AI Practice Knowledge BaseAI Self-Growth System
Keyword Matrix Design
PremiumExpand one seed term into 100 rankable long-tail keywords
Keyword Matrix Design
Most people only have a "keyword list". You need a "keyword matrix".
What you will get in this chapter
- The minimum process to build a keyword matrix
- A keyword scoring model (quantitative filtering)
- Structured standards for the task list
One-sentence definition
Keyword matrix = seed terms * variant dimensions * intent labels.
A matrix is multiplication, not addition.
Minimum viable keyword matrix (MVS)
| Step | You need | Acceptance result |
|---|---|---|
| Seed terms | 1-3 core terms | Split into 30+ variants |
| Dimensions | 2-3 dimensions | Multiplicative expansion |
| Scoring | Keyword scoring model | Can filter P0/P1/P2 |
| Task list | Structured JSON | Ready for generation |
Qualified signal: produce 50-200 executable tasks within 1 hour.
Four-step matrix build
- Pick seeds: clear intent, rich variants
- Set dimensions: style/industry/scenario/format
- Expand: use tools to generate long-tail
- Score: model scoring, prioritize
Keyword scoring model (core)
Scoring dimensions
| Dimension | Meaning | Range |
|---|---|---|
| VolumeScore | Search volume (log normalized) | 0-1 |
| DifficultyScore | 1 - KD/100 | 0-1 |
| IntentScore | Commercial intent strength | 0-1 |
| DataScore | Has data support | 0-1 |
| SERPScore | How beatable the SERP is | 0-1 |
Scoring formula (ready to use)
KeywordScore = 0.35*VolumeScore
+ 0.25*DifficultyScore
+ 0.20*IntentScore
+ 0.10*DataScore
+ 0.10*SERPScoreIntent scoring examples
- Includes
tool / generator / free= 1.0 - Includes
best / top / compare= 0.8 - Includes
how to / what is= 0.5 - Pure informational query = 0.3
Priority thresholds
- P0 (do first): Score >= 0.65
- P1 (doable): 0.5 <= Score < 0.65
- P2 (later): Score < 0.5
Core metrics (must track)
Definition (default):
- Time window: unless stated otherwise, use the last 7 days rolling.
- Data source: use one trusted source (GA4/GSC/platform console/logs) and keep it consistent.
- Scope: only the current product/channel, exclude self-tests and bots.
| Metric | Meaning | Pass line |
|---|---|---|
| P0 Ratio | Share of P0 keywords | 10%-30% |
| Avg Score | Average score | >= 0.5 |
| Coverage | Total keywords | >= 50 |
| Task Ready Rate | Fields complete and ready | >= 80% |
Task list structure (recommended JSON)
{
"meta": {
"seed": "ai image generator",
"generatedAt": "2025-12-25",
"total": 80
},
"tasks": [
{
"slug": "anime",
"keyword": "anime ai image generator",
"volume": 50,
"kd": 15,
"intent": "tool",
"score": 0.68,
"priority": "P0",
"status": "pending"
}
]
}Acceptance checklist
Seed terms and dimensions are defined and reusable
Scoring model is running and graded (P0/P1/P2)
Task list is structured and ready for generation
Common mistakes
- Only look at search volume -> KD too high, cannot rank
- Too few dimensions -> cannot scale
- No scoring model -> all keywords look equally important
- No data support -> even if generated, you cannot write well
Summary
Key takeaways
1. A matrix is multiplicative expansion, not a stacked list.
2. Scoring quickly filters low-value keywords.
3. A structured JSON task list is the prerequisite for automation.
Next chapter: Drip Release Strategy - how to steadily release matrix content to search engines.