Imported from buda-ai/bunny-agent (templates/seo-keywords-analytics/.claude/skills/keyword-mining/SKILL.md). Install upstream with npx skills add buda-ai/bunny-agent --skill keyword-mining. Copyright stays with the author.

name: keyword-mining description: Analyzes Google Search Console data to extract core keywords, identify strategic opportunities (quick wins, CTR gaps, growth keywords), and generate actionable SEO reports with priority scoring and topic clusters. CRITICAL: This skill MUST be invoked for ANY keyword analysis request - do not call MCP tools directly.

Keyword Mining

IMPORTANT: When This Skill Should Be Used

This skill is the ONLY way to perform keyword analysis.

MANDATORY: Invoke this skill when the user asks for:

• "分析关键词" / "关键词情况" / "analyze keywords"
• "挖掘关键词" / "keyword mining"
• "GSC分析" / "GSC analysis" / "search console analysis"
• "搜索表现" / "search performance"
• "识别关键词机会" / "identify keyword opportunities"
• "关键词策略报告" / "keyword strategy report"

DO NOT skip this skill and directly call gsc_search_analytics MCP tool. The skill manages:

• Safe 14-day default data range (avoids rowLimit issues)
• Proper data validation and error handling
• Strategic analysis requirements
• Report language detection and generation

This skill performs comprehensive keyword analysis using Google Search Console data to extract core keywords, identify strategic opportunities, and generate prioritized reports for SEO decision-making.

CRITICAL: You are acting as an expert SEO analyst, NOT a data processing tool. Your goal is to provide strategic insights and actionable recommendations, not just data tables.

When to Use This Skill

Trigger Keywords: When user asks to:

• "analyze keywords" / "关键词分析"
• "keyword mining" / "挖掘关键词"
• "GSC analysis" / "GSC数据分析"
• "search console report" / "搜索分析报告"
• "identify keyword opportunities" / "识别关键词机会"
• "keyword strategy report" / "关键词策略报告"

Use Cases:

• Generating keyword strategy reports
• Identifying website's core keyword themes
• Mining untapped keyword opportunities
• Creating data-driven SEO roadmaps
• Understanding keyword performance and intent
• Prioritizing keywords for content strategy
• Building topic cluster strategies

What This Skill Does

1. Core Keyword Extraction: Identifies primary keywords driving search visibility
2. Opportunity Mining: Discovers high-potential keywords (quick wins, CTR gaps, growth keywords)
3. Intent Mapping: Classifies keywords by search intent (informational, commercial, transactional)
4. Cluster Creation: Groups related keywords into topic themes
5. Priority Scoring: Ranks keywords by potential impact and effort
6. Gap Analysis: Identifies missing keyword coverage
7. Strategic Reporting: Generates actionable keyword reports

Analysis Depth Requirements

CRITICAL MINDSET: Think like an SEO consultant reviewing a client's data. For EVERY finding:

Required Analysis Depth

For EACH significant opportunity you identify, you MUST provide:

1. Why It Matters: Explain the business impact in specific terms

   • ❌ BAD: "High impression keyword"
   • ✅ GOOD: "9,977 impressions at position 5 but only 2 clicks - this is a MASSIVE anomaly indicating the page ranking doesn't match user intent or has a misleading title"

2. Root Cause Analysis: Diagnose WHY the opportunity exists

   • ❌ BAD: "Low CTR"
   • ✅ GOOD: "CTR is 0.02% vs expected 7-9% for position 5. Likely causes: (1) Wrong page ranking for this query, (2) Title doesn't match search intent, (3) URL may not match user expectation"

3. Specific Action Items: Provide concrete, implementable steps

   • ❌ BAD: "Optimize title tag"
   • ✅ GOOD: "Update title from 'Bika - Automation' to 'AI-Powered Telegram Bot Builder | Bika.ai' to emphasize key value prop and include primary keyword naturally"

4. Expected Outcomes: Give specific estimates with reasoning

   • ❌ BAD: "Should improve traffic"
   • ✅ GOOD: "Moving from position 12 to position 6-8 could increase CTR from 0.8% to 4-5%, resulting in ~10-12 additional clicks per week (+500% increase)"

5. Business Context: Connect findings to existing content/pages

   • ❌ BAD: "Create content for this keyword"
   • ✅ GOOD: "You already have /template/ai-batch-image-recognition ranking at position 10! This is a quick win - optimize the existing page by adding '图片识别' to the title and creating a FAQ section"

Analysis Quality Standards

Always explain the "So What?":

• Don't just show data - interpret it
• Connect patterns across keywords
• Identify anomalies and investigate them
• Consider competitive landscape
• Think about user intent and journey

Be specific and actionable:

• Reference actual URLs/pages when possible
• Suggest exact title rewrites, not just "improve title"
• Explain why a keyword is important to THIS specific business
• Prioritize based on effort vs impact trade-offs

Think strategically:

• Group related opportunities into campaigns
• Consider resource constraints
• Flag high-priority items as "URGENT" with justification
• Suggest testing approaches for uncertain opportunities

How to Use

Full Keyword Analysis

Generate keyword analysis report for [URL]
Mine keywords from my GSC data
Create keyword strategy report

Targeted Analysis

Find core keywords driving traffic
Identify quick win opportunities
Show me keyword gaps in my content

Strategic Planning

Build keyword clusters for content strategy
Prioritize keywords for next quarter
Map keywords to content funnel stages

Data Requirements

Primary Method (Recommended): Real-time GSC API via MCP

• Fetch data directly from Google Search Console using MCP tools
• SAFE DEFAULT: Last 14 days (prevents hitting rowLimit on high-traffic sites)
• IMPORTANT: MCP has a rowLimit: 25000 hard constraint
  • Start with 14 days to assess data volume
  • If rows < 10K: can extend to 30 days
  • If rows > 20K: keep to 14 days or use 7-day chunks
  • NEVER fetch 90+ days in single request
• Automatically includes Query, Clicks, Impressions, CTR, Position, Page

Fallback Method: Local CSV Export

• Use only if MCP is unavailable or user provides specific CSV
• CSV location: ./data/gsc-export-YYYY-MM-DD.csv
• Required fields: Query, Clicks, Impressions, CTR, Position, Page (URL)

Report Language

Auto-detect from user request: The report language must match the language of the user's request

• If user asks in Chinese → Generate report in Chinese
• If user asks in English → Generate report in English
• If user asks in other languages → Generate report in that language
• Apply to ALL sections: executive summary, analysis, recommendations, and action items

Mining Process

1. Data Loading and Validation

CRITICAL: Always prefer MCP over local CSV files!

Step A: Try MCP First (Primary Method)

When user requests analysis without specifying a CSV file:

1. Ask for the website URL if not provided
2. CRITICAL - Start with 14-day range to avoid rowLimit issues
   • Default: Last 14 days (safer than 30 days for high-traffic sites)
   • If 14 days has < 10,000 rows, can extend to 30 days
   • If 14 days has > 20,000 rows, stick to 7-14 day range
3. Handle rowLimit constraint: MCP max rows = 25000
   • NEVER fetch 90+ days in one call - this will hit the limit
   • Start conservative (14 days), check data volume, then adjust
   • If you need longer period, fetch in weekly chunks and analyze separately
4. NEW - Use MCP-side filtering to reduce tokens:
   • Set rowLimit: 10000 (balanced - covers most valuable keywords)
   • Use dimensions: "query,page" (minimum required - more dimensions = more data)
   • Consider type: "web" only (skip image/video/news unless specifically requested)
   • After fetch, if rows == 10000 (hit limit), can increase to 15000 or reduce date range
5. Dimensions: query, page (minimum required)

Step B: DATA PRE-FILTERING (NEW - CRITICAL for performance)

After fetching data, ALWAYS pre-filter to reduce processing load:

# PROGRESS: Step 1/4 - Pre-filtering data...
echo "⏳ [1/4] Pre-filtering data to reduce processing load..."

# Extract rows and save filtered version
cat mcp-response.json | jq '.[0].text' | jq '.rows' > raw-rows.json

# Count total rows
TOTAL_ROWS=$(cat raw-rows.json | jq '. | length')
echo "   📊 Total rows fetched: $TOTAL_ROWS"

# Apply value-based filters to reduce dataset:
# - Keep only keywords with >= 10 impressions (removes very long tail noise)
# - Keep only keywords with position <= 100 (include more ranking keywords)
cat raw-rows.json | jq '[.[] | select(.impressions >= 10 and .position <= 100)]' > filtered-rows.json

FILTERED_ROWS=$(cat filtered-rows.json | jq '. | length')
REMOVED_ROWS=$((TOTAL_ROWS - FILTERED_ROWS))
echo "   ✅ Filtered to $FILTERED_ROWS rows (removed $REMOVED_ROWS low-value rows)"
echo "   💾 Using filtered data for all subsequent analysis"

Pre-filtering Benefits:

• Reduces jq processing time by 40-60%
• Focuses analysis on high-value keywords
• Prevents timeout on large datasets
• Still captures all meaningful opportunities

Example MCP call pattern - SAFE APPROACH (Token-Optimized):

# Step 1: Start with 14 days + ADAPTIVE rowLimit
Use gsc_search_analytics to fetch data for:
- Site: https://example.com
- Start date: [14 days ago]
- End date: [today]
- Dimensions: query, page
- Row limit: 10000 (balanced - good coverage without excessive tokens)
- Type: web (skip image/video unless requested)

# Step 2: Check result and ADAPT if needed
After receiving data, check actual rows:
├─ If rows == 10000 (hit limit)
│  └─ Option A: Increase rowLimit to 15000 and refetch
│  └─ Option B: Reduce to 7-day range for better coverage
│  └─ Option C: Proceed with 10000 (covers most valuable keywords)
│
├─ If rows 5000-9999 (excellent coverage)
│  └─ ✅ Proceed: This is optimal for comprehensive analysis
│
└─ If rows < 5000 (light coverage)
   └─ Can extend to 21-30 days if more historical data needed

# Step 3: Client-side pre-filtering
After fetch, filter for: impressions >= 10 AND position <= 100
This removes very low-value noise while preserving most opportunities
Note: If filtered results are still < 100 keywords, further relax to impressions >= 5

# Decision Matrix:
┌─────────────┬──────────┬─────────────────────────────────┐
| Actual Rows | Coverage | Action                          │
├─────────────┼──────────┼─────────────────────────────────┤
| 10000 (hit) | ~70-85%  │ Can extend to 15000 for more   │
| 5000-9999   | ~85-95%  │ ✅ Excellent - proceed         │
| <5000       | ~95-100% │ ✅ Complete - can extend days  │
└─────────────┴──────────┴─────────────────────────────────┘

Key Insight: 10000 rows captures 85-95% of valuable keywords while keeping token usage reasonable. Using impressions >= 10 and position <= 100 captures a comprehensive set including emerging opportunities.

Step B: Fallback to CSV (Only if MCP fails or user specifies CSV)

Only check for local CSV if:

• MCP tools are not available
• MCP fetch fails with error
• User explicitly says "use the CSV file" or "analyze the data in ./data/"

# Check for GSC data file (ONLY as fallback)
ls -la ./data/gsc-export-*.csv

# Validate data completeness
head -50 ./data/gsc-export-*.csv

Verify:

• Required fields present (Query, Clicks, Impressions, CTR, Position)
• Sufficient data volume (> 100 queries)
• Date range specified

2. PARALLEL ANALYSIS (NEW - Critical for performance)

CRITICAL: Execute analysis tasks in parallel using background processes to reduce total execution time from 30+ minutes to 5-10 minutes.

# PROGRESS: Step 2/4 - Running parallel analysis...
echo "⏳ [2/4] Running parallel analysis (4 tasks)..."
echo "   🔄 This will run multiple analyses simultaneously to save time"
echo ""

# Create output directory for intermediate results
mkdir -p ./analysis-cache

# Run ALL analysis tasks in parallel using background jobs (&)
# Each task saves results to a separate cache file

# Task A: Core Keywords Analysis
echo "   ▶ Task 1/4: Extracting core keywords..."
(cat filtered-rows.json | jq -r '.[] | "\(.keys[0])|\(.clicks)|\(.impressions)|\(.ctr*100)|\(.position)"' | \
  sort -t'|' -k2 -nr | head -30 > ./analysis-cache/core-keywords.txt) &
PID_CORE=$!

# Task B: Quick Wins (Page 2 Rankings)
echo "   ▶ Task 2/4: Identifying quick wins..."
(cat filtered-rows.json | jq -r '.[] | select(.position > 10 and .position < 20 and .impressions >= 100) | "\(.keys[0])|\(.position)|\(.impressions)|\(.clicks)|\(.keys[1])"' | \
  sort -t'|' -k3 -nr | head -20 > ./analysis-cache/quick-wins.txt) &
PID_QUICK=$!

# Task C: CTR Opportunities
echo "   ▶ Task 3/4: Finding CTR optimization opportunities..."
(cat filtered-rows.json | jq -r '.[] | select(.position <= 10 and .impressions >= 500) | "\(.keys[0])|\(.position)|\(.impressions)|\(.ctr*100)|\(.keys[1])"' | \
  awk -F'|' 'BEGIN{bench[1]=28;bench[2]=18;bench[3]=11;bench[4]=8;bench[5]=7;bench[6]=5;bench[7]=4;bench[8]=3;bench[9]=2;bench[10]=1.5}
    {pos=int($2);exp=bench[pos];if(exp && $4<exp*0.7) print $0}' > ./analysis-cache/ctr-ops.txt) &
PID_CTR=$!

# Task D: High Volume Keywords
echo "   ▶ Task 4/4: Identifying high-volume opportunities..."
(cat filtered-rows.json | jq -r '.[] | select(.impressions >= 1000) | "\(.keys[0])|\(.impressions)|\(.position)|\(.clicks)|\(.keys[1])"' | \
  sort -t'|' -k2 -nr | head -30 > ./analysis-cache/high-volume.txt) &
PID_VOL=$!

echo "   ⏸️  All tasks started. Waiting for completion..."

# Wait for all tasks and show completion
wait $PID_CORE && echo "   ✅ Task 1/4 complete: Core keywords"
wait $PID_QUICK && echo "   ✅ Task 2/4 complete: Quick wins"
wait $PID_CTR && echo "   ✅ Task 3/4 complete: CTR opportunities"
wait $PID_VOL && echo "   ✅ Task 4/4 complete: High volume analysis"

echo ""
echo "   🎉 All parallel analysis tasks completed!"
echo "   📂 Results cached in ./analysis-cache/"

Parallel Processing Benefits:

• 4x faster than sequential processing
• Reduces total analysis time from 20-30 minutes to 5-10 minutes
• Each analysis runs independently in the background
• Results cached for reuse during report generation

Progress Tracking Pattern:

# Always show progress with these emojis:
echo "⏳ [Step X/Total] Description..."    # In progress
echo "   ▶ Subtask N/M: Details..."        # Starting subtask
echo "   ✅ Task complete"                  # Finished
echo "🎉 All tasks completed!"              # All done
echo "📊 Result: X rows, Y keywords"        # Summary
echo "💾 Saved to: ./path/to/file"         # Location

3. Extract Core Keywords (Using Cached Results)

## Core Keyword Identification

### Primary Traffic Drivers

Keywords generating majority of your organic clicks:

| Keyword | Clicks | Impressions | CTR | Position | Value |
|---------|--------|-------------|-----|----------|-------|
| [keyword] | [X] | [X] | [X%] | [X] | ⭐⭐⭐⭐⭐ |

**Analysis**:
- Total core keywords: [X]
- Combined clicks: [X] ([Y]% of total)
- Average position: [X]
- Average CTR: [X%]

**Theme Identification**:
- Theme 1: [Topic] ([X] keywords, [Y] clicks)
- Theme 2: [Topic] ([X] keywords, [Y] clicks)

3. Classify Keywords by Intent (Using Filtered Data)

## Keyword Intent Analysis

| Intent | Keywords | Clicks | Avg CTR | Avg Position | Strategy |
|--------|----------|--------|---------|--------------|----------|
| Informational | [X] | [X] | [X%] | [X] | Educational content |
| Commercial | [X] | [X] | [X%] | [X] | Comparison content |
| Transactional | [X] | [X] | [X%] | [X] | Product pages |
| Navigational | [X] | [X] | [X%] | [X] | Brand pages |

### Intent Patterns

**Informational** (what, how, why, guide):
- [query examples]
- Content need: Comprehensive guides, tutorials

**Commercial** (best, review, vs, compare):
- [query examples]
- Content need: Comparisons, reviews, case studies

**Transactional** (buy, price, discount):
- [query examples]
- Content need: Product pages, pricing, CTAs

4. Mine Keyword Opportunities (Using Cached Results)

CRITICAL: For each opportunity type, you MUST provide detailed analysis, not just tables!

## Keyword Opportunity Mining

### Priority 1: Quick Wins (Page 2 Rankings)

**Criteria**: Position 11-20, Impressions > 100/month

These keywords rank positions 11-20 with decent impression volume. Small optimizations (title tags, content updates, internal linking) could push them to page 1.

| Query | Position | Impressions | Current Clicks | Potential Clicks* | Impact |
|-------|----------|-------------|----------------|-------------------|--------|
| [keyword] | [15] | [X] | [X] | [~Y] | +Z clicks |

*Potential clicks calculated based on moving from position 11-20 to position 6-10 (3-5% CTR)

**REQUIRED FOR EACH TOP QUICK WIN** (pick top 3-5):

**[Keyword Name]** (Impressions: X, Position: Y)
- Current: X clicks/week (Y% CTR)
- Potential at position 6-8: ~Z clicks/week (W% CTR)
- **Gain: +N clicks/week (+M%)**

**Root Cause**: [Why is it stuck at position 12-15? Thin content? Weak backlinks? Missing keywords in title?]

**Action Items**:
1. [Specific action with the actual page URL if possible]
2. [Concrete recommendation, not generic advice]
3. [Expected timeline and success metrics]

**Example of GOOD analysis**:
> **"twitter automation"** (247 impressions, position 12.35)
> - Current: 2 clicks/week (0.81% CTR)
> - Potential at position 8: ~12 clicks/week (5% CTR)
> - **Gain: +10 clicks/week (+500%)**
>
> **Root Cause**: The page `/template/x-ai-automated-tweets` has good content but the title is buried. It's ranking #12 because competitors have "Twitter Automation" in their H1 and title tags.
>
> **Action Items**:
> 1. Optimize `/template/x-ai-automated-tweets` page for "twitter automation" keyword
> 2. Update title tag to move "Twitter Automation" to the beginning: "Twitter Automation Tool | AI-Powered Tweet Scheduling | Bika"
> 3. Add internal links from high-authority pages (homepage, blog) using anchor text "twitter automation tool"
> 4. Add FAQ section: "How to automate twitter posts", "Best twitter automation tools 2026"
> 5. Timeline: 2-3 weeks to see position improvement

---

### Priority 2: CTR Optimization

**Criteria**: Position ≤ 10, CTR < 70% of expected benchmark

High-impression keywords with CTR significantly below expected benchmarks for their position.

**CTR Benchmarks by Position**:
- Position 1: 28-35% | Position 2: 15-20% | Position 3: 10-13%
- Position 4-5: 7-9% | Position 6-10: 3-5%

| Query | Position | Impressions | CTR | Expected CTR | CTR Gap | Potential Gain |
|-------|----------|-------------|-----|--------------|---------|----------------|
| [keyword] | [5] | [X] | [Y%] | [Z%] | [-W%] | +N clicks |

**REQUIRED FOR EACH TOP CTR OPPORTUNITY** (pick top 3-5):

**[Keyword Name]** (Impressions: X, Position: Y)
- Current: X clicks (Y% CTR) at position Z
- Expected: W clicks (V% CTR) for position Z
- **Missing: N-M clicks per week**

**Root Cause Analysis**: [Why is CTR so low? Wrong page ranking? Generic title? Misleading meta? Competitor has rich snippets?]

**Urgent Investigation Required** (if CTR gap > 5%): [What specifically needs to be checked in GSC?]

**Specific Fix**:
- Current title: "[exact current title]"
- Suggested title: "[exact new title with reasoning]"
- Meta description update: "[specific recommendation]"
- Other actions: [schema markup, featured snippet optimization, etc.]

**Example of GOOD analysis**:
> **"混元" (Tencent Hunyuan AI)** (903 impressions, Position 5.96)
> - Current: 1 click (0.11% CTR) at position 6
> - Expected: 27-36 clicks (3-4% CTR) for position 6
> - **Missing: 25-35 clicks per week**
>
> **Root Cause**: The ranking page is likely a generic integration reference page, not dedicated content about Hunyuan. Users searching "混元" want to learn about the AI model, not see it as one line in a features list.
>
> **Urgent Investigation Required**: Check in GSC which page is ranking. If it's just a brief mention in a larger page, this explains the terrible CTR.
>
> **Specific Fix**:
> - If current page is integration list: Create dedicated "如何在Bika中使用腾讯混元AI" guide page
> - If dedicated page exists: Update title from generic "Bika AI Integration" to "腾讯混元AI自动化集成指南 | 完整教程 | Bika.ai"
> - Add compelling meta: "通过Bika.ai连接腾讯混元大模型，实现智能对话、文本生成等AI能力。5分钟快速接入，附完整代码示例。"
> - Add structured FAQ schema for "什么是混元AI", "如何使用混元"
> - Timeline: Immediate if just title/meta fix; 1-2 weeks if new content needed

### Type 3: High Volume, Low Position

**Criteria**: Impressions > 1000/month, Position > 20

| Keyword | Impressions | Position | Est. Monthly Searches | Potential at Pos 5 |
|---------|-------------|----------|----------------------|-------------------|
| [keyword] | [5,000] | [35] | [~30,000] | [~1,500 clicks] |

**Why**: Huge traffic potential
**Effort**: High (major content creation)
**Timeline**: 4-12 weeks

### Type 4: Emerging Trends

**Criteria**: Impression growth > 200% in last 90 days

| Keyword | Current Impressions | Growth | Position | Opportunity |
|---------|-------------------|--------|----------|-------------|
| [keyword] | [800] | +450% | [18] | First mover |

**Why**: Growing demand, low competition window
**Effort**: Medium
**Timeline**: 1-2 weeks (time-sensitive)

### Type 5: Branded vs Non-Branded Balance

| Type | Queries | Clicks | % of Total | Health Check |
|------|---------|--------|------------|--------------|
| Branded | [X] | [X] | [X%] | [assessment] |
| Non-Branded | [X] | [X] | [X%] | [assessment] |

**Healthy Mix**: 60-80% non-branded clicks for SEO growth

5. Build Topic Clusters

CRITICAL: Don't just list keywords - provide strategic content recommendations!

## Topic Cluster Strategy

**Analysis Approach**:
1. Group related keywords by topic/theme
2. Identify pillar content opportunities (high-volume head terms)
3. Map supporting content (long-tail variations)
4. Assess current content coverage and gaps
5. Provide specific content creation recommendations

### Cluster [N]: [Topic Name] ([Performance Level]: Core Strength / Growth Opportunity / Weak Coverage)

**Performance Summary**:
- **Total Keywords**: [X] variations
- **Total Impressions**: [X]/week (or /month)
- **Total Clicks**: [X]
- **Average Position**: [X]
- **Overall Assessment**: [1-2 sentences on current state]

**Current Content**:
- Pillar page: [URL if exists, or "MISSING" if needed]
- Supporting pages: [List 2-3 key pages, or note gaps]
- Content quality: [Strong/Moderate/Weak - with brief reasoning]

**Keyword Breakdown**:
| Keyword | Impressions | Position | Clicks | Status |
|---------|-------------|----------|--------|--------|
| [main keyword] | [X] | [Y] | [Z] | [Pillar candidate] |
| [variation 1] | [X] | [Y] | [Z] | [Supporting] |
| [variation 2] | [X] | [Y] | [Z] | [Gap - create content] |

**Strategic Recommendation**:

[For STRONG clusters]:
**Leverage & Expand**: You're already ranking well for [topic]. Focus on:
1. [Specific action to consolidate authority]
2. [Cross-linking strategy]
3. [Content refresh priorities]

[For GROWTH clusters]:
**Build Authority**: Significant search demand but weak current presence. Priority actions:
1. **Create pillar content**: "[Exact title suggestion]" targeting [main keyword]
   - Required sections: [list key topics to cover]
   - Target length: [X] words
   - Key differentiators: [what makes this better than competitors]
2. **Supporting content** (create [N] articles):
   - "[Article title 1]" for [keyword]
   - "[Article title 2]" for [keyword]
3. **Internal linking structure**: [Hub-and-spoke diagram or explanation]
4. **Expected timeline**: [realistic timeframe]
5. **Estimated impact**: +[X] clicks/month after 3-6 months

[For WEAK/COMPETITIVE clusters]:
**Strategic Caution**: High competition or low relevance. Consider:
- Whether to compete at all (resource vs ROI)
- Long-tail alternatives instead of head terms
- Partnership/guest post opportunities

**Example of GOOD cluster analysis**:

> ### Cluster 1: Telegram Automation (Core Strength)
>
> **Performance Summary**:
> - **Total Keywords**: 150+ variations (telegram bot, tg机器人, telegram频道, etc.)
> - **Total Impressions**: ~2,500/week
> - **Total Clicks**: 65/week
> - **Average Position**: 5.8
> - **Overall Assessment**: This is your strongest content cluster with dominant rankings (positions 3-8) across most Telegram-related queries. High CTR indicates strong content-search intent match.
>
> **Current Content**:
> - Pillar page: `/help/guide/automation/telegram-send-message-action` (ranking well)
> - Supporting pages: Multiple template pages, integration guides
> - Content quality: Strong - comprehensive guides with code examples
>
> **Strategic Recommendation**:
>
> **Leverage & Expand**: You're the authority for Telegram automation in your niche. Focus on consolidating:
>
> 1. **Content Refresh Priority**: Update the pillar guide with 2026 Telegram API changes
> 2. **Fill Content Gaps**: Create these high-demand guides:
>    - "Telegram Channel Management Bot - Complete Setup Guide" (targeting "telegram频道机器人", 231 impressions, position 5.5)
>    - "How to Build a Telegram Group Moderation Bot" (targeting "telegram group bot", 103 impressions, currently not ranking with dedicated page)
> 3. **Internal Linking**: Cross-link all 15+ Telegram templates back to the main automation guide. Currently seeing fragmented authority.
> 4. **Comparison Content**: "Telegram vs Discord for Business Automation" - capitalize on "telegram vs" queries (47 impressions)
> 5. **Expected Impact**: Consolidating authority could improve positions 5-8 keywords to 3-5, gaining +15-20 clicks/week

---

> ### Cluster 2: AI Model Integrations (Growth Opportunity)
>
> **Performance Summary**:
> - **Total Keywords**: 混元, 通义千问, doubao, qianwen variations
> - **Total Impressions**: ~3,500/week
> - **Total Clicks**: 4/week (0.11% CTR - TERRIBLE)
> - **Average Position**: 6.4
> - **Overall Assessment**: MASSIVE opportunity - ranking well (positions 5-8) but getting almost zero clicks. Pages ranking don't match user intent.
>
> **Current Content**:
> - Pillar page: MISSING (likely just integration list page ranking)
> - Supporting pages: Generic AI integration mentions
> - Content quality: Weak - not dedicated content for these models
>
> **Strategic Recommendation**:
>
> **Build Dedicated Content Immediately**: This is your highest ROI opportunity.
>
> 1. **Create pillar content** (Priority: URGENT):
>    - "如何在Bika中使用中国AI大模型完整指南" (How to use Chinese AI models in Bika)
>    - Target length: 2,500-3,000 words
>    - Cover: 通义千问, 腾讯混元, 豆包, API setup, use cases, examples
>
> 2. **Supporting content** (create 3 individual guides):
>    - "腾讯混元AI自动化集成教程" for 混元/腾讯混元 (1,047 combined impressions)
>    - "通义千问API接入指南" for 千问/通义千问 (1,930 combined impressions)
>    - "字节豆包AI集成方案" for doubao queries (42 impressions - emerging)
>
> 3. **Technical requirements**:
>    - Each guide needs: API key setup, code examples, workflow templates, troubleshooting
>    - Add FAQ schema for common questions
>    - Include video walkthrough if possible
>
> 4. **Internal linking**:
>    - Hub page linking to individual model guides
>    - Cross-link from general automation guides
>    - Feature in homepage "Popular Integrations" section
>
> 5. **Timeline**: 2-3 weeks for all content
> 6. **Estimated Impact**: Fix CTR from 0.11% to 3-4% = +105-140 clicks/week (+2,500%!)

6. Priority Scoring

## Keyword Prioritization

### Scoring Formula

**Impact Score** (0-100):
- Click potential: 40%
- Impression volume: 30%
- Current position: 20%
- Strategic value: 10%

**Effort Score** (0-100):
- Content exists: 10-40 points
- Content needs creation: 60-80 points
- Technical complexity: varies

**ROI Score** = Impact / Effort

### Top 20 Priority Keywords

| Rank | Keyword | Impact | Effort | ROI | Type | Timeline |
|------|---------|--------|--------|-----|------|----------|
| 1 | [keyword] | 85 | 25 | 3.4 | Quick Win | 2 weeks |
| 2 | [keyword] | 90 | 30 | 3.0 | CTR Fix | Immediate |

### Recommended Focus

**This Month** (Quick ROI):
- Quick Wins: [X] keywords, est. [Y] clicks/month
- CTR Fixes: [X] keywords, est. [Y] clicks/month

**Next Quarter** (Growth):
- High-volume targets: [X] keywords
- Topic clusters: [X] clusters to build

7. Gap Analysis

## Keyword Coverage Analysis

### Missing Opportunities

| Keyword Theme | Est. Volume | Current Ranking | Gap Type | Priority |
|---------------|-------------|----------------|----------|----------|
| [theme] | High | Not ranking | Content gap | High |
| [theme] | Medium | Pos 40+ | Optimization | Medium |

### Intent Coverage

| Intent | Coverage | Industry Avg | Gap | Recommendation |
|--------|----------|--------------|-----|----------------|
| Informational | [X%] | 70% | [gap] | [action] |
| Commercial | [X%] | 60% | [gap] | [action] |
| Transactional | [X%] | 40% | [gap] | [action] |

8. Generate Report

IMPORTANT - Language Detection: Before generating the report, detect the language from the user's original request:

• Chinese request → Generate entire report in Chinese
• English request → Generate entire report in English
• Other languages → Match the user's request language

NEW - STAGED REPORT GENERATION (Critical for avoiding timeouts):

IMPORTANT: Generate the report in stages AUTOMATICALLY without requiring user confirmation for each stage. The stages are only for progress tracking, NOT for approval gates.

Generate the complete report by:

1. Using the Write tool to create the file
2. Building the report content progressively through the Write tool
3. Showing progress updates between major sections
4. DO NOT ask for user approval between stages

Process:

1. Create report file with header (use Write tool)
2. Progress: "▶ Writing executive summary..."
3. Append executive summary (use Write tool with append mode or Edit)
4. Progress: "✅ Complete. ▶ Writing performance overview..."
5. Continue for all sections automatically
6. Final: "🎉 Report saved to ./output/keyword-mining-report-[date].md"

Staged Generation Benefits:

• Prevents timeout from generating massive content in one operation
• User sees progress (stages) instead of hanging
• Each stage is small and fast
• Report file is built incrementally
• NO user intervention required

Using Cached Analysis Results: When writing the Opportunity Mining section, load from cache instead of re-analyzing:

# Read from cache files during analysis
QUICK_WINS=$(cat ./analysis-cache/quick-wins.txt | wc -l)
CTR_OPS=$(cat ./analysis-cache/ctr-ops.txt | wc -l)
# Use these counts in your report generation

Save to: ./output/keyword-mining-report-[date].md

# Keyword Mining Report: [Domain]

**Date**: [date]
**Data Period**: [date range]
**Data Source**: Google Search Console
**Report Language**: [detected from user request]

## Executive Summary

- Total keywords analyzed: [X]
- Core keywords identified: [X] driving [Y]% of clicks
- Quick wins found: [X] keywords on page 2
- CTR opportunities: [X] keywords with [Y] impressions
- High-volume targets: [X] keywords with major potential
- Recommended focus: [priority areas]

## Performance Overview

| Metric | Value | Context |
|--------|-------|---------|
| Total Queries | [X] | [vs industry] |
| Total Clicks | [X] | [trend] |
| Total Impressions | [X] | [trend] |
| Overall CTR | [X%] | [vs benchmark] |
| Avg Position | [X] | [assessment] |

## Core Keywords
[Section 2 content]

## Intent Analysis
[Section 3 content]

## Opportunity Mining
[Section 4 content]

## Topic Clusters
[Section 5 content]

## Priority Rankings
[Section 6 content]

## Gap Analysis
[Section 7 content]

## Implementation Roadmap

### Month 1: Quick Wins
- [ ] Optimize [X] page-2 keywords
- [ ] Fix [X] title tags for CTR
- Expected: +[Y] clicks/month

### Month 2-3: Growth
- [ ] Create content for [X] high-volume keywords
- [ ] Build topic cluster for [theme]
- Expected: +[Y] clicks/month

### Month 4-6: Authority
- [ ] Complete [X] topic clusters
- [ ] Target competitive keywords
- Expected: +[Y] clicks/month

## Tracking Metrics

**Weekly**:
- Position changes for top 50 keywords
- CTR trends for optimized pages

**Monthly**:
- Total clicks and impressions growth
- New keywords ranking top 20
- Topic cluster performance

**Quarterly**:
- Overall organic visibility improvement
- Market share in target topics

Output Files

IMPORTANT - SINGLE OUTPUT FILE ONLY:

Generate ONLY ONE file: ./output/keyword-mining-report-[date].md

• ✅ DO: Create a comprehensive markdown report with all analysis
• ❌ DO NOT: Create separate CSV files
• ❌ DO NOT: Create multiple markdown files
• ❌ DO NOT: Export data to external formats

All keyword data, tables, and analysis should be embedded directly in the single markdown report using markdown tables.

Report Location: ./output/keyword-mining-report-[date].md

Success Metrics

30 Days

• keywords moved from page 2 to page 1
• [Y]% increase in overall CTR
• [Z]% increase in total clicks

90 Days

• [X]% increase in organic clicks
• [Y] new keywords in top 10
• [Z] topic clusters established

Tips

1. Focus on impressions > 100/month - Lower volume too variable
2. Page 2 keywords = gold - Easiest ROI
3. CTR gaps = quick wins - Immediate impact
4. Group into themes - Build topical authority
5. Track weekly - Positions change fast
6. Intent matters - Match content to search intent
7. Update monthly - New opportunities emerge constantly

CRITICAL EXECUTION REMINDERS

❌ What NOT to Do

Don't be a data dumper:

• ❌ Don't just create tables and move on
• ❌ Don't use generic advice like "optimize title tag"
• ❌ Don't skip the "why" and only show the "what"
• ❌ Don't ignore business context and existing content
• ❌ Don't write Python scripts to "process the data"

Example of BAD analysis:

## Quick Wins
| Keyword | Position | Impressions |
|---------|----------|-------------|
| twitter automation | 12.3 | 247 |

Action: Optimize title tag and add internal links.

✅ What TO Do

Be an SEO consultant:

• ✅ Analyze data deeply and explain findings
• ✅ Provide specific, actionable recommendations
• ✅ Connect to business goals and existing assets
• ✅ Diagnose root causes, not just symptoms
• ✅ Give concrete examples and estimates

Example of GOOD analysis:

## Quick Wins - Detailed Analysis

**Top Opportunity: "twitter automation"** (247 impressions, position 12.35)

**Current State**:
- 2 clicks/week (0.81% CTR)
- Ranking page: `/template/x-ai-automated-tweets`
- Position: Bottom of page 2

**Root Cause**:
The page has good content but loses to competitors because:
1. Title tag buries the keyword: "Bika | AI Tweet Templates" instead of "Twitter Automation..."
2. No internal links from high-authority pages
3. Missing FAQ content that could capture featured snippet

**Specific Actions**:
1. Update title tag: "Twitter Automation Tool | AI-Powered Tweet Scheduling | Bika"
2. Add 3 internal links from:
   - Homepage (use anchor "twitter automation tool")
   - Blog post about automation (if exists)
   - Main features page
3. Add FAQ section with schema markup:
   - "How to automate Twitter posts in 2026?"
   - "Best Twitter automation tools comparison"
4. Add "Last updated: [date]" to show freshness

**Expected Outcome**:
- Position improvement: 12.3 → 6-8 (in 2-3 weeks)
- CTR increase: 0.81% → 4-5%
- Traffic gain: +10-12 clicks/week (+500%)
- Total effort: 2-3 hours of work

**ROI**: VERY HIGH - Minimal effort, significant return

Processing Instructions

CRITICAL - NO PYTHON SCRIPTS ALLOWED:

❌ ABSOLUTELY FORBIDDEN:

• Do NOT write Python files to process data
• Do NOT create analyze_topics.py, export_csv.py or any other Python scripts
• Do NOT use Python for data analysis, filtering, or CSV generation

✅ MANDATORY - USE BASH TOOLS ONLY:

All data processing MUST be done with bash commands:

• head, tail - preview data
• wc - count lines/words
• grep - search and filter
• awk - text processing and calculations
• sort - sorting data
• uniq - remove duplicates
• cut - extract columns
• jq - JSON processing (essential for MCP data)

NEW - PERFORMANCE OPTIMIZATION REQUIREMENTS:

✅ PARALLEL PROCESSING (MANDATORY for large datasets):

• ALWAYS run analysis tasks in parallel using background jobs (&)
• Use wait to track completion
• Cache results to ./analysis-cache/ for reuse
• This reduces processing time from 30+ minutes to 5-10 minutes

✅ PROGRESS FEEDBACK (MANDATORY for all operations):

• ALWAYS show progress with emoji indicators
• Use format: ⏳ [Step X/Total] Description...
• Show subtasks: ▶ Subtask N/M: Details...
• Show completion: ✅ Task complete
• Show final summary: 🎉 All tasks completed!

✅ DATA PRE-FILTERING (MANDATORY after data fetch):

• ALWAYS filter data after fetching: impressions >= 10 AND position <= 100
• If filtered results < 100 keywords, relax to: impressions >= 5 AND position <= 100
• This reduces processing load while preserving meaningful opportunities
• Prevents timeout on large datasets

✅ TOKEN OPTIMIZATION (MANDATORY for cost control):

• Set MCP rowLimit: 10000 (not 25000) - balances coverage and cost
• Use 14-day default range - shorter period = less data
• Avoid verbose intermediate output - don't print full datasets
• Limit deep analysis to Top 10-15 opportunities (not every single keyword)
• Use cached results - don't re-read raw JSON multiple times
• Target: Keep total token usage under 150K per report

Token Budget Breakdown:

Optimized Data Processing Workflow with MCP JSON Data:

# Step 0: PRE-FILTERING (CRITICAL - Do this FIRST)
echo "⏳ [1/4] Pre-filtering data to reduce processing load..."
cat mcp-response.json | jq '.[0].text' | jq '.rows' > raw-rows.json
TOTAL_ROWS=$(cat raw-rows.json | jq '. | length')
cat raw-rows.json | jq '[.[] | select(.impressions >= 10 and .position <= 100)]' > filtered-rows.json
FILTERED_ROWS=$(cat filtered-rows.json | jq '. | length')
echo "   ✅ Filtered to $FILTERED_ROWS rows (removed $((TOTAL_ROWS - FILTERED_ROWS)) low-value rows)"
# If still too few keywords, relax filter further
if [ "$FILTERED_ROWS" -lt 100 ]; then
  echo "   ⚠️  Only $FILTERED_ROWS keywords, relaxing filter..."
  cat raw-rows.json | jq '[.[] | select(.impressions >= 5 and .position <= 100)]' > filtered-rows.json
  FILTERED_ROWS=$(cat filtered-rows.json | jq '. | length')
  echo "   ✅ Expanded to $FILTERED_ROWS rows"
fi

# Step 1: PARALLEL ANALYSIS (All tasks run simultaneously)
echo "⏳ [2/4] Running parallel analysis..."
mkdir -p ./analysis-cache

# Task A: Core keywords
echo "   ▶ Task 1/4: Core keywords..."
(cat filtered-rows.json | jq -r '.[] | "\(.keys[0])|\(.clicks)|\(.impressions)|\(.ctr*100)|\(.position)"' | \
  sort -t'|' -k2 -nr | head -30 > ./analysis-cache/core-keywords.txt) &
PID_CORE=$!

# Task B: Quick wins
echo "   ▶ Task 2/4: Quick wins..."
(cat filtered-rows.json | jq -r '.[] | select(.position > 10 and .position < 20 and .impressions >= 100) | "\(.keys[0])|\(.position)|\(.impressions)|\(.clicks)|\(.keys[1])"' | \
  sort -t'|' -k3 -nr | head -20 > ./analysis-cache/quick-wins.txt) &
PID_QUICK=$!

# Task C: CTR opportunities
echo "   ▶ Task 3/4: CTR opportunities..."
(cat filtered-rows.json | jq -r '.[] | select(.position <= 10 and .impressions >= 500) | "\(.keys[0])|\(.position)|\(.impressions)|\(.ctr*100)|\(.keys[1])"' | \
  awk -F'|' 'BEGIN{bench[1]=28;bench[2]=18;bench[3]=11;bench[4]=8;bench[5]=7;bench[6]=5;bench[7]=4;bench[8]=3;bench[9]=2;bench[10]=1.5}
    {pos=int($2);exp=bench[pos];if(exp && $4<exp*0.7) print $0}' > ./analysis-cache/ctr-ops.txt) &
PID_CTR=$!

# Task D: High volume
echo "   ▶ Task 4/4: High volume keywords..."
(cat filtered-rows.json | jq -r '.[] | select(.impressions >= 1000) | "\(.keys[0])|\(.impressions)|\(.position)|\(.clicks)|\(.keys[1])"' | \
  sort -t'|' -k2 -nr | head -30 > ./analysis-cache/high-volume.txt) &
PID_VOL=$!

# Wait for all parallel tasks
wait $PID_CORE && echo "   ✅ Task 1/4 complete"
wait $PID_QUICK && echo "   ✅ Task 2/4 complete"
wait $PID_CTR && echo "   ✅ Task 3/4 complete"
wait $PID_VOL && echo "   ✅ Task 4/4 complete"
echo "🎉 All parallel analysis completed!"

# Step 2: USE CACHED RESULTS for report generation
echo "⏳ [3/4] Analyzing cached results..."
# Load from ./analysis-cache/*.txt files instead of re-processing
echo "📊 Analysis complete: $(cat ./analysis-cache/quick-wins.txt | wc -l) quick wins identified"

# Step 3: REPORT GENERATION (Automatic, no user confirmation)
echo "⏳ [4/4] Generating comprehensive report..."
echo "   📝 This will proceed automatically through all stages"
# Generate report using Write tool progressively
# Show progress but DO NOT ask for approval
echo "🎉 Report saved to ./output/keyword-mining-report-$(date +%Y%m%d).md"

Key Performance Improvements:

Progress Indicators (Use these emojis):

IMPORTANT - jq is your best friend for MCP data:

• Always use jq to parse JSON responses from MCP tools
• Use jq -r for raw output (no quotes) when piping to other tools
• Chain jq commands for complex filtering
• Combine with sort, uniq, head for analysis

If you need complex analysis, use inline bash commands with jq, NOT separate Python files

Report Quality Checklist

Before finalizing your report, verify:

• Each top opportunity has detailed root cause analysis
• Specific action items (not "optimize", but HOW to optimize)
• Referenced actual URLs/pages where possible
• Included traffic gain estimates with reasoning
• Connected to business context ("You already have X page...")
• Flagged urgent items with clear justification
• Grouped related opportunities into campaigns
• Provided both quick wins AND long-term strategy
• Report reads like a consultant's analysis, not a data dump

Success = Strategic Value

Your goal is for the reader to finish your report and think:

✅ "Wow, this is incredibly actionable" ✅ "I know exactly what to do next week" ✅ "The ROI calculations help me prioritize" ✅ "They really understand my business"

NOT:

❌ "This is just my data in table form" ❌ "These recommendations are too generic" ❌ "I don't understand why this matters" ❌ "A Python script could have done this"