Skip to content

KNOWLEDGE_BASE.md

Latest commit

 

History

History
87 lines (65 loc) · 6.64 KB

KNOWLEDGE_BASE.md

File metadata and controls

87 lines (65 loc) · 6.64 KB
Raw

seo-superpower — Knowledge Base

Generated: 2026-05-22 (night-shift R2 enrichment, 2 sub-agents) Status: Public Claude Code plugin (github.com/benskamps/seo-superpower), v0.3.0 shipping v3 roadmap. This KB grounds v3 launch + post-launch direction. Depth in docs/enrichment-2026-05-22/.

Shards

Shard Subdomain Sources
geo-landscape.md GEO landscape 2026 — AI search citation, AIO/AI Mode state, agentic search, llms.txt verdict 33 sources, Princeton paper A-tier
plugin-ecosystem.md CC plugin distribution, marketplace dynamics, competitor positioning, release mechanics 10 sources

Cross-shard load-bearing facts

GEO landscape (these reshape v3's substance):

  1. 🚨 Reddit overtook Wikipedia as #1 AI citation source (40.1% vs 26.3% per Jun 2025 cross-platform study; Reddit citation share grew 73% Oct 2025 → Jan 2026). The "47.9% Wikipedia" framing in existing SOURCES.md is stale.
  2. Google launched "Community Perspectives" in AI Overviews on 2026-05-07 — pulls Reddit/forum quotes inline. Plus 4 other AIO/AI-Mode updates the same week. This is fresh material the April enrichment missed.
  3. 🔑 Agentic search is the May 2026 doctrine shift. Mike King's patent analysis + iPullRank essays argue every major engine now does query fan-out (6-8 synthetic queries per user query). Implication: optimize for the planner's follow-ups, not just the surface query. Recommended new skill/MCP: query-fan-out simulator (Gemini-API-backed, free tier; Qforia is iPullRank's reference).
  4. llms.txt verdict tightened to "skip unless you're a docs site." SERanking 300k-domain study: no measurable citation lift. 500M AI-bot-visit analysis: only 408 requested llms.txt. Don't ship an llms.txt generator skill.
  5. AIO CTR rebounded (Seer: 1.76% → 0.61% → 2.4% by Feb 2026) and AI-referred conversion is 14.2% vs 2.8% organic (5.1× advantage). The "AIO killed organic" panic was transient.
  6. Entity graph (Wikidata QID + Crunchbase + Schema sameAs chain) is the prerequisite layer for citation likelihood — 3-6 month establishment timeline. Recommended new skill: establishing-entity-presence.
  7. LinkedIn became ChatGPT's 5th-most-cited source Dec 2025 → Feb 2026. Sweet spot: 500-2000 word articles. Missing B2B track.

Plugin ecosystem (these reshape v3's launch):

  1. 🚨 Version coherence is broken right now: plugin.json says 0.3.1, VISION.md says "v3 SHIPPED ✅". Marketplace users see the literal version field. Either bump to 1.0.0 (since v3-the-roadmap shipped) or rename internal phases. Highest-leverage cleanup before v3 launch.
  2. Two real competitors: claude-seo (AgriciDaniel, 25 skills + 18 agents, requires paid DataForSEO) and SearchFit SEO (11 skills + 3 agents, 6,096 installs, SaaS). seo-superpower's wedge is real and defensible: free-tier + open-source + dev-first kickoff.
  3. Official marketplace UI now shows context-cost + last-updated + "will install" preview (CC v2.1.143+). 13 skills × ~100 tokens of frontmatter = ~1,300 always-on. Audit descriptions; lazy-load in skill bodies.
  4. No seo / marketing / content category exists in the official enum. Best fit is productivity.
  5. Install counts ARE publicly visible (feature-dev: 89,000+; SearchFit: 6,096). Track it as a competitive metric.
  6. Three official Anthropic marketplaces: claude-plugins-official (editorial), claude-plugins-community (automated validation, in-app submission form), claude-code-plugins (demo). Target: community marketplace.
  7. Free distribution surfaces: add claude-code-plugin GitHub topic; PR to ComposioHQ/awesome-claude-plugins; PR to quemsah/awesome-claude-plugins (automated indexer).
  8. SessionStart hook printing a kickoff message is the single highest-ROI plugin addition for v3 onboarding.

v3 release checklist (compressed from plugin-ecosystem shard §11)

Pre-release hygiene:

  • Reconcile plugin.json version with VISION.md. Pick: bump to 1.0.0 OR rename phases internally.
  • Audit all 13 skill descriptions — slim frontmatter, push detail into bodies.
  • Category: set to productivity (until SEO/marketing enum exists).

Discoverability:

  • Add claude-code-plugin GitHub topic.
  • PR to awesome-claude-plugins indexes.
  • Submit to claude-plugins-community marketplace.

Onboarding:

  • Add SessionStart hook with kickoff message.
  • README: "smallest first win" demo above the fold.
  • Demo GIF or short video.

Content / GEO substance:

  • Update SOURCES.md with the Reddit-overtook-Wikipedia number.
  • Add query-fan-out simulator skill/MCP (or document as v3.1).
  • Add establishing-entity-presence skill (Wikidata + Crunchbase + sameAs chain).
  • Drop or deprecate any llms.txt skill — net-negative recommendation.
  • Add LinkedIn B2B track (500-2000 word articles, ChatGPT #5 source).

Skill audit (from shard §10 of geo-landscape):

  • For each of the 13 existing skills, mark: keep / edit / deprecate. Rationale per skill is in the shard.

Recommended next moves

Land before v3 ships:

  1. Fix the version coherence (plugin.json vs VISION.md). Single highest-impact cleanup.
  2. Audit SOURCES.md and refresh with the Reddit + Community Perspectives + agentic-search facts.
  3. Add the SessionStart kickoff hook.

Land with v3: 4. Two new skills: query-fan-out and establishing-entity-presence. 5. Deprecate any llms.txt generator skill (or mark as docs-sites-only). 6. Add LinkedIn B2B skill.

Post-launch: 7. Discoverability blast: GitHub topic + awesome-list PRs + community-marketplace submission. 8. Track install count + AI citations to scarecares.com / vibecrafting.ai (Ben's preferred public examples).

Tensions to honor

  • Free vs paid wedge: claude-seo and SearchFit both lean on paid API tiers (DataForSEO, etc.). seo-superpower's free-tier + open-source posture IS the moat. Resist any "v3 requires API key" creep.
  • Comprehensive vs focused: 13 skills is already in the heavy zone. Adding 2-3 (fan-out, entity, LinkedIn) is fine; adding 8 more is bloat.
  • Trend-chasing vs canonical: GEO is fluid. v3 should encode the May 2026 doctrine; v3.x should reserve room for the next pivot (agentic-everything seems inevitable).

Provenance

Night-shift R2 enrichment 2026-05-22. Sub-agents: 2 × general-purpose. 43 unique sources between them, including the Princeton GEO paper, iPullRank, Profound, Mike King, Aleyda Solis. IN-USE.md in place; skills/, agents/, MCP server code, .claude-plugin/ untouched.