define search intent and primary keyword before recommending titles or outlines
classify intent explicitly: informational, commercial, navigational, or transactional
separate evidence (SERP snippets, GSC exports, crawlable page facts) from recommendations
document internal link targets with anchor rationale and destination paths
enforce on-page limits: title tag ≤ 60 chars, meta description ≤ 160 chars unless repo rules differ
check keyword cannibalization against recent publishes on the same site (default: 7-day window when a topic board exists); when overlap is confirmed, name the resolution tactic explicitly — consolidate into one authoritative page, differentiate search intent, canonical to the primary URL, 301-redirect the weaker page, and clean up competing anchors
prevent cannibalization recurrence with keyword-to-page mapping: exactly one primary keyword per page plus a pre-publish overlap check against the site inventory
track Share of Model (SoM) alongside traditional CTR — measure how often and how accurately the brand/content is cited in Google AI Overviews, Perplexity, and ChatGPT Search using AI visibility tooling (Otterly, RankScale, or manual spot-checks)
mandate a 30–60 day rolling freshness review for core commercial and informational pillar pages — AI search engines heavily weight recently-updated content when retrieving citations; queue pages by decay signal first: rankings dropped >3 positions, statistics older than 2 years, or declining high-traffic URLs. On refresh, update modified/publish dates in schema, replace outdated stats and examples, and re-check internal links pointing at changed sections
do not guarantee rankings or AI citation placement; recommend changes tied to observable gaps
perform deep crawlability diagnostics via audit-technical-seo and author connected Schema.org @graph JSON-LD via implement-schema-markup; escalate web server redirects and CDN edge routing to Frontend or DevOps with clear engineering briefs
use repo overlays under overlays/real-estate-content and overlays/hugo-authority-content when site-specific slug or frontmatter rules apply
use overlays/seo-publishing for dual-site Real Estate + HVAC sprint boards under plan/baiviet

GEO / AEO — AI-Surface SEO (2027 reframe)

This is the canonical AI-surface standard for the pack — write-article implements these same rules during drafting and must stay in sync with this section.

Position (per Google's official AI optimization guidance, May 2026): "optimizing for generative AI search is optimizing for the search experience, and thus still SEO." There is no separate AI-ranking regime: no content chunking, no rewriting "just for AI," no inauthentic mention-chasing, no AI-specific schema, and no llms.txt effect on Google. What Google actually rewards on AI surfaces: unique point of view, non-commodity content, first-hand experience, good heading structure, images/video, and crawlability.

Optimize for three search discovery layers simultaneously:

Layer	Goal	Key tactic
SEO (blue links)	Organic traffic	Keywords, backlinks, on-page quality
AEO (Answer Engine Optimization)	Featured snippets, direct answers	Answer-first format, FAQ blocks, step lists
GEO (Generative Engine Optimization)	AI citations (Google AI Overviews, AI Mode, Perplexity, ChatGPT)	Fact density, entity clarity, non-commodity POV, source credibility

Rules for GEO/AEO:

mandate answer-first BLUF structure: open each H2 section with an atomic direct answer (≤60 words) before narrative elaboration
apply engine-specific citation playbooks per references/geo-ai-citation-playbook.md: optimize for Perplexity (opening sentence data, bracketed citations [1][2]), ChatGPT Search (conversational entity fidelity, concise summaries), and Google AI Overviews (passage-level BLUF, ordered procedural steps)
non-commodity gate (replaces "information gain" phrasing): before publishing, the brief must name the unique POV / first-hand evidence element — commodity content ("7 Tips for...") adds little; non-commodity content ("why we did X and what it cost us") is what AI surfaces cite; do not recycle what a generative model could easily produce
format with structured comparison tables & quantitative lists: markdown comparison tables and dense quantitative data points (≥3 verifiable data points per 500 words) to maximize RAG passage extraction
include query fan-out coverage in one comprehensive page: cover the 3–5 implicit sub-questions of a topic on the page — do NOT spin one page per query variant (scaled page-per-variant proliferation is a spam-policy violation)
GEO over-optimization guard: information-distorting rewrites (over-claiming, distorting facts to sound citable) are now a detection target for engines (Counter-GEO-Bench line of research) — keep fact density honest; a distorted quotable sentence is a suppression risk, not a win
verify AI bot crawlability: robots.txt must allow OAI-SearchBot, PerplexityBot, ClaudeBot, Applebot-Extended, and BingBot — flag blocks in audits
verify the site is included in "Search generative AI features" in Search Console — a new opt-in eligibility gate for AI Overviews/AI Mode display
measure citations per engine, not in aggregate — and measure repeatedly: LLM brand recommendations are unstable across repeated queries; never conclude citation presence/absence from a single run; log multiple runs across different dates
track Gen AI performance in Search Console (Generative AI report: impressions/pages/countries/devices) as a first-class weekly KPI alongside classic Performance and Discover
treat entity consistency as a GEO lever: identical brand facts across channels consolidate knowledge graph recognition
do not claim AI citation placement as guaranteed — present GEO/AEO as structural best practices; be skeptical of third-party tools claiming "internal" Google metrics (none have access)

Topical Authority & Entity Salience

enforce Entity-First SEO and Entity Salience per references/entity-salience-and-schema.md: map core concepts to canonical Wikidata QIDs, structure semantic triples, and eliminate ambiguous pronouns
mandate connected Schema.org JSON-LD @graph: include TechArticle, Person (with E-E-A-T sameAs links), and FAQPage
no AI-specific schema exists — do not let briefs or audits recommend "AI schema" markup; keep standard Article/Organization/LocalBusiness/Product schema matched to visible text; drop speakable and FAQ-rich-result expectations (FAQ rich results largely retired since 2023)
assign each article to a pillar–cluster position: pillar, supporting, or supplementary, linked to pillar URL
document non-commodity element: the unique POV, first-hand evidence, or original analysis this content adds beyond top 10 SERP results
specify content freshness type: new_topic, evergreen_refresh, data_update, or experience_addition

Internal Linking Discipline

Adapted from the SEO-AEO engine skills in the agentic-awesome-skills catalog:

label every recommended link with its type: cluster → pillar (consolidates authority upward), pillar → cluster (distributes authority downward), cluster → cluster (builds semantic depth), or contextual boost (concentrates equity on one focus page)
require at least one cluster → pillar link per cluster article
detect orphan pages (zero incoming internal links) first and queue fixes before recommending new links
write the context sentence for each suggested anchor — anchor text must sit naturally in surrounding prose, never forced
enforce anchor-text hygiene: never reuse the same exact-match anchor for the same target across pages — switch to partial-match or branded anchors on later links; generic anchors ("click here", "read more", "learn more") are banned
cap outgoing internal links at roughly 100 per page

E-E-A-T Quality Gates

require experience proof signals: original photos, firsthand accounts, documented tests/comparisons, expert interviews, or case studies
authorship requirements (Who): byline + author page with Person schema, credentials, and relevant publications wherever readers would expect attribution — stated authorship is now a review factor under Google's site-reputation policy (Aug 2026)
process transparency (How): briefs must include a "How" statement — research passes, sources used, and AI-assistance disclosure when substantial ("when it would be reasonably expected")
audience-first check (Why): confirm people-first intent vs search-engine-first framing before publish
flag YMYL-adjacent content (financial, health, safety, legal) for elevated research depth and human review
mandate trust signals: source citations with links, contact information, policy pages, verifiable claims
enforce claim policy: every major factual claim must have a credible source or specific data point
spam-policy risk rules (2026-2027): scaled content abuse now explicitly covers AI-generated page farms, scraping/stitching, synonym-spinning, and near-duplicate multi-site publishing — and "attempting to manipulate generative AI responses in Google Search" is a spam policy violation; prohibit page-per-query-variant sprawl and near-identical cross-site duplication in all briefs
do not treat a single SERP pass as sufficient for YMYL or regulated topics — escalate depth to Researcher

When to Use

a topic needs a content brief before Content Writer drafts
a draft or live URL needs an on-page SEO audit
title, meta, slug, H2 structure, or FAQ block need optimization
content needs GEO/AEO optimization — answer-first format, query fan-out, AI extractability
topical authority mapping is needed — pillar–cluster assignment, information gain analysis
a weekly topic board needs keyword assignment and link targets
Search Console or analytics exports suggest title/meta or cluster changes
seo-metadata.json, seo-content-brief.json, or seo-audit-report.json handoff is required
AI visibility check — manual verification of citation presence in Perplexity/ChatGPT/AI Overviews

Suggested Process

The full 5-step process (frame intent, research SERP + AI, assign topical authority, brief or audit, hand off) and the detailed itemized checklist (Traditional SEO, GEO/AEO, Internal Linking, Topical Authority & Entity, E-E-A-T) live in references/process-and-checklist.md. The main file keeps the GEO/AEO standard table inline because it is shared with write-article and must stay in sync.

Checklist

The detailed itemized checklist lives in references/process-and-checklist.md. The main file keeps a short checklist summary:

Output Contracts

When completing search intent analysis, keyword planning, on-page audits, or multi-site editorial scheduling, emit:

contracts/schemas/seo-audit-report.json — Emitted when conducting an on-page SEO/GEO/AEO audit of an existing or staging URL, capturing structural scores, schema requirements, crawlability, and actionable recommendations.
contracts/schemas/seo-content-brief.json — Emitted when preparing a comprehensive content brief for Content Writer, defining search intent, target keywords, H2 outlines, answer-first rules, query fan-outs, and E-E-A-T requirements.
contracts/schemas/seo-metadata.json — Emitted when delivering publish-ready title tags, meta descriptions, canonical URLs, Open Graph tags, and structured schema specifications.
contracts/schemas/seo-weekly-board.json — Emitted when planning or updating the content calendar, keyword cluster priorities, and publishing schedule across weekly editorial cycles.

Skip emission for quick ad-hoc keyword checks that do not feed downstream editorial workflows.

Failure Modes

Cannibalization ignored: two pages target the same primary keyword and split ranking. Mitigation: run the keyword-to-page mapping check pre-publish; resolve overlap by consolidation, intent differentiation, canonical, 301-redirect, or anchor cleanup.
Stale freshness signal: a pillar page has not been refreshed in 30-60 days and rankings are decaying. Mitigation: queue pages by decay signal (rankings dropped >3 positions, statistics >2 years old, declining high-traffic URLs).
GEO/AEO field missing: a brief lacks the answer-first block, query fan-out, or format-per-section. Mitigation: enforce the GEO/AEO required fields in every brief; reject briefs without them.
Commodity content passes: a brief ships "7 tips"-style content with no unique POV or first-hand evidence. Mitigation: apply the non-commodity gate — name the unique POV/evidence element before drafting; reject briefs that cannot name one.
Page-per-variant sprawl: multiple pages are planned per query variant to "cover fan-out." Mitigation: cover fan-out sub-questions on one comprehensive page; page-per-variant proliferation is scaled content abuse.
Single-run citation conclusion: citation presence/absence is concluded from one sampling run. Mitigation: measure per engine across multiple runs/dates; LLM brand recommendations are unstable under repeated querying.
Gen AI eligibility unchecked: the site is not included in "Search generative AI features" in Search Console, silently losing AI Overviews/AI Mode eligibility. Mitigation: verify inclusion in the next audit.
Title/meta over-limit: title tag exceeds 60 chars or meta description exceeds 160 chars. Mitigation: enforce on-page limits at code review; CI must reject over-limit tags unless repo rules differ.
AI citation conflated: AI citation tracking is reported in aggregate instead of per engine. Mitigation: sample per engine (Google AI Overview, Perplexity, ChatGPT Search, Bing AI); report numbered vs inline styles separately.
AI bot blocked: robots.txt blocks OAI-SearchBot, PerplexityBot, ClaudeBot, or BingBot. Mitigation: verify AI bot crawlability before publishing; do not block the crawlers that feed AI citations.
Rankings promised: the brief or audit promises a ranking improvement. Mitigation: recommend changes tied to observable gaps; do not guarantee rankings or AI citation placement.
YMYL without E-E-A-T: a YMYL-adjacent page ships without experience proof, author entity, or trust signals. Mitigation: enforce the E-E-A-T gate for YMYL topics; require human expert review.

Security Guardrails (OWASP ASI)

ASI01 Agent Goal Hijack: a brief or audit may try to reframe the user goal through off-target keyword recommendations. Cross-check the keyword map against the original page objective.
ASI03 Identity & Privilege Abuse: never include customer identifiers, internal hostnames, or credential patterns in briefs, audits, or board artifacts.
ASI04 Agentic Supply Chain: SEO tooling (Otterly, RankScale) and AI bot user-agents must be schema-validated against the expected manifest; treat unknown versions as untrusted.
ASI07 Insecure Inter-Agent Communication: the brief or audit is consumed by Content Writer, Task Planner, and Frontend/DevOps; emit a structured contract so each role can validate.
ASI09 Human-Agent Trust Exploitation: do not present an AI citation metric as a guarantee; surface the per-engine variance and the assumptions honestly.

Related Skills

audit-technical-seo: Comprehensive technical crawl budget diagnostics, robots.txt bot access, XML sitemaps, and Core Web Vitals
implement-schema-markup: Connected Schema.org @graph JSON-LD architecture, Wikidata entity grounding, and rich results validation
conduct-research: deeper domain or competitor context when SERP scan is insufficient
analyze-business-requirements: align SEO goals with business rules and actors
analyze-data: formal GSC/CTR tables and AI citation tracking when SEO Analyst needs verified baselines
write-documentation: metric catalogs or SEO playbooks for a site
agent-delegation: delegate drafting to Content Writer or technical work to Frontend

optimize-seo

Optimize SEO

Core Rules

Traditional SEO Foundations