ACTION_ID: youtube_extract_video_transcript
NAME: Youtube: Extract YouTube Video Transcript
CATEGORY: Web Research
CREDITS: 1

Extract a YouTube video's transcript as raw text or timestamped
segments, with control over language and whether to prefer
auto-generated or author-uploaded captions.

1. INPUTS

youtube_video_or_share_url (type: url, required)
  Youtube Video or Share URL. Complete YouTube video URL or share URL
  for extracting transcript

transcript_structure (type: string, optional, resolve: enum)
  Transcript Structure. Concatenated body of text or segmented with
  timestamps. Default is concatenated and recommended for AI analysis.
  Values:
    - { value: "concatenate", label: "Raw Text (recommended)" }
    - { value: "segment", label: "Text with Timestamps" }

transcript_language (type: string, optional)
  Transcript Language. Language for extracted transcript. Default is
  English.

transcript_type (type: string, optional, resolve: enum)
  Transcript Type. Defines the transcript preference when both
  auto-generated and manually uploaded transcripts are available.
  Default is auto-generated.
  Values:
    - { value: "auto", label: "Auto-Generated by Youtube (recommended)" }
    - { value: "manual", label: "Manually Uploaded by Author" }

2. OUTPUTS

raw_transcript (string) — Transcript with raw text

transcript_with_timestamps (raw_array) — Transcript with text segmented
via timestamps

search_parameters (json) — Search parameters

languages (raw_array) — Available transcript languages

3. HOW TO CONFIGURE

Configure Action body:

{
  "inputs": {
    "youtube_video_or_share_url": "{{input.video_url}}",
    "transcript_structure": "concatenate",
    "transcript_type": "auto"
  }
}

Field-by-field:
  - transcript_structure  `concatenate` returns one text blob (best for
                          LLM analysis, and the default);
                          `segment` returns timestamped segments.
  - transcript_type       `auto` prefers YouTube's auto-captions;
                          `manual` prefers author-uploaded ones.
  - transcript_language   Defaults to English. No dynamic-options
                          resolver — check `available_languages` on the
                          output after a first run.

4. KEY NOTES

- Both output fields are always present but only one is populated —
  `transcript_concatenated` for `concatenate`, `transcript_segmented`
  for `segment`.
- `available_languages` lists what the video actually offers. Run once,
  read it, then set `transcript_language` accordingly.
- `transcript_type: manual` fails on videos with no author-uploaded
  captions, which is most of them. Leave it on `auto` unless caption
  accuracy matters.
- `transcript_segmented` and `search_parameters` are nested types — use
  bracket access inside a `formula`.

5. WHERE IT FITS IN A WORKFLOW

Pattern: enrich_youtube_video (metadata + filter on views) ->
youtube_extract_video_transcript (segmented, for quoting with
timestamps) -> llm_models.

6. WHEN TO USE

Use youtube_extract_video_transcript when the transcript is the point
and you need timestamps or a non-English language.

7. WHEN NOT TO USE

Also need views, likes and channel stats
  -> action-detail/enrich_youtube_video.txt

The recording is a sales call, not a video
  -> action-detail/gong_get_call_transcript.txt

Last updated: 2026-08-03.