Claude keyword research skill: 10 SEO workflows you can run today

dataforseo

Attach real search volume and CPC to each candidate the skill surfaces.

seo-content-writer

Hand the tagged list straight to drafting once you pick winners.

Difficulty + opportunity scoring of an existing list

Feed in a keyword list and get each row scored for difficulty and opportunity, so the easy-but-valuable terms float to the top instead of the vanity head terms.

ForTeams with a 200-row export and no idea which rows to attack first.

The prompt

Use the keyword-research skill. Read keywords.csv (columns: keyword, volume). Score each row for difficulty (1-100, based on SERP competitiveness signals) and opportunity (a blend of volume, intent match to a SaaS product, and how winnable the SERP looks for a mid-authority site). Sort descending by opportunity. Add a 'Verdict' column: Attack now / Build authority first / Skip. Output the top 25 rows as a table and a two-sentence summary of where the quick wins cluster. Save to scored-keywords.md.

What slides.md looks like

## Scored keywords (top 25 by opportunity)

| Keyword | Volume | Difficulty | Opportunity | Verdict |
|---------|--------|-----------|-------------|---------|
| claude code skills | 1,300 | 28 | 86 | Attack now |
| mcp directory | 880 | 22 | 81 | Attack now |
| best seo mcp | 210 | 19 | 74 | Attack now |
| seo mcp server | 320 | 47 | 58 | Build authority first |
| seo | 410,000 | 96 | 11 | Skip |

Quick wins cluster around low-difficulty, product-aligned long-tail terms (difficulty < 30).

One-line tweak

Swap the SaaS-product assumption for 'an affiliate review site' and the opportunity score re-weights toward commercial-intent terms instead.

Pairs with

ahrefs

Pull real Keyword Difficulty so the score isn't a pure estimate.

semrush

Cross-check difficulty and volume against a second data provider.

Cluster keywords into content briefs

Collapse a flat keyword list into topic clusters, then turn each cluster into a one-page content brief — the target term, the supporting terms, the angle, and the questions to answer.

ForEditors who need to turn a keyword dump into assignable articles.

The prompt

Use the keyword-research skill. Read scored-keywords.md. Group the keywords into 6-10 topic clusters by semantic relevance AND shared search intent — do not merge an informational cluster with a transactional one even if the words look similar. For each cluster, output a content brief: cluster name, primary target keyword, 4-6 supporting keywords, the search intent, the recommended page type (guide / comparison / tool page), and 3 questions the page must answer. One brief per cluster. Save each to briefs/<cluster-slug>.md.

What slides.md looks like

## Brief: "Claude skills for SEO"

- Primary keyword: claude code skills
- Supporting: keyword research skill, seo audit skill, seo skill claude
- Intent: Informational (with a soft commercial edge)
- Page type: Guide
- Must answer:
  1. What is a Claude skill vs an MCP server?
  2. Which SEO skills exist and what does each do?
  3. How do you install one?

Cluster size: 11 keywords · combined volume 4,200

One-line tweak

Add 'and reject any cluster smaller than 3 keywords as not worth a standalone page' to avoid thin-content briefs.

Pairs with

content-strategy

Sequence the briefs into a publishing order with dependencies.

seo-content-writer

Each brief becomes the spec for a drafted article.

SERP-intent classification (info / nav / commercial / transactional)

Run a list purely through intent classification, so you stop writing a how-to guide for a query that wants a product page (and losing every time).

ForAnyone whose 'great' article ranks page two because it mismatched intent.

The prompt

Use the keyword-research skill. Classify the search intent of each keyword in queries.txt. For each, return the dominant intent (informational / navigational / commercial / transactional), a confidence note, and the SERP feature most likely to dominate (featured snippet, product carousel, local pack, video). Where a keyword is mixed-intent, say so and recommend whether to split it across two pages. Output a table plus a short list of the queries where my current page type is probably wrong. Save to intent-map.md.

What slides.md looks like

## Intent map

| Keyword | Intent | Likely SERP feature | Mixed? |
|---------|--------|---------------------|--------|
| how to install mcp server | Informational | Featured snippet | No |
| best mcp server | Commercial | Carousel + listicles | No |
| mcp server | Informational | Definition + PAA | Yes — split |
| buy mcp hosting | Transactional | Ads + product pages | No |

Probable mismatches: 'mcp server' is being targeted with a product page; it wants a definition first.

One-line tweak

Add 'and rewrite each title tag to match the classified intent' to ship the fix in the same pass.

Pairs with

seo-meta-optimizer

Rewrite titles and meta to match the classified intent.

serper-search-and-scrape

Pull the live SERP so the intent read is grounded in real results.

Competitor gap analysis

Compare your keyword footprint to a competitor's and surface the terms they rank for that you don't — the content gaps worth closing.

ForMarketers who know a rival is eating their lunch but can't see where.

The prompt

Use the keyword-research skill. I rank for the keywords in mine.csv. A competitor ranks for competitor.csv. Find the gap: keywords the competitor covers that I don't, filtered to ones that match my product's intent and that a site at my authority level could realistically win. For each gap keyword, note the competitor's likely page type and a one-line angle I could take to do it better. Sort by estimated opportunity. Output the top 20 gaps as a table. Save to gap-analysis.md.

What slides.md looks like

## Competitor gap (top 20)

| Gap keyword | Their page type | My angle | Opportunity |
|-------------|----------------|----------|-------------|
| mcp server vs rest api | Blog explainer | Add a decision table they lack | 79 |
| claude skill examples | Listicle | Live install cards, not screenshots | 72 |
| self-host mcp server | Tutorial | Cover Docker + Windows native | 68 |

20 gaps · 12 winnable at current authority · 8 need link-building first.

One-line tweak

Point it at three competitor exports at once and ask for 'keywords at least two of the three rank for' to find the consensus gaps.

Pairs with

content-gap-analysis

Go deeper on the topical holes the keyword gap implies.

firecrawl

Crawl the competitor's site to derive their footprint when you have no export.

Long-tail mining from a head term

Pull the long tail out of a single competitive head term — the specific, lower-volume, higher-intent variants a smaller site can actually rank for.

ForNew sites that can't win 'crm' but can win 'crm for solo real estate agents'.

The prompt

Use the keyword-research skill. Head term: 'mcp server'. Mine the long tail: generate 50 specific multi-word variants by combining the head term with audiences, use cases, modifiers (for/with/without/vs/best/free/self-hosted), and question framings. For each, estimate difficulty as Low / Medium / High and tag the intent. Keep only Low and Medium difficulty rows. Output them grouped by the modifier pattern that produced them. Save to long-tail-mcp-server.md.

What slides.md looks like

## Long-tail — head: "mcp server"

### Pattern: "for <audience>"
- mcp server for data analysts — Low — Informational
- mcp server for solo founders — Low — Commercial

### Pattern: "<modifier>"
- free self-hosted mcp server — Medium — Transactional
- mcp server without docker — Low — Informational

50 generated · 31 kept (Low/Medium) · 19 dropped (High difficulty).

One-line tweak

Add 'and cluster the kept rows into the 5 articles that would cover them all' to jump straight from mining to a plan.

Pairs with

programmatic-seo

Turn a repeating long-tail pattern into a templated page set.

dataforseo

Validate that the mined long-tail terms have non-zero real volume.

GEO / answer-engine query expansion

Expand a topic into the conversational, question-shaped queries people type into AI answer engines — the GEO layer the skill's description calls out by name.

ForTeams who want to be cited by ChatGPT, Perplexity, and AI Overviews, not just Google's ten blue links.

The prompt

Use the keyword-research skill in GEO mode. Topic: 'choosing an MCP server'. Expand into the natural-language questions a user would ask an AI assistant (not the terse keywords they'd type into Google). Generate 25 conversational prompts across the funnel — definitional, comparative, troubleshooting, and recommendation-seeking. For each, note the answer format an AI engine would prefer (a list, a table, a direct one-sentence answer) so I can structure my page to be quotable. Output grouped by funnel stage. Save to geo-queries.md.

What slides.md looks like

## GEO query expansion — "choosing an MCP server"

### Recommendation-seeking
- "What's the best MCP server for SEO work?" → Answer format: ranked list
- "Should I use an MCP server or build my own integration?" → direct answer + table

### Troubleshooting
- "Why won't my MCP server connect in Claude Code?" → numbered steps

25 prompts · 8 definitional · 6 comparative · 6 troubleshooting · 5 recommendation.

One-line tweak

Add 'and draft the one-sentence answer an AI engine could lift for each' to pre-write your quotable snippets.

Pairs with

geo-content-optimizer

Optimize the page so answer engines actually quote it.

ai-seo

Round out the GEO play with AI-visibility tactics beyond keywords.

Content calendar from a cluster

Turn a set of keyword clusters into a sequenced publishing calendar — what to write first, what depends on what, and a realistic cadence.

ForSolo founders and small content teams who need an order, not just a backlog.

The prompt

Use the keyword-research skill. Read briefs/ (the cluster briefs from earlier). Build a 12-week content calendar. Sequence by a blend of opportunity score and topical dependency — pillar pages before the supporting posts that link up to them. For each week, list the article title, its primary keyword, the cluster it belongs to, and which earlier piece it should internally link back to. Flag any week that has no internal-link target yet as a dependency risk. Output as a week-by-week table. Save to content-calendar.md.

What slides.md looks like

## 12-week content calendar

| Week | Article | Primary keyword | Cluster | Links back to |
|------|---------|-----------------|---------|---------------|
| 1 | What are Claude skills? | claude code skills | Skills (pillar) | — |
| 2 | Keyword research skill guide | keyword research skill | Skills | Week 1 |
| 3 | Best SEO MCP servers | best seo mcp | MCP/SEO (pillar) | Week 1 |
| 4 | SEO audit skill guide | seo audit skill | Skills | Week 2 |

Dependency risk: none — every week after 1 has an internal-link target.

One-line tweak

Change '12-week' to 'two posts a week for 6 weeks' and the calendar re-paces without losing the dependency order.

Pairs with

content-strategy

Layer goals and KPIs onto the raw publishing schedule.

content-trend-researcher

Slot timely, trending topics into open calendar weeks.

Title and meta generation per cluster

Generate intent-matched title tags and meta descriptions for every keyword in a cluster, respecting length limits and keeping the primary term near the front.

ForAnyone shipping a batch of pages who refuses to hand-write 30 meta descriptions.

The prompt

Use the keyword-research skill. For the 'Claude skills for SEO' cluster, write a title tag (<= 60 chars, primary keyword front-loaded, intent-matched) and meta description (<= 155 chars, one benefit + one reason to click) for each of the 6 keywords. Do not keyword-stuff — each meta should read like a human wrote it. Flag any title that collides with another in the cluster so I don't create duplicate-title pages. Output as a table: Keyword, Title (char count), Meta (char count). Save to titles-meta.md.

What slides.md looks like

## Titles + meta — "Claude skills for SEO"

| Keyword | Title (chars) | Meta (chars) |
|---------|---------------|--------------|
| claude code skills | What Are Claude Code Skills? A Plain Guide (43) | The 60-second explainer... (132) |
| keyword research skill | Claude Keyword Research Skill: 10 Workflows (44) | Run seed expansion... (141) |
| seo audit skill claude | Claude SEO Audit Skill Guide (2026) (35) | 10 prompts that audit... (138) |

No title collisions in this cluster.

One-line tweak

Add 'and generate two title variants per keyword for A/B testing' to feed a CTR experiment.

Pairs with

seo-meta-optimizer

Refine and A/B-test the generated titles against live CTR.

brand-voice

Keep every meta description on-brand across the batch.

Internal-link map from clusters

Derive a concrete internal-linking plan from the cluster structure — which page links to which, with the anchor text to use, so topical authority actually flows.

ForSEOs who built the content but never wired the internal links that make it rank.

The prompt

Use the keyword-research skill. Read briefs/ and content-calendar.md. Build an internal-link map: for each planned page, list which other pages it should link TO and which should link back to IT, with a suggested anchor (the supporting keyword, not 'click here'). Enforce a hub-and-spoke shape — pillar pages are hubs, cluster posts are spokes that link up and sideways to siblings. Flag any orphan page with no inbound internal links. Output an adjacency table: Source page, Target page, Anchor text. Save to internal-links.md.

What slides.md looks like

## Internal-link map

| Source page | Target page | Anchor text |
|-------------|-------------|-------------|
| Keyword research skill guide | What are Claude skills? | what a Claude skill is |
| Keyword research skill guide | SEO audit skill guide | run an SEO audit |
| What are Claude skills? (hub) | Keyword research skill guide | keyword research skill |

Orphans: none. Hub 'What are Claude skills?' has 6 inbound spokes.

One-line tweak

Add 'and limit each page to 4 outbound internal links to the most relevant siblings' to avoid diluting link equity.

Pairs with

seo-research

Audit the live internal-link graph to compare plan against reality.

programmatic-seo

Auto-generate the link map when the cluster is a templated page set.

Community signal

Three voices on the parts of this workflow that matter most. The first two are on intent as the qualifying filter — the principle the skill bakes in by tagging intent on expansion — and the third is on why AI ideation needs real data behind it.

“If the intent doesn't align with something your site can realistically serve and convert, the keyword doesn't belong in your plan, regardless of how attractive the volume looks.”

Despina Gavoyannis (Ahrefs) · Blog

Ahrefs' keyword-intent guide arguing intent is a hard qualifying filter that overrides raw search volume — the principle behind use cases 1 and 4.

“Keyword intent is your first filter, not your last.”

Despina Gavoyannis (Ahrefs) · Blog

The closing takeaway of the same piece — intent classification belongs at the start of the workflow, which is why the skill tags intent on expansion.

“Paired with traditional keyword research tools, ChatGPT for keyword research can be a powerful tool in your arsenal.”

Daniel Smullen (Search Engine Journal) · Blog

SEJ framing AI as an augment to data-backed tools, not a replacement — the same reason every use case here pairs the skill with a data MCP.

The contrarian take

Not everyone thinks keyword research should start with keywords at all. The sharpest version of that critique is from Lazhar Ichir (Topicseed):

“Focusing on search volumes makes you lose sight of the larger picture: topical authority.”

Lazhar Ichir (Topicseed) · Blog

From Topicseed's argument that volume-first keyword research undersells topical authority.