/youtube-comment-analyzer

YouTube Comment Analyzer

Analyze YouTube video comments for brand relevance using a 5-layer AI scoring model. Identifies purchase intent, product questions, brand risks, competitor mentions, and sentiment — all scoped to YOUR brand context.

When to Use

• When the user wants to analyze YouTube comments for brand relevance
• When the user wants to understand comment sentiment and business categories on a video
• When the user asks about KOL/influencer video comment quality
• When the user wants to find high-priority comments (purchase intent, brand risks)
• Trigger phrases: "analyze comments", "YouTube comment analysis", "brand relevance", "comment sentiment"

Configuration

Required environment variables:

• `ANTHROPIC_API_KEY` — Your Anthropic API key for Claude AI analysis
• `YOUTUBE_API_KEY` — Your YouTube Data API v3 key for fetching comments

Input Parameters

The user should provide: 1. **YouTube Video URL** (required) — e.g., `https://www.youtube.com/watch?v=VIDEO_ID` 2. **Brand Name** (required) — The brand/product to analyze relevance for 3. **Collaboration Type** (optional, default: "auto") — One of:

• `dedicated` — Entire video is about this brand/product
• `sponsored` — Brand segment within the video
• `mention` — Brief mention of the brand
• `placement` — Product visible but not discussed
• `auto` — Let AI determine relevance thresholds

4. **Brand Context** (optional) — Additional context: website URL, product names, keywords 5. **Output Format** (optional, default: "markdown") — `markdown`, `csv`, or `json` 6. **Language** (optional, default: "auto") — Output language (`en`, `zh`, `ja`, `ko`, etc.)

Instructions

Follow these steps to analyze YouTube video comments:

Step 1: Extract Video ID and Fetch Comments

Extract the video ID from the URL, then use the YouTube Data API to fetch comments.

# Extract video ID from URL (handles youtube.com/watch?v=, youtu.be/, etc.)
VIDEO_ID="<extracted from user URL>"

# Fetch comments via YouTube Data API v3
curl -s "https://www.googleapis.com/youtube/v3/commentThreads?part=snippet&videoId=${VIDEO_ID}&maxResults=100&order=relevance&key=${YOUTUBE_API_KEY}"

Parse the response to extract:

• Comment ID
• Comment text
• Author name
• Published date
• Like count
• Reply count

Also fetch the video title:

curl -s "https://www.googleapis.com/youtube/v3/videos?part=snippet&id=${VIDEO_ID}&key=${YOUTUBE_API_KEY}"

Step 2: Analyze Comments with Claude AI

Send comments in batches of 15 to Claude for analysis. Use the following system prompt and scoring model.

**Model:** `claude-haiku-4-5-20251001` (cost-efficient for batch analysis)

**System Prompt:**

You are an expert in analyzing YouTube comments for brand relevance in KOL/influencer sponsored videos.

## Your Task
Given a brand context and YouTube video context, analyze each comment using a multi-layer scoring model.

## Step 1: Semantic Relevance (semantic_score: 0.0-1.0)
Does the comment relate to the product, brand, or sponsored content?
- 0.9-1.0: Directly discusses the product/brand by name
- 0.7-0.89: Discusses product features, use cases, or category closely tied to the brand
- 0.5-0.69: Tangentially related (general product category discussion)
- 0.0-0.49: Unrelated (off-topic, "first!", personal KOL comments, video production)

## Step 2: Intent Classification (business_category)
Classify into ONE:
- purchase_intent: wants to buy, asks price/discount/where to get it
- product_question: asks about features, compatibility, bugs, pricing, privacy
- brand_risk: false claims, scam links, privacy concerns, KOL misrepresentation
- competitor_mention: compares with or mentions competitor products
- positive_feedback: praises product, recommends it
- negative_feedback: complains, disappointed, wants refund
- irrelevant: unrelated to the brand/product

sub_category (for product_question and brand_risk only):
- product_question → feature | bug | compatibility | pricing | privacy
- brand_risk → false_claim | scam | privacy_concern | kol_misrepresentation

## Step 3: Entity Detection (entity_match)
Check comment for:
- "exact_brand": brand name explicitly mentioned
- "partial": indirect reference ("this tool", "the app", product feature name)
- "competitor": competitor brand mentioned
- "none": no brand/product entity detected

## Step 4: Video Relevance (video_relevance_score: 0.0-1.0)
How closely does the VIDEO TITLE relate to the brand/product?
- 0.9-1.0: Video is entirely about this brand/product
- 0.6-0.89: Video covers a category the brand is in
- 0.3-0.59: Loosely related topic
- 0.0-0.29: Video has nothing to do with the brand
This score should be THE SAME for all comments in the same batch (same video).

## Step 5: Output per comment
Return a JSON array. Each object MUST include:
- id: the comment ID
- business_category, sub_category
- semantic_score: 0.0-1.0
- video_relevance_score: 0.0-1.0
- entity_match: "exact_brand" | "partial" | "competitor" | "none"
- priority: "high" | "medium" | "low"
  - high: brand_risk; strong negative emotion; scam/phishing
  - medium: purchase_intent; product_question; competitor_mention; mild negative
  - low: positive_feedback; neutral; irrelevant
- suggested_action: "reply" | "escalate" | "monitor" | "ignore"
- key_insight: one-line summary
- evidence_quoted_text: key phrase (max 50 chars)
- classification_reason: one sentence why
- sentiment: positive | neutral | negative | toxic
- intent: question | complaint | praise | spam | scam | phishing | brand_mention | competitor_mention | general
- threat_level: safe | review | auto_hide | auto_delete
- confidence: 0.0-1.0
- competitor_mentioned: competitor name or null

IMPORTANT: Analyze comments in their ORIGINAL language. Support all languages including English, Chinese, Japanese, Korean, Hindi, Arabic, Spanish, Portuguese, etc. Output fields (key_insight, classification_reason) should match the user's preferred language.

Respond ONLY wit