YouTube Corpus Audit & RAG Readiness Report

Turn public YouTube captions from videos, playlists, or channels into a decision-ready RAG corpus audit. The report focuses on caption coverage, missing-caption risk, chunking quality, retrieval QA actions, and the next run needed to move from raw transcript extraction to usable AI retrieval.

Use corpus_snapshot for a compact coverage checklist, or rag_readiness when you need retrieval QA actions and prioritized fixes. The legacy chunks mode is still available for timestamped transcript rows.

Store Quickstart

Recommended first run:

{
  "videoUrls": [
    "https://www.youtube.com/watch?v=dQw4w9WgXcQ"
  ],
  "language": "en",
  "reportTier": "corpus_snapshot",
  "maxChargeUsd": 9,
  "maxVideos": 1,
  "delivery": "dataset",
  "dryRun": false
}

Input Examples

Corpus Snapshot Report

{
  "videoUrls": [
    "https://www.youtube.com/watch?v=dQw4w9WgXcQ"
  ],
  "reportTier": "corpus_snapshot",
  "maxChargeUsd": 9,
  "maxVideos": 1,
  "delivery": "dataset",
  "dryRun": false
}

RAG Readiness Report

{
  "channelUrls": [
    "https://www.youtube.com/@OpenAI"
  ],
  "reportTier": "rag_readiness",
  "maxChargeUsd": 29,
  "maxVideos": 5,
  "chunkSize": 1200,
  "chunkOverlap": 150,
  "delivery": "dataset",
  "dryRun": false
}

Legacy Transcript Chunks

{
  "videoIds": [
    "dQw4w9WgXcQ"
  ],
  "reportTier": "chunks",
  "language": "en",
  "delivery": "webhook",
  "webhookUrl": "https://example.com/webhook",
  "dryRun": false
}

Sample Output

{
  "meta": {
    "actorName": "youtube-channel-transcript-rag-intelligence",
    "actorTitle": "YouTube Channel Transcript RAG Intelligence",
    "fetchedAt": "2026-05-09T00:00:00.000Z",
    "totalRows": 2
  },
  "rows": [
    {
      "rowType": "corpus_audit_report",
      "reportTier": "rag_readiness",
      "status": "success",
      "chargedEvent": "rag_readiness_report",
      "chargedUsd": 29,
      "decisionSummary": "RAG readiness report: 4/5 videos have usable public captions, coverage score 80, RAG readiness 76/100, missing-caption risk medium.",
      "coverageScore": 80,
      "ragReadinessScore": 76,
      "missingCaptionRisk": "medium",
      "chunkingRisks": [
        {
          "severity": "low",
          "code": "chunking_ok",
          "action": "Use current chunking as a baseline for retrieval QA."
        }
      ],
      "retrievalQaChecklist": [
        {
          "check": "caption_coverage",
          "status": "pass",
          "action": "Confirm target videos have public captions before indexing."
        },
        {
          "check": "grounding",
          "status": "required",
          "action": "Run golden Q&A prompts and require timestamped source citations."
        }
      ],
      "actionList": [
        "Replace captionless or unavailable videos and keep their warning rows as no-charge source evidence.",
        "Build 5-10 golden retrieval questions and verify each answer cites a timestamped source chunk.",
        "Tag missing-caption videos as ingestion blockers before scheduling recurring corpus updates."
      ],
      "previewReport": {
        "nextRunInput": {
          "channelUrls": ["https://www.youtube.com/@OpenAI"],
          "reportTier": "rag_readiness",
          "maxChargeUsd": 29,
          "dryRun": false
        }
      },
      "sourceUrls": ["https://www.youtube.com/@OpenAI"]
    }
  ],
  "warnings": []
}

Output Fields

Report rows include:

• decisionSummary
• coverageScore
• ragReadinessScore
• missingCaptionRisk
• chunkingRisks
• retrievalQaChecklist
• actionList and prioritizedActions
• previewReport.nextRunInput
• status, chargedEvent, chargedUsd, reason
• sourceUrls, warnings, errors

Supporting transcript chunks are included as no-charge evidence rows in report mode. Legacy chunk mode keeps timestamped rag_chunk rows with video metadata, chunk text, timestamps, and source URLs.

Pricing and No-Charge Rules

• corpus_snapshot emits corpus_snapshot_report.
• rag_readiness emits rag_readiness_report.
• Report mode charges at most one paid event per run. Supporting chunks are no-charge evidence rows.
• dryRun, demoMode, caption failures, source failures, and maxChargeUsd limit rows are no-charge.
• The recurring watch summary is planned and proof-gated. It is not selectable in the public input schema and is not promoted until paid proof exists.

Compliance Guardrails

• Uses public YouTube pages and public caption tracks only.
• No account session, private video, member-only, paywalled, or login-only access is used.
• No CAPTCHA or rate-limit bypass is attempted.
• Do not position output as a replacement for rights-managed transcript licensing.
• Do not claim ranking, sales, or revenue improvements from the report.
• Do not use provider emblems or wording that implies upstream approval.

See Also

• youtube-transcript-bulk-api
• app-release-review-regression-monitor
• website-content-extractor