SEO Keyword Niche

Example output: examples/seo-keyword-niche-espresso-machine-reviews-20260514/KEYWORD-NICHE-PLAN.md

Keyword Niche

Mine the long tail for content opportunities. Pulls longtail variants, question-intent keywords, related keywords, and SERP samples to find under-served niche keywords; clusters them by intent; proposes a content tier (template + URL pattern + sample pages + thin-content quality gates) that can be published programmatically or as a steady editorial cadence.

Prerequisites

• DataForSEO MCP server connected.
• User provides: a seed topic (e.g. "running shoes", "tax preparation"), or 2–5 seed keywords. Target country (default us). Optional: minimum volume threshold (default: 50/mo for niche skill — lower than seo-keyword-cluster's 100), maximum KD (default: 40 for accessibility).

Process

1. Validate & preflight

   • Confirm seeds make sense (not too broad, not branded, not single-letter).
   • DataForSEO bills per call; this run issues ~N calls. Use the documented limit/filters params to cap. This skill is heavier than most: pulling longtail at depth issues more calls than pulling head terms.

2. Longtail expansion mcp__dataforseo__dataforseo_labs_google_keyword_ideas

   • For each seed: pull longtail variants (typically 3+ words, lower individual volume, lower KD).
   • Target: 200–500 longtail candidates per seed.

3. Question expansion mcp__dataforseo__dataforseo_labs_google_keyword_ideas (filter to interrogatives: who/what/why/how/can/does/is/are…)

   • For each seed: pull question-phrased keywords.
   • These are gold for content mining — explicit user intent in the keyword.

4. Related expansion mcp__dataforseo__dataforseo_labs_google_related_keywords and mcp__dataforseo__dataforseo_labs_google_keyword_suggestions

   • For each seed: pull related + similar keywords.
   • Catches semantic neighbours that longtail expansion missed.

5. Filter and clean

   • Remove keywords below min_volume and above max_kd.
   • Strip branded keywords the seed-domain doesn't own (unless the user supplied a branded seed).
   • Tag each keyword with detected intent: informational, commercial, transactional, navigational.
   • De-duplicate across seeds.

6. SERP sample for representative keywords mcp__dataforseo__serp_organic_live_advanced

   • For 5–10 representative keywords (one per emerging cluster), pull top 10.
   • Identify the dominant page type for each cluster (informs the template proposal in step 8).

7. Cluster by intent + theme

   • Group keywords by semantic similarity + intent.
   • Each cluster: name, primary keyword, supporting keywords, total volume, weighted KD, dominant page type.
   • Target 8–20 clusters (more granular than seo-keyword-cluster's 5–12 — this skill is for niche tiers).

8. Propose a content tier

   • Template structure: which fields make each page unique. For {city} apartment rentals, the template fields are {city}, {neighborhood}, {price_range}, {property_count}. For {tool} alternatives it's {tool}, {competitor_count}, {primary_use_case}.
   • URL pattern: /{category}/{slug}/ or /{topic}/{tool}-alternatives/ etc.
   • Sample pages: 3 fully-spec'd page wireframes for representative cluster keywords.
   • Internal-linking automation: how to link pages within the tier (hub-spoke from a category page, or peer-to-peer for genuinely flat structures).

9. Quality gates (anti-thin-content guardrails)

   • Minimum unique data per page: template fields must produce ≥ 5 unique attributes (counts, prices, names, dates, etc.). Pages without 5+ unique attributes are duplicates in disguise — skip those keyword variants.
   • Minimum word count: 600 words effective content (excludes navigation, footer, boilerplate).
   • Schema requirement: every templated page gets Article (or relevant type) + BreadcrumbList.
   • Index/noindex split: if the variant doesn't pass the unique-data threshold, generate the page but noindex it.

9a. Programmatic publishing — extra gates

Most users discover this skill because they want to ship pages at scale (programmatic SEO, data-driven directories, location pages). Programmatic publishing is the highest-risk path for spam-classifier blowback. Apply these in addition to the gates above whenever the proposed tier expects to ship 50+ templated pages.

• Per-row uniqueness threshold (≥ 30% varying fields). Of the template's content-producing fields (excluding nav/footer boilerplate), at least 30% must hold values that differ from the median sibling page. A 12-field template where 9 fields are identical across pages is templated mush — Google's spam systems are well-tuned for this since the 2024 core update. Compute: varying_fields / content_fields ≥ 0.30 per row vs the cluster median.
• Min unique-fact count vs parent + sibling (≥ 5 facts). Each row carries at least 5 facts that do not appear on the parent hub page or any sibling page in the same cluster. Facts = numbers, dates, named entities, original quotes, photos. Synonym shuffles don't count. Sample 10 rows manually before greenlighting the tier.
• Data-source independence. Don't auto-publish from a single source (one CSV, one API, one scrape). If the page's only differentiator is a row from cities.csv, the page is a CSV row dressed as content — likely thin. Combine ≥ 2 independent data sources per page.
• Index-bloat circuit-breaker. After the first 50 pages ship, monitor GSC. If the index-coverage rate drops below 60% (Google indexed <60% of submitted), pause the tier and re-audit. Continuing past this signal compounds bloat across the rest of the tier.
• Crawl-budget honesty. Sites under 50k pages can usually crawl whatever you ship. Sites >50k must factor in crawl-budget cost: every thin programmatic page steals attention from the cornerstone content. If the site is in this band and the unique-fact count is borderline, default to noindex.

These gates are not negotiable for the programmatic path. If the proposed tier can't pass them, the right answer is fewer pages, not lower thresholds.

10. Synthesise KEYWORD-NICHE-PLAN.md

Output format

Create a folder seo-keyword-niche-{target-slug}-{YYYYMMDD}/ with:

seo-keyword-niche-{target-slug}-{YYYYMMDD}/
├── KEYWORD-NICHE-PLAN.md           (synthesised plan — primary deliverable)
├── keywords.csv                    (all enriched keywords with cluster + intent — load-bearing CSV the publishing team paste into CMS/sheets)
├── 06-template-spec.md             (fields, URL pattern, sample pages — load-bearing reference writers consult directly)
├── 07-quality-gates.md             (thin-content guardrails — load-bearing reference release-gate readers consult directly)
└── evidence/
    ├── 01-seed-expansion.md        (raw expansion per seed — raw step output)
    ├── 02-question-keywords.md     (question-keyword expansion)
    ├── 03-filtered-keywords.md     (post min-vol / max-kd filter)
    ├── 04-cluster-assignment.md    (every keyword and its cluster)
    └── 05-serp-samples.md          (top 10 for representative cluster keywords)

Top-level: KEYWORD-NICHE-PLAN.md + keywords.csv + 06-template-spec.md + 07-quality-gates.md. Writers consult the template spec directly when authoring; release gates consult the quality gates directly. The 01–05 step files preserve raw API/clustering outputs in evidence/.

KEYWORD-NICHE-PLAN.md shape:

# Keyword Niche Plan: {topic}

> Generated {YYYY-MM-DD} · Country: {country} · Seeds: {list}

## Inventory
- Longtail keywords mined: {n}
- Question keywords: {n}
- After filter (min-vol {n}, max-kd {n}): {n}
- Clusters formed: {n}
- Estimated combined monthly volume: {n}

## Recommended content tier

### Template
**URL pattern:** `/{category}/{slug}/`

**Required fields per page:**
- `{slug}` — URL-friendly identifier
- `{H1}` — page primary heading
- `{TL;DR}` — first 200 words direct answer
- `{primary_keyword}` — target keyword
- `{related_keywords}` — secondary keywords from cluster
- `{unique_attributes}` — list of ≥ 5 differentiating attributes (numbers, names, dates)
- `{related_pages}` — internal-link list (3–5 sibling pages)
- `{schema_type}` — Article/Product/Other
- `{datestamp}` — last-updated date

**Sample pages:** see `06-template-spec.md` for 3 fully-spec'd wireframes.

## Cluster build order (top 10 by priority)

| Rank | Cluster | Volume (combined) | Weighted KD | Dominant page type | Pages to ship |
|---|---|---|---|---|---|
| 1 | {cluster name} | {n} | {kd} | {type} | {n} |
| ... |

## Quality gates (do not ship pages that fail)

1. **Unique-data threshold:** each page has ≥ 5 unique attributes vs sibling pages in the same cluster.
2. **Minimum word count:** 600 words effective content.
3. **Schema:** `Article` + `BreadcrumbList` (or relevant `@type` for the page).
4. **Internal links:** ≥ 3 outbound to siblings, ≥ 1 inbound from category hub.
5. **Index/noindex:** noindex pages that don't pass the unique-data threshold.

### Programmatic-only gates (apply when tier ships 50+ pages)

6. **Per-row uniqueness ≥ 30%:** at least 30% of content-producing fields hold values that differ from the cluster median.
7. **Min unique facts vs parent + sibling (≥ 5):** five facts per row that don't appear on the parent hub or any sibling.
8. **Data-source independence:** ≥ 2 independent data sources per page (no single-CSV pages).
9. **Index-bloat circuit-breaker:** pause tier if GSC index-coverage <60% after the first 50 pages.

## Scaling estimate
- Clusters: {n}
- Pages per cluster (median): {n}
- Total pages: {n}
- At {pages/week} cadence: {n weeks} to ship the tier.
- Crawl-budget impact (for sites > 10k pages): noindex strategy keeps thin variants out of the index.

## Risks / monitoring
- **Thin-content penalty:** the quality gates above are the guardrail. Audit at scale via `seo-technical-audit` after first 100 pages ship.
- **Index bloat:** monitor in GSC; if newly indexed pages don't accrue impressions in 60 days, candidate for noindex/consolidate.
- **Cannibalization:** monitor with `seo-subdomain` + `seo-page` after first 50 pages.

## Recommended next step
Build a small pilot — 10 pages from the top cluster — before committing to the full tier. Apply quality gates rigorously to the pilot. Re-audit after 60 days; iterate the template based on which pilot pages indexed and ranked.

keywords.csv columns: keyword,volume,kd,cpc,intent,cluster,role_in_cluster,dominant_page_type,unique_attributes_estimate

Tips

• DataForSEO allows up to 2,000 calls/min, 30 concurrent. Skills pace sequentially. Steps 2–4 fan out across all seeds.
• DataForSEO bills per call; this run issues ~N calls (roughly several per seed across steps 2–6). Use the documented limit/filters params to cap. Large dataforseo_labs_* lists can exceed the MCP inline limit — cap with limit and filters; slice saved files with jq.
• Programmatic SEO is risky. Pages that don't pass the unique-data threshold are dead weight at best, penalty bait at worst. The quality gates in step 9 are not optional.
• The pilot recommendation is critical. 10 pages with thoughtful templates outperform 1000 pages of templated mush.
• For e-commerce / inventory-driven content (city pages, product variations), the unique-data threshold is usually easy to hit.
• For "best X for Y" content (where Y has many variants), pick variants that have genuinely distinct content — don't templatise across near-synonyms.
• Pair with seo-keyword-cluster for the broader pillar+spoke architecture (this skill complements, not replaces).
• Pair with seo-content-brief to expand individual pillar topics into full editor briefs.
• Pair with seo-technical-audit after first 100 pages ship to catch any thin-content / canonicalization issues.
• Don't auto-publish. The template, URL pattern, and quality gates ship from this skill; the content authoring + review remains a human responsibility.