/seo-keyword-niche

name: seo-keyword-niche
description: Mine longtail keywords + question keywords for a topic at scale to surface niche content opportunities. Outputs a content tier plan (template + URL pattern + sample pages + quality gates) suitable for programmatic publishing or steady editorial output. Distinct from `seo-keyword-cluster` (which plans pillar+spoke architecture from broad seeds) — this skill goes deeper into the long tail. Use when the user asks "longtail keywords", "question keywords", "niche content", "content opportunities at scale", "programmatic SEO", "content mining", or "what should I write about for {topic}".

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

• SE Ranking 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).
• `DATA_getCreditBalance` — surface remaining credits. This skill is heavier than most: pulling longtail at depth costs more than pulling head terms.

2. **Longtail expansion** `DATA_getLongTailKeywords`

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

3. **Question expansion** `DATA_getKeywordQuestions`

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

4. **Related expansion** `DATA_getRelatedKeywords` and `DATA_getSimilarKeywords`

• 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** `DATA_getSerpResults`

• 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-cover