Overview
This is the skill of deciding "what to target" before starting new content, pages, or campaigns. It prioritizes keywords based on search volume, competition, and search intent, and groups related keywords into topic clustersto create a keyword map that simultaneously targets Naver, Google, and AI search (GEO).
In 2026, search has moved from competing over single keywords to competing over topic and query groups. Google AI search expands a single query into multiple related queries (query fan-out), and AI search cites well-structured pages even if they are not in the top 3. Therefore, from the research stage, you should approach it not as individual keywords but as clusters + search intent.
Original reference: github.com/coreyhaines31/marketingskills (MIT — full notice: repository root NOTICE) — the methodologies of the ai-seo, programmatic-seo, and competitors skills have been restructured to fit the Korean search market (Naver, Google, GEO).
Trigger Keywords
keyword research, keyword discovery, search term discovery, search intent, long-tail keywords, topic clusters, pillar clusters, keyword difficulty, search volume analysis, content selection, query fan-out, seed keywords
Practical Keyword Research Patterns (2026 Naver, Google, GEO)
1. Four Search Intent Categories (Search Intent)
For keywords, intentcomes before search volume. If the intent does not match, traffic will not convert even if it comes in.
| Intent | Query Pattern | Content to Target | Distance to Conversion |
|---|---|---|---|
| Informational (Informational) | "What is OO", "How to OO", "Why OO" | Guides, explanations, how-tos | Far (awareness stage) |
| Navigational (Navigational) | "OO brand", "OO login" | Official pages, landing pages | — (already aware) |
| Commercial Investigation (Commercial) | "OO comparison", "OO recommendations", "OO reviews" | Comparison tables, reviews, listicles | Close (consideration stage) |
| Transactional (Transactional) | "Buy OO", "OO price", "Apply for OO" | Product, pricing, application pages | Closest (decision) |
Key point: For new sites and low-authority domains, start with informational long-tail + commercial investigation. Transactional big keywords are too competitive to rank for early on.
2. Long-tail First Strategy (Long-tail)
- Big keyword (head): high search volume and extreme competition, ambiguous intent → hard to capture
- Long tail (3+ words): low search volume but low competition and clear intent → high conversion rate
- Naver-specific: specific queries of 15+ characters doubled (2026). Local, specific queries like "Gangnam 40s women Pilates recommendation" are the long-tail goldmine
- When multiple long-tail keywords come together to cover a single head topic, it is promoted to a cluster
3. Query Fan-out & Topic Cluster (Query Fan-out)
Google AI Search, for a single query like "lawn care method," simultaneously generates related queries such as weeding, chemicals, and weed prevention related queries simultaneously. If you target only individual keywords, you will be omitted from the fan-out branches.
Pillar-cluster structure:
[필러 페이지] 이메일 마케팅 기초 (헤드, 정보형)
├─ [클러스터] 이메일 리스트 만드는 법
├─ [클러스터] 이메일 세그먼트 전략
├─ [클러스터] 제목 A/B 테스트
└─ [클러스터] 이메일 도달률 높이기
When the pillar links the cluster with internal links, topical expertise (Naver C-Rank, Google topical authority) accumulates. One cluster usually consists of 5-10 related queries.
4. Keyword Difficulty Assessment (Keyword Difficulty)
Practical estimation method when there is no precise KD score (Ahrefs, Semrush):
| Signal | Interpretation |
|---|---|
| Top 10 are only large media and official domains | High difficulty → avoid |
| Personal blogs and small sites are mixed in the top ranks | Entry possible → worth targeting |
| Query is specific and long-tail | Low difficulty |
| Commercial investigation type but comparison content is weak | Opportunity |
| AI Search cites you if the structure is good even if you are not in the top 3 | Ranking difficulty ≠ AI citation difficulty |
AI citation depends less on ranking than on extractable structure(definition blocks, tables, 40-60 character answer paragraphs). Even if ranking difficulty is high, you can bypass it with GEO.
5. Korea's 3-channel keyword sources
| Channel | Discovery tools and methods |
|---|---|
| Naver | Naver Search Ad keyword tool (search volume, competition), autocomplete, related searches, popular topics |
| Google Keyword Planner, Search Console queries, People Also Ask, autocomplete | |
| AI Search (GEO) | Query ChatGPT and Perplexity directly to collect "follow-up questions AI asks back," prioritizing comparison and list-type queries |
6. Keyword Priority Matrix
Score each candidate by search volume × competition × intent fit × conversion distanceand sort them.
우선순위 = (검색량 정규화 + 의도 적합) − (경쟁 난이도) + 전환 가치 가중
For the first 12 weeks, concentrate resources on long-tail keywords with "medium search volume + low competition + clear intent" to secure quick wins, then expand to head keywords.
Keyword research anti-patterns
- Betting everything on big keywords based only on search volume (new domains will not rank)
- Ignoring intent (transactional CTA on informational traffic → bounce)
- Listing individual keywords only, no clusters (missing fan-out and topical authority)
- Naver only or Google only (the three channels have different intent)
- Research premised on keyword stuffing (10% drop in AI visibility, Naver penalty)
Workflow
Step 1: Confirm research scope
Confirm the topic and industry, target channels (Naver/Google/AI search/all), business goals (traffic/leads/conversions), target readers, and 2-5 seed keywords. Also check whether search volume data is accessible (Naver Search Ad account, GSC, tools).
Step 2: Seed expansion
From the seed keywords, expand candidates using autocomplete, related searches, People Also Ask, and AI follow-up questions. Tag each candidate's expected search intent with one of four categories.
Step 3: Search intent classification & long-tail decomposition
Divide candidates into informational/navigational/commercial investigation/transactional, and decompose head terms into long tails. Mark local and specific long tails separately.
Step 4: Topic clustering
Group related keywords into a pillar-cluster structure. Compose one set of 1 pillar + 5-10 clusters, and specify internal link directions.
Step 5: Difficulty & priority calculation
Assign each keyword a difficulty (high/medium/low) and a priority score. Separate the initial 12-week targets from the mid-term expansion targets.
Step 6: Generate research deliverables
Organize a keyword list table (search volume, competition, intent, cluster), a topic cluster map, a priority roadmap, and targeting strategies by channel (Naver/Google/GEO).
Usage examples
Example 1: "Pilates studio blog, pull keywords for what to write first" → Seed expansion → intent classification → local long-tail discovery → 3 clusters → 12-week targets
Example 2: "It's a B2B SaaS, research the query candidates that ChatGPT and Perplexity will cite" → Collect comparison- and list-type queries → map query fan-out → extraction structure requirements → GEO priorities
Output format
- Keyword list table (keyword · search volume · competition · search intent · cluster it belongs to)
- Topic cluster map (pillar + clusters + internal link directions)
- Priority roadmap (initial 12 weeks / mid-term expansion)
- Targeting strategies by channel (Naver · Google · GEO)
Precautions
| Situation | Response |
|---|---|
| When there is no search volume data | Guide them to Naver Search Ad and Google Keyword Planner, then present priorities using relative estimates (autocomplete appearances, related search term density) |
| When there is no precise keyword difficulty score | Estimate high/medium/low from the composition of top 10 domains. Recommend recalibrating after acquiring a tool |
| When the seed keywords are ambiguous | First derive seed candidates from industry, target, and goals, then expand |
| Requests to diagnose SEO problems on existing pages | Outside the scope of this skill. moai-marketer:marketing-seo-auditguide them to |
Post-processing (required chain — narrative recommendations only)
Keyword strategy description, cluster rationale, channel strategy, etc. Prose deliverablesgo through the post-processing chain before publication. Keyword lists, search volume figures, priority score tables, etc. Numeric/table deliverables are not subject to this.
Recommended chain (narrative recommendations): marketing-keyword-research(서술형 권고) → moai-coworker:ai-slop-reviewer → moai-writer:korean-humanize → 최종 검수
Related skills
| Skill | When to use |
|---|---|
moai-marketer:marketing-seo-audit |
Diagnose the actual page/site SEO health after keywords are decided (run separately) |
moai-marketer:content-blog |
Write an SEO blog using the discovered keywords |
moai-marketer:content-editorial-calendar |
Expand clusters into a content publishing schedule |
moai-marketer:marketing-campaign-planner |
Expand keywords into search/content campaigns |
moai-coworker:ai-slop-reviewer |
Post-chaining (required, narrative recommendations): AI pattern review of strategy prose |
moai-writer:korean-humanize |
Post-chaining (required, narrative recommendations): remove Korean AI tells as the next step after ai-slop-reviewer |
When not to use this skill
- Diagnose the SEO health of already-published pages/sites:
moai-marketer:marketing-seo-audituse - Write actual blog body text for a specific keyword:
moai-marketer:content-bloguse - Publishing schedule planning:
moai-marketer:content-editorial-calendaruse