Keyword Research
Usage
Use when planning content around a topic before writing, building a keyword map for a new content area, finding question-based keywords for GEO optimization, or identifying gaps where competitors rank and you don't.
Process
Step 1: Gather Inputs
Ask the user for:
- Seed topic or keyword — e.g., "AI search optimization", "B2B SaaS customer acquisition"
- Country code (optional) — for localised volumes (default: "us")
- Competitor domain(s) (optional) — to identify keyword gaps
- Number of clusters (optional) — default: let the data dictate
Step 2: Validate & Prepare
- Confirm the Keywords Everywhere MCP server is connected. If not configured, tell the user — this skill requires it.
- Check credit balance with Get Credit Balance before starting — warn the user if credits are low.
Credit estimate formula: (related_count + pasf_count) * 2 credits for expansion + total_unique_keywords * 1 credit for metrics. A typical run with 100 related + 100 PASF keywords costs ~500 credits. Warn if balance would drop below 1,000 after the run.
Step 3: Expand the Seed
Run two Keywords Everywhere tools against the seed keyword:
- Get Related Keywords (num: 100) — returns a list of keyword strings (no metrics yet)
- Get "People Also Search For" Keywords (num: 100) — returns a list of keyword strings (no metrics yet)
If PASF returns empty results: This is common for newer or niche terms. Proceed with the related keywords only. If the combined list is thin (< 30 keywords), consider running a second expansion on a broader variant of the seed.
Combine the results into a single deduplicated keyword list.
If the seed returns fewer than 20 keywords total, it may be too narrow. Suggest broader alternatives to the user.
Step 3b: Pull Metrics
Run Get Keyword Data on the deduplicated keyword list to get volume, CPC, competition, and trend data for every keyword.
Batch in groups of 50 keywords per API call to avoid oversized responses. Run batches in parallel where possible.
Parameters: country: "us" (or from user input), currency: "usd", dataSource: "cli" (includes clickstream data for more accurate volumes).
Step 4: Competitor Gap Analysis (optional)
If competitor domain(s) were provided:
- Get Domain Keywords for each competitor — returns keywords they rank for
- Cross-reference with the expanded keyword list from Step 3
- Flag keywords where competitors rank but the user's domain doesn't — these are gaps
- Add any high-volume competitor keywords that didn't appear in the Step 3 expansion
If no competitors provided, skip this step.
Step 5: Cluster by Intent
Group the full keyword list into semantic clusters. Each cluster represents a potential piece of content.
Pre-clustering: Filter brand/navigational noise
Before clustering, separate out brand-specific and product-name keywords. These are navigational queries for specific tools, not topics you'd write content about. List them in an "Excluded: Brand/Navigational Keywords" section at the end — they're useful market intelligence but shouldn't inflate your topic clusters.
Clustering rules:
- Group keywords that would be answered by the same piece of content
- Name each cluster after its core topic (not the highest-volume keyword)
- Assign an intent to each cluster:
- Informational — "what is", "how to", "why does" — answered by blog posts, guides
- Commercial — "best", "vs", "review", "pricing" — answered by comparison pages, landing pages
- Navigational — brand-specific or product-specific queries — answered by product/feature pages
- Transactional — "buy", "sign up", "get started" — answered by landing pages, pricing pages
- A cluster should have 3–20 keywords. If larger, split into sub-clusters. If smaller, consider merging.
- Keywords with very low volume (< 10/mo) can be grouped into clusters but shouldn't form their own cluster
- Drop zero-volume keywords entirely
Within each cluster, identify:
- The primary keyword — highest volume keyword that best represents the cluster's intent
- Question keywords — any keywords phrased as questions (valuable for H2 headings and FAQ sections)
- Long-tail keywords — lower volume, more specific phrases (valuable for weaving into content)
Step 6: Prioritize
Score each cluster for content priority:
| Signal | What to look at |
|---|---|
| Total volume | Sum of all keyword volumes in the cluster |
| Competition | Average competition score (lower = easier to rank) |
| Gap opportunity | Are competitors ranking here and you're not? (from Step 4) |
| Intent fit | Does this cluster match content you'd actually create? |
| Question density | Clusters with more question keywords are better for GEO |
Rank clusters by a blended priority — not just volume. A low-competition cluster with good question density and a clear gap often beats a high-volume, high-competition cluster.
Output Format
# Keyword Research: [Seed Topic] **Seed:** [seed keyword] **Date:** [current date] **Total keywords found:** [X] **Clusters:** [X] --- ## Cluster 1: [Cluster Name] **Intent:** Informational / Commercial / Navigational / Transactional **Primary keyword:** [keyword] ([volume]/mo) **Total cluster volume:** [X]/mo **Avg competition:** [X] **Gap opportunity:** Yes / No **Priority:** High / Medium / Low | Keyword | Volume | CPC | Competition | Type | |---------|--------|-----|-------------|------| | [keyword] | [vol] | [cpc] | [comp] | Primary | | [question keyword]? | [vol] | [cpc] | [comp] | Question | | [long-tail keyword] | [vol] | [cpc] | [comp] | Long-tail | --- ## Cluster 2: [Cluster Name] [Same format...] --- ## Summary | Cluster | Intent | Primary Keyword | Volume | Competition | Priority | |---------|--------|----------------|--------|-------------|----------| | [name] | [intent] | [keyword] | [vol] | [comp] | High | | [name] | [intent] | [keyword] | [vol] | [comp] | Medium | ## Recommended Next Steps - [Which clusters to write first and why] - [Suggested content type for each high-priority cluster] - [Any gaps that need competitor research first]
Rules
- Never invent keyword volumes or competition scores — all data must come from the Keywords Everywhere API.
- Never cluster keywords you haven't actually retrieved — don't pad clusters with guesses.
- Never present unclustered keyword dumps — always group and prioritize.
- If the MCP server is not connected, stop and tell the user.
- If credit balance is low (< 100 credits), warn before starting.
- If the seed returns fewer than 20 keywords, suggest broadening.
- If the seed returns 1000+ keywords, confirm they want the full expansion or suggest narrowing.
- Country code matters — volumes vary significantly by market. Default to "us" if no country specified.
- Run this skill periodically (quarterly) on core topics to catch new keywords and shifting volumes.
