content-gap-analysis

Layer 1b of the keyword research pipeline. Finds keyword opportunities by comparing the brand's blog against competitors AND by expanding seeds + modifiers via Semrush (phrase_fullsearch / phrase_rela

SKILL.md

GitHub

Content Gap Analysis Skill

DATA LAYER RULING (2026-06-12). Every mcp__semrush__<kebab-name> tool named below is FICTIONAL (never existed on the live server), and DataForSEO is retired. Before making any data call, read ../keyword-research-pipeline/references/semrush-data-layer.md — it maps this layer's old calls to the real Semrush MCP pattern (get_report_schema -> execute_report). The logic below (filters, thresholds, output schema) remains binding; only the data calls changed.

Use Semrush's multi-mode Keyword Gap and Keyword Magic Tool to surface keywords competitors rank for that the brand doesn't, plus seed-modifier expansion of the brand's own keyword universe. The output feeds /keyword-prioritization, which then feeds /blog-pipeline for the chosen keywords.

Threshold reminder. All KD% thresholds in this skill are recalibrated for Semrush per .claude/skills/keyword-research-pipeline/references/semrush-metric-translation.md. Semrush KD% is materially stricter than Ahrefs KD — do not transplant Ahrefs thresholds. Read the metric-translation doc before tuning any number here.

Input

/content-gap-analysis <competitor-domain> [<competitor-domain> ...] [--our-domain <domain>]

Examples:

  • /content-gap-analysis competitor1.com competitor2.com
  • /content-gap-analysis competitor1.com --our-domain mybrand.com

If --our-domain isn't provided, read it from brand-config.md.

Process

  1. Parse input. Extract competitor domains (CLI args or auto-discover) and read the brand domain from brand-config.md.

  2. Resolve competitors. In order of preference:

    • CLI args (e.g. /content-gap-analysis competitor1.com competitor2.com)
    • brand-config.md competitor list
    • Autonomous fallback: call mcp__semrush__organic-competitors for the brand domain, take top 3 by organic-keyword intersection
    • Cache the resolved competitors to content-pipeline/0-keywords/cache/competitors.json so Layer 1c (/keyword-aio-gap) reuses the same set without re-querying
  3. Read brand context. Audience, products — used downstream to filter the gap list to relevant intent (Layer 2's BID).

  4. Run multi-mode Keyword Gap analysis via mcp__semrush__keyword-gap. Semrush's Keyword Gap exposes five modes; Layer 1b uses all five and tags every returned row with the originating gap_mode column. Pass to each call:

    • our_domain = brand domain
    • competitors = 1–3 competitor domains (Semrush accepts up to 4 in addition to the seed)
    • country = US (or read from brand-config.md if specified)
    • min_volume = 20 (default)
    • max_kd_percent = 70 (Semrush KD% — equivalent to legacy Ahrefs KD ≤ 60 per the metric-translation doc)
    • Exclude branded terms for competitors (use the negative-keyword filter on competitor brand strings)

    Phase 4b — multi-mode Keyword Gap

    Run the gap report once per mode and concatenate the rows. Tag the originating mode on each row:

    gap_mode What it means What to do with it
    missing Competitors rank, brand doesn't (the classic content gap) Default candidates — feed straight into the merged pool
    weak Brand ranks 11+, competitors top-10 Small-effort wins — feed into the pool, prioritization will boost
    unique Only one competitor ranks (single point of failure to displace) High-leverage — feed into the pool, prioritization will boost
    common Everyone ranks (saturated SERPs) Feed into the pool with a penalty applied at Layer 5 — keep for completeness, deprioritize
    strong Brand ranks top-3, competitors don't (already won) Do NOT feed into the writing queue. Route to content-pipeline/0-keywords/cache/strong-positions.csv for tracking only

    Run order: missing → weak → unique → common. Skip strong mode for the writing pool; run it separately and write its rows to cache/strong-positions.csv with the same column shape so downstream tooling can read them. Cap each mode at 200 rows sorted by traffic_potential descending.

  5. Pull the gap keyword list. Aim for ~400–800 raw rows across the four pool-eligible modes (missing/weak/unique/common). Tag each with source=competitor_gap and the corresponding gap_mode.

  6. Auto-relax filters if pool is small. If fewer than 50 candidates come back at the default filters, automatically re-run once with relaxed thresholds: volume ≥ 5, KD% ≤ 85 (Semrush — equivalent to legacy Ahrefs KD ≤ 80). Log the relaxation. If still under 50, continue with what we have — Layer 1c (AIO gap) and Layer 1a-driven seed expansion will widen the pool.

  7. Seed-modifier expansion (Layer 1a integration). If content-pipeline/0-keywords/seeds.json exists:

    • For each seed, call mcp__semrush__keyword-magic-broad with the seed as the keyword and the modifiers list as an include filter (terms include any of the listed strings). Use mcp__semrush__keyword-magic-phrase and mcp__semrush__keyword-magic-related for breadth where the broad search is thin; use mcp__semrush__keyword-magic-exact only when a modifier is a multi-word phrase that needs exact-match expansion.
    • Pull volume, KD%, traffic_potential, first_keyword_group (or KSB cluster id when available), and the intents array from mcp__semrush__keyword-overview for each result that lacks an intent label.
    • Tag each row with source=seed_modifier and gap_mode=seed_modifier (sentinel value — not one of the five Semrush gap modes).
    • Cap per-seed expansion at 100 results to keep the merged pool manageable.
  8. Merge and dedupe. Combine competitor-gap rows with seed-modifier rows. Keep one row per unique keyword:

    • If a keyword appears in both a competitor-gap mode and seed-modifier expansion, set source=both and retain BOTH gap_mode (from the competitor-gap row) and competitor_top_position from the gap row, plus the seed_modifier metadata.
    • If a keyword appears in multiple competitor-gap modes (e.g. it shows up in both missing and weak because two different competitors are scored differently), keep the row with the higher-priority gap_mode. Priority order: unique > weak > missing > common.
    • Otherwise keep the source-specific row as-is.
  9. Add columns the downstream layers need:

    • keyword (from Semrush)
    • volume (monthly searches)
    • kd_percent (Semrush Keyword Difficulty %)
    • traffic_potential (estimated traffic of the #1-ranking page from mcp__semrush__keyword-overview or domain-organic-pages of the top result)
    • competitor_top_position (the best position any competitor holds; null for seed_modifier-only rows)
    • cluster_id (Keyword Strategy Builder cluster id when available; otherwise empty — Layer 5 uses this for cluster_authority_gap boost)
    • first_keyword_group (Semrush Keyword Magic group; fallback when cluster_id is empty)
    • intents (array — Semrush's per-keyword intent classification: informational / navigational / commercial / transactional. Layer 2 uses this as the primary BID-Intent signal)
    • source (competitor_gap / seed_modifier / both / aio_gap — Layer 1c will append aio_gap rows)
    • gap_mode (missing / weak / unique / common / seed_modifier — populated for every row; strong-mode rows go to cache/strong-positions.csv, not this CSV)
    • Empty columns for priority_score, brand_fit, product_fit, notes, plus the BID/AIO/redteam columns Layers 2-4 will fill
  10. Save as CSV to content-pipeline/0-keywords/keyword-ideas.csv. Use UTF-8, headers in row 1.

  11. Print a one-line summary (autonomous mode) or suggest running /keyword-prioritization next (interactive mode). The summary breaks out row counts per gap_mode so the orchestrator can see whether the pool leans missing vs weak vs unique.

Output

content-pipeline/0-keywords/keyword-ideas.csv

A CSV with one row per gap keyword, every row tagged with a gap_mode, plus the columns listed above.

content-pipeline/0-keywords/cache/strong-positions.csv (parallel — gap_mode=strong rows; tracking only, not the writing queue).

Quality checklist

  • CSV has at least 50 rows (the gap is real)
  • Every row has a gap_mode populated (no nulls)
  • No competitor branded terms in the list (e.g. competitor product names)
  • All keywords have volume ≥ 20 (or ≥ 5 if auto-relaxed)
  • All keywords have kd_percent ≤ 70 (or ≤ 85 if auto-relaxed)
  • cluster_id or first_keyword_group populated (used by prioritization for cluster_authority_gap)
  • intents array populated for ≥ 90% of rows (Layer 2 needs it as the primary BID-Intent signal)
  • File is valid UTF-8 CSV (opens in Excel / Sheets without garbage)
  • cache/strong-positions.csv exists when gap_mode=strong rows were returned

When the gap is small

If fewer than 20 results come back, either:

  • The brand and competitors are too similar in coverage (good problem) — try different competitors
  • The filters were too tight — relax volume to ≥ 5 and KD% to ≤ 85
  • Semrush returned a partial result — re-run

Autonomous behavior

When invoked from /keyword-research-pipeline (or with BLOG_AGENT_AUTONOMOUS=1 set):

  • Auto-discover competitors via mcp__semrush__organic-competitors if neither CLI args nor brand-config.md provide any. Take top 3 by organic-keyword intersection. Cache to cache/competitors.json so Layer 1c reuses the same set.
  • Run all five gap modes and tag rows with gap_mode. Route strong to cache/strong-positions.csv; the rest go to the writing pool.
  • Auto-relax filters if pool is < 50 (volume ≥ 5, KD% ≤ 85; once only).
  • Auto-merge seed-modifier expansion if seeds.json exists.
  • No human prompt — never ask "which competitors?" or "is this enough?"; the orchestrator can't answer.

When competitors can't be auto-discovered

If mcp__semrush__organic-competitors returns empty (very rare — usually means the brand domain isn't indexed by Semrush, or it's brand-new):

  • Log to cache/competitor-discovery-failed.log
  • Fall through to seed-modifier expansion alone (which doesn't need competitors). Every row will have source=seed_modifier and gap_mode=seed_modifier.
  • Downstream layers handle the empty competitor_top_position column gracefully

Interactive mode (legacy / dev-only)

If neither CLI args nor brand-config nor auto-discovery yields competitors AND BLOG_AGENT_AUTONOMOUS is not set, fall back to asking the user. This branch only fires when a human is at the keyboard.

Tool naming

Tool names follow the placeholder convention from .claude/skills/keyword-research-pipeline/references/semrush-mcp-tool-inventory.md. After running the discovery prompt described in that file, replace any name in this skill that doesn't match the verified inventory. No mcp__ahrefs__* calls — Ahrefs is retired; an Ahrefs call from this skill is a bug, not a fallback.