How to Get a Podcast Transcript via API (No Manual Transcription Needed)

July 14, 2026

If you’re building an AI agent, a search tool, or a RAG pipeline over podcast content, you need transcripts — and you probably assumed you’d have to run every episode through a speech-to-text model. Often, you don’t. A large and growing share of podcasts already publish transcripts themselves, right inside their RSS feed. The hard part isn’t transcription; it’s extracting and normalizing what’s already there.

Podcasts already ship transcripts

The Podcasting 2.0 namespace added a <podcast:transcript> tag that publishers place in each episode’s RSS <item>. It points to a transcript file with a URL and a MIME type:

<podcast:transcript
  url="https://example.com/ep342/transcript.vtt"
  type="text/vtt" />

That sounds simple until you look across real feeds. Publishers use JSON (the Podcasting 2.0 schema), WebVTT, SubRip (SRT), or plain text/HTML — and some list several formats or languages for the same episode.

Why parsing it yourself is tedious

To do this by hand for arbitrary shows, you end up writing and maintaining:

  • A parser for each format — the JSON segment schema, VTT cue timings with <v Speaker> voice tags, SRT’s indexed blocks, and a plain-text fallback.
  • Timestamp handling for the different notations you’ll hit (HH:MM:SS.mmm vs MM:SS,mmm).
  • Selection logic when a feed lists multiple transcripts, plus fetching, retries, and graceful handling of dead or malformed source URLs.

That’s a lot of plumbing before you extract a single usable word.

PodKit normalizes all of it into one JSON response

PodKit reads the feed, picks the best available transcript (JSON › VTT › SRT › text), fetches it, and parses it into one consistent shape every time: a format, a flat text string, and segments with startTime, endTime, speaker, and text.

curl "https://podkitapp.com/v1/episode/42/transcript" \
  -H "x-api-key: pk_your_key"
{
  "episodeId": 42,
  "format": "vtt",
  "text": "Welcome back to the show. Today we're talking about AI agents...",
  "segments": [
    { "startTime": 0.0, "endTime": 4.2, "speaker": "Host",  "text": "Welcome back to the show." },
    { "startTime": 4.2, "endTime": 9.8, "speaker": "Guest", "text": "Thanks for having me." }
  ]
}

The {id} is a PodKit episode id — you get it from GET /v1/podcast/{itunesId}, which lists a show’s episodes. If a source file can’t be fetched or parsed, you still get a clean 200 with text and segments set to null, plus sourceUrl and parseError: true, so you can fall back to the raw file. Episodes with no published transcript return 404 — that’s the case where you’d reach for your own speech-to-text.

Use text for embeddings, segments for citations

For RAG, embed the flat text (or chunk the segments). For agents that cite sources, the segments give you timestamped, speaker-attributed spans you can link back to the exact moment in the audio. Because PodKit caches parsed transcripts, repeat reads are fast and don’t re-hit the origin feed.

See the Transcript API →   Read the full docs

← All posts