citation-intel

name	citation-intel
description	Agentic citation intelligence and AI-search optimization for long-form publication. Trigger on /citation-intel, "is this citation ready", "what schema for", "optimize for citation", or any request to check discoverability / schema / token efficiency / citation readiness before publication. CENTROID: article-specific citation intelligence — what makes AI agents cite this article in 2026, not generic SEO IS: - extractable-claim check (is the primary claim in the first 500 tokens?) - entity-density scan (proper nouns + concrete stats — agents cite 3× more) - FAQPage schema generator (AI cites FAQ content heavily) - community-language alignment (Perplexity cites Reddit at 24%) - token-efficiency audit (posts <10k, articles <20k, centroid <15k) - site-level AI-access stack audit (robots.txt allowlist, llms.txt) IS NOT: collapse_zones: - prestige SEO advice (keyword-stuffed listicles are tier-3, skipped entirely) - generic content marketing (must be article-specific, never templated) - llms.txt evangelism (llms.txt is one signal, not the whole game) - "add FAQ schema" boilerplate (FAQ must be article-derived, no new claims) - publication action (citation-intel checks readiness; publish is downstream) sibling_overlaps: - /tom (both pre-publication; tom re-expresses, citation-intel optimizes for citation) - /complement (both check for missing pieces; complement on structure, citation-intel on discoverability) WHEN: - before publishing long-form to AI-searchable surfaces (blog, docs, paper) - when an article should be cited by Perplexity / ChatGPT / Claude in response to topic queries - when the sidecar router injects a `citation_ready_due` lane directive into additionalContext - on site-level audit request (robots.txt / llms.txt / AI crawler allowlist) NOT WHEN: - for internal-only documents (no citation surface exists) - for short-form (tweet / chat / single-screen — token efficiency moot) - for non-AI-searchable formats (PDF behind login, internal wiki)
user-invocable	true

/citation-intel — Citation-Ready Audit

Article-specific citation intelligence, not generic SEO. What actually moves AI citation in 2026:

• Extractable claim in first 500 tokens — agents skip preamble
• Entity density — proper nouns + concrete stats cited 3× more
• Synthesis content — centroid articles are citation gold
• FAQPage schema — AI cites FAQ content heavily
• Community language alignment — Perplexity cites Reddit at 24%
• Token efficiency — posts <10k, articles <20k, centroid <15k
• robots.txt AI allowlist — 96% of top domains haven't done this

Inputs

slug:          [identifier]
article-type:  BIZ | INFRA | CENTROID
headline:      [plain text]
lede:          [plain text]
draft-body:    [full draft]
skip-research: false  # set true to return schema only (quick mode)

Outputs

status:          ok | blocked
blocked-reason:  null | [string]

primary-claim:   "[single extractable sentence]"
claim-in-500:    true | false
token-estimate:  [integer]
entities:        [list of proper nouns + concrete stats]
community-lang:  [2-4 phrases the community uses for this topic]

faq:
  - q: "[question]"
    a: "[answer — from article body, not new claims]"

schema-type:     BlogPosting | Article
schema-faq-block: |
  [FAQPage JSON-LD string, ready to insert]
schema-centroid-additions: |
  [mentions + about additions for CENTROID articles]

citation-tier1:  [primary source URLs]
citation-tier2:  [Reddit / HN URLs for community context]
flags:           [non-blocking notes]

Six-Step Research Loop

Run all 6 unless skip-research: true (then run steps 1+6 only).

Step 1 — Claim extraction Identify the single most extractable sentence. Token-estimate the draft (words ÷ 0.75). Is the claim in first 500 tokens? If not: flag (not block).

Step 2 — Entity scan Extract all proper nouns, named systems, concrete statistics. Minimum 3 entities. If fewer: flag — article may be too abstract for citation.

Step 3 — Category search

[topic] [voice] 2026 site:reddit.com OR site:news.ycombinator.com
[topic] [article-type context] 2026 research

Target: 2 tier-2 sources (Reddit / HN) + 1 tier-1 (primary data). Skip prestige SEO blogs — they are tier-3.

Step 4 — Competitive scan

"[primary claim keywords]" thought leadership 2026

Is the claim already saturated? If yes: sharpen, don't soften. If unique: note as citation advantage.

Step 5 — FAQ generation Generate 2-3 questions a real person would search that this article answers. Rules: questions must be answerable from article body, answers must contain the primary claim, combined FAQ text <1500 tokens, no new claims introduced.

Step 6 — Schema assignment Assign by article-type:

• BIZ → BlogPosting + FAQ
• INFRA → BlogPosting + FAQ + speakable
• CENTROID → Article + mentions + FAQ

Site-Level Audit

Check and report on:

• robots.txt — are all major AI crawlers explicitly allowed? (GPTBot, ClaudeBot, PerplexityBot, anthropic-ai, Google-Extended, OAI-SearchBot, Meta-ExternalAgent)
• llms.txt — present? current? all published articles listed?
• Headers — Content-Type for text/plain; charset=utf-8 on llms.txt?
• Internal link graph — do articles link to 3+ related pieces?
• Token efficiency — spot-check most recent 3 articles for token estimate

Returns: per-check status + one action item per failure.

Quick Reference

token-rule:  claim in first 500 tokens — always
schema:      BIZ → BlogPosting+FAQ | INFRA → BlogPosting+FAQ+speakable |
             CENTROID → Article+mentions+FAQ
site-files:  robots.txt + llms.txt + Content-Type headers
reddit:      Perplexity cites Reddit 24% — community language = citation surface
claudebot:   up 800% early 2026 — assume crawled, make it extractable