate
Start with transcript search when you want to discover conversations across many shows, find a guest's appearances, or locate a discussion of a topic. An existing transcript can save you from processing the audio again. Exa's podcast search includes episode transcripts contributed by its data partner Particle.
On-demand transcription is useful when you already know the episode and need a document made from its audio. Podtyper tries to find supported public episode audio, transcribes it, and produces text you can review and export. It does not require a matching transcript in a search catalog.
The requirement is audio access. A public Spotify listing does not guarantee that another service can retrieve the recording. Processing takes time, and your available minutes determine how much of the episode can be transcribed.
What our episode checks found
We make Podtyper. On 18 November 2026, we searched Exa for 17 distinct Spotify episodes: five recent episodes with saved Podtyper text, plus 12 episodes whose audio we had failed to retrieve. We searched by title and show name and fetched transcript bodies for exact matches. Related conversations and descriptions did not count as transcripts.
These are selected examples, not a random coverage benchmark. The five rows below were chosen because Podtyper had already generated text; their results cannot be used to calculate its overall success rate. Four were partial outputs under the available allowance. The evidence file includes all 17 episodes, search-result metadata, query methods and rate-limit handling.

| Episode | Podtyper's saved output | Exa result in our check |
|---|---|---|
| Unison Church Podcast — Discussing Messianic Judaism, episode 17 | First 30 minutes; partial | No exact transcript found |
| Money Tree Investing — Stagflation is Back... Oh Shit | First 30 minutes; partial | Show found; episode not found |
| Sister Act 40k — Big Dogma in Little London | Full episode, about 110 minutes | Transcript retrieved, extending to about 110 minutes |
| A Question of Science — What's going on inside the teenage brain? | First 30 minutes; partial | No exact transcript found |
| eComFuel Podcast — How an AI Army Replaced His Ad Agency | First 30 minutes; partial | Transcript retrieved, extending to about 47 minutes |
Exa supplied substantial text for two recordings, including more of the eComFuel episode than the selected partial Podtyper output. For three others, generating from audio produced text that our transcript searches did not find.
A search miss is not proof that an episode is absent from Exa's or Particle's underlying catalog. We also did not compare word accuracy, speaker attribution, processing speed or price.
A newly published episode illustrates the difference
Money Tree Investing's publisher feed dated Stagflation is Back... Oh Shit to 18 November at 06:00 UTC. Podtyper had generated a partial transcript that day. At our check, the Exa show page listed releases only through 13 November.
That is consistent with a discovery, processing or search-visibility delay. It does not measure Exa's usual update time. For someone who needs text from a newly released episode, however, generating from accessible audio provides another route to try.
What if the audio cannot be retrieved?
We also checked 12 distinct episodes that Podtyper could not retrieve. Our Exa searches returned no verified transcript matches for those recordings either. Several results covered similar subjects or featured the same speaker, but they were different episodes.
Podtyper needs supported public audio. Its error messages reported that a public feed could not be found; that alone does not prove the shows were Spotify exclusives. The failed sample included smaller and non-English shows, and does not represent Spotify as a whole.
Particle also offers on-demand transcription for requestable back-catalog episodes. We did not test that service directly. These checks compare our observed Podtyper outputs with Exa search and retrieval, rather than establish everything Particle can deliver.
How to get text when your transcript search comes up empty
- Check the publisher and your podcast app. A reviewed publisher transcript or an available native transcript may already meet your needs. Confirm the show, episode title and date before using text from a search result.
- Generate from supported audio. If you know the episode and need a reading document, try on-demand transcription. Our Spotify transcript guide explains the steps and access requirements.
- Try another source for the same recording. If retrieval fails, check whether the publisher provides a public Apple Podcasts episode or a direct audio link. Verify that it is the same recording. Paying for transcription does not make inaccessible audio retrievable.
- Check coverage and review the text. Make sure the output reaches the passage you need, then listen back to names, numbers and quotations. A partial transcript can be useful, but it should not be mistaken for the whole episode.
For accuracy measurements, our separate podcast transcription research describes its recordings and scoring. Finding a transcript and checking whether it is correct are separate tasks.
If you already have an episode link and want a document, generate a transcript from a supported Spotify episode. Your first 30 minutes are free. A longer recording may return a partial result; completing it requires sufficient available minutes and processing the remainder.