varnan-tech/seo-keyword-research
SEO keyword research workflow for blog generation using Google Trends data. Use when writing blog posts, planning content calendars, or optimizing articles for search engines. Finds breakout keywords, builds content structure, and generates SEO-optimized blog outlines targeting tech and developer audiences.
npx skills add https://github.com/Varnan-Tech/opendirectory --skill seo-keyword-research
Find trending, high-opportunity keywords BEFORE writing blog content. This skill turns generic blog topics into SEO-optimized content that ranks.
Always research keywords BEFORE generating blog content.
BAD: Write blog -> Hope it ranks -> Usually doesn't
GOOD: Research keywords -> Find breakout opportunity -> Write optimized blog -> Ranks well
When analyzing Google Trends RELATED_QUERIES results, prioritize keywords in this order:
formatted_value: "Breakout" = 5000%+ growthformatted_value: "+100%" or higherformatted_value: "+50%" to "+99%"Query RELATED_QUERIES with the user's blog topic:
curl -s "https://serpapi.com/search?engine=google_trends&q=USER_TOPIC&data_type=RELATED_QUERIES&date=today+3-m&api_key=$SERPAPI_KEY"
From the response, extract:
Select the primary keyword:
Query RELATED_TOPICS with the same topic:
curl -s "https://serpapi.com/search?engine=google_trends&q=USER_TOPIC&data_type=RELATED_TOPICS&date=today+3-m&api_key=$SERPAPI_KEY"
Extract topic titles from rising + top results. These become your H2 section headings (pick 3-5).
Only if choosing between multiple candidate keywords or validating viability:
curl -s "https://serpapi.com/search?engine=google_trends&q=KEYWORD&data_type=TIMESERIES&date=today+12-m&api_key=$SERPAPI_KEY"
Compare recent 2-month average vs. earlier 2-month average. If recent > earlier, trend is rising — proceed. If declining, consider a different keyword.
Build the outline using this structure:
Title: [Primary Keyword] — [Benefit/Number] [Year]
(max 60 characters, must include primary keyword)
Meta Description: (150-160 chars, primary + 1-2 secondary keywords)
# [H1 — same as or variation of title]
## Introduction (150 words)
- Primary keyword in first 100 words
- Hook with a problem or question
- Preview what they'll learn
## [H2: Related Topic 1 from Step 2]
### [H3: Long-tail question from Step 1]
Content answering the question (150-200 words)
### [H3: Another long-tail question]
Content (150-200 words)
## [H2: Related Topic 2]
### [H3: Long-tail question]
### [H3: Long-tail question]
## [H2: Related Topic 3]
### [H3: Long-tail question]
### [H3: Long-tail question]
## Conclusion (100 words)
- Summarize key points
- Primary keyword mentioned once
- Call-to-action
Target: 1500-2500 words total
| Location | Rule |
|----------|------|
| Title | Include primary keyword, max 60 chars |
| H1 | Same as title or slight variation |
| H2 headings (3-5) | Use related topics, natural language |
| H3 headings (8-12) | Use long-tail keywords, question format |
| First paragraph | Primary keyword in first 100 words |
| Body content | Primary keyword 1-2% density, secondary 0.5-1% |
| Conclusion | Primary keyword once |
| Meta description | Primary + 1-2 secondary, 150-160 chars |
Never keyword-stuff. Content must read naturally. Google penalizes unnatural repetition.
Before generating the blog, verify:
If no breakout or high-growth keywords exist for the topic, inform the user that SEO opportunity is limited and suggest alternative angles or related topics that do have growth.
Free tier: 250 searches/month
| Strategy | Calls/Blog | Monthly Capacity |
|----------|-----------|-----------------|
| Minimal (recommended) | 2 | 125 blogs |
| Standard | 3 | 83 blogs |
| Complete | 4 | 62 blogs |
Default to 2 calls (RELATED_QUERIES + RELATED_TOPICS). Only add TIMESERIES or GEO_MAP when specifically needed.
For a complete working example, run:
python scripts/blog_seo_research.py "your blog topic"
See scripts/blog_seo_research.py for the implementation.
Take varnan-tech/seo-keyword-research from the repository into ~/.claude/skills for personal
use, or into .claude/skills inside a project.
The agent identifies a skill by the name field in its header. Two skills with the
same name cannot sit side by side — one of them will be ignored.