AI YouTube Transcript Analyzer

Turn one YouTube video into one sharp answer.

Video URL + Prompt = { "answer": "❤️‍🔥" }

No setup maze. No prompt engineering circus. No transcript cleanup work on your side.

Drop in a YouTube URL, tell the Actor what you want, and it will:

• fetch the transcript
• understand the full video
• chunk long videos safely behind the scenes
• return either plain text or strict JSON

Input

{
  "videoUrl": "https://www.youtube.com/watch?v=dQw4w9WgXcQ",
  "prompt": "Summarize the video in 3 concise bullet points.",
  "jsonschema": {
    "title": "video_summary",
    "type": "object",
    "additionalProperties": false,
    "properties": {
      "topic": { "type": "string" },
      "highlights": {
        "type": "array",
        "items": { "type": "string" }
      }
    },
    "required": ["topic", "highlights"]
  }
}

What You Get

• One YouTube video per run
• One custom prompt per run
• Plain text output, or strict JSON with your schema
• Clean dataset output shaped like { "result": ... }

Pricing, But Human

We price by transcript analysis.

• Most short videos usually fit into 1 transcript analysis
• A short video with jsonschema still usually fits into 1 transcript analysis
• If the video is long, we automatically split the transcript into chunks behind the scenes
• Each processed chunk counts as 1 transcript analysis
• After all chunks are processed, the final combined answer counts as 1 more transcript analysis
• Internal fallback retries are not billed as extra transcript analyses

Big video? No panic.

We use the transcript's flattened text, cut it into safe pieces when needed, analyze every piece, and stitch the final answer back together. The goal is simple: no lost context, no dropped data, no messy partial answer.

Example Dataset Item

{
  "result": {
    "topic": "Product strategy",
    "highlights": [
      "The speaker explains...",
      "The video compares...",
      "The conclusion emphasizes..."
    ]
  }
}

Example OUTPUT Record

{
  "summary": {
    "requested": 1,
    "succeeded": 1,
    "failed": 0,
    "chunkCount": 3,
    "fallbackUsed": false,
    "llmCallCount": 4,
    "llmAttemptCount": 4,
    "aiCallsRequested": 4,
    "aiCallsCharged": 4,
    "chargeLimitReached": false,
    "finalModel": "mistralai/mistral-medium-3-instruct"
  },
  "error": null,
  "meta": {
    "pricing": {
      "eventName": "ai-call",
      "requestedAiCalls": 4,
      "chargedAiCalls": 4,
      "chargeLimitReached": false
    }
  }
}

Environment Variables

• WORKER_BASE_URL required
• WORKER_API_TOKEN optional
• REQUEST_TIMEOUT_MS optional

Apify Pricing Setup

• 1 ai-call event = 1 transcript analysis
• you choose the price of each transcript analysis
• long videos cost more only because they truly require more analysis work
• transcripts are split into chunks of 12,000 characters; each chunk is processed as a separate LLM request and billed individually; after all chunks are processed, a final merge request combines the results and is also billed

Example scenario

• A 20-minute video has a transcript of 15,000 characters

• This gets split into:

  • 2 chunks (12,000 + 3,000)
  • 1 merge step

• Total cost = 3 units (2 chunk analyses + 1 merge)