/kai-topical-map — the map of pages, entities, and passages a team can start building

Objective

A topical map built for AI search and traditional search together: which entities the brand must own, which sub-queries each pillar has to satisfy, where the brand can say something nobody else can, which pages that implies, and in what order they publish. Every decision traces to retrievability, entity clarity, source quality, and measurable demand.

AI search visibility is sampled, volatile, and engine-specific. This skill does not promise citations. It builds the thing that can be measured — pages to create, entities to clarify, passages to make retrievable, sources to cite, and the follow-up checks that show whether any of it worked.

Done when

Work type strategy-plan (also_covers: topical-map) — floor E3/C3/O1 (harness/eco-floors.yaml).

• E3 — a named human approved the exact content node architecture: entity clusters make strategic sense, hub-and-spoke groupings are right, information gain angles are achievable with data the brand actually has, no critical sub-query is missing, and the priority order by Citation Impact Score is correct.
• C3 — banned_word_check clean, every structural gate below met, and a named non-producer read the map end to end.
• O1 — the baseline AI Presence Scorecard is recorded before any page ships, with the re-read scheduled at 30/60/90 days. Each priority node names the query set it targets and who owns publishing it.

Structural gates, all mandatory:

Gate	Bar
Entity map	Every Tier 1 entity has a proposed Entity Home URL and a Wikidata action plan
Fan-out matrix	Every pillar has ≥6/8 sub-queries identified; pillars under 4/8 covered are priority gaps
Information gain	Every pillar has ≥2 "High" novelty opportunities, or is flagged for research before content creation
Schema	Every prescribed schema type has every relevant attribute populated

Constraints

• Read MARKETING.md from the project root before asking the user anything. If it does not exist, build it from the codebase — CLAUDE.md, README.md, PROJECT.md, package.json, landing pages, email/ad/analytics config — using the template from /kai-email-system, and confirm the draft.
• Know these before mapping: the brand entity, the topic space, the existing content inventory (blog URLs, episodes, landing pages, guides), 2–3 competitors currently winning citations in the space, the target queries, and the current AI presence.
• Baseline the AI presence with recorded conditions. Sample 3–5 category queries across ChatGPT, Perplexity, Bing/Copilot, and Google AI surfaces where available. Record prompt, location, date, engine, account state, citations, mentions, and missing-data caveats. This becomes the "before" measurement; re-run at 30/60/90 days.
• Never paste a benchmark into a client-facing map without a source URL, retrieval date, evidence tier, confidence label, and fit note. Treat Knowledge Graph, Wikidata, vector-search, traffic, citation, and conversion claims as source-dependent context, not promised outcomes.
• Case-study evidence needs method, measurement, source quality, applicability, and caveat before it informs a recommendation: what changed and whether a control existed; engine, prompt set, sample size, dates, geography, account state, and citation definition; official study vs vendor report vs internal measurement; whether the client's category, authority, depth, and distribution match; and what is still hypothesis.
• Distribution must be transparent. Participate in communities openly. No astroturfing, no seeded fake threads, no hidden ownership, no manufactured consensus.
• Do not compare AI and organic conversion rates without channel definitions, attribution windows, sample sizes, and confidence labels. Track AI traffic through source/medium rules, referrer inspection, landing-page cohorts, assisted conversions, and qualitative lead-source notes.
• Schema does not substitute for useful visible content, and llms.txt is useful for cooperative agents, not a Google AI Overview ranking requirement.
• This skill plans. It does not write pages or publish anything. The calendar hands off to /kai-content-calendar and /kai-write.

Context

Need	Load
AEO strategy foundation	knowledge/frameworks/aeo-ai-search/aeo-ai-search-playbook-2026.md
Entity tiers, Knowledge Graph, sameAs	knowledge/frameworks/aeo-ai-search/entity-seo-knowledge-graph-deep-dive.md
Query decomposition	knowledge/frameworks/aeo-ai-search/query-fan-out-guide.md
Page vs heading vs sentence	knowledge/frameworks/content-copywriting/qdp-qdh-qds-content-architecture.md
Novelty scoring	knowledge/frameworks/aeo-ai-search/patent-information-gain-US12013887B2.md + knowledge/frameworks/aeo-ai-search/hidden-aeo-edges.md
Retrieval and ranking behavior	knowledge/frameworks/aeo-ai-search/geo-academic-research-synthesis.md + knowledge/frameworks/aeo-ai-search/perplexity-ranking-reverse-engineered.md
Brief structure	harness/brief-schema.md
Calendar format compatibility	harness/skills/kai-content-calendar/SKILL.md
Product, ICP, voice, competitors	MARKETING.md (project root)

Entity tiers — every entity lands in one:

Tier	Meaning	Treatment
1 — Own	Brand, product, founder, proprietary methodology names	Entity Home URL + Wikidata QID (or submission plan) + Schema.org markup
2 — Associate	Industry terms, use cases, methodologies, problem categories, competitor categories	Content cluster topics; co-occurrence builds context vectors that signal expertise
3 — Reference	Research institutions, standards bodies, regulators, recognized experts	Cite for provenance and passage usefulness — no promised visibility lift

The site's @graph roots at Organization or Person with sameAs links (Wikidata, LinkedIn, Crunchbase, socials), knowsAbout listing Tier 2 entities, and mentions/about connecting pages to entities.

Query fan-out. Google's AI Mode decomposes a complex query into roughly 8 parallel sub-queries and synthesizes one answer with layered citations (Liz Reid, Google I/O 2025). Mine PAA 3–4 levels deep and categorize each sub-query by facet:

Facet	Example for "AI phone answering"
Definition	"What is an AI phone answering service?"
Cost	"How much does AI phone answering cost?"
Process	"How does AI phone answering work?"
Comparison	"AI phone answering vs live receptionist"
Safety/Risk	"Are AI phone answering services reliable?"
Timeline	"How long to set up AI phone answering?"
Alternatives	"Best AI phone answering services 2026"
Technical	"AI phone answering integrations with CRM"

Route each sub-query by demand: QDP (high demand + distinct intent) gets its own URL, QDH (moderate) becomes a section inside a hub or spoke, QDS (low) becomes an inline sentence.

Information gain. The patent (US12013887B2) scores novelty against existing content via embeddings — paraphrase scores low however it is worded. Find the consensus across the top 5–10 results, then place the brand's angle:

IG category	What it is	Why it works
Proprietary data	Original research, internal metrics, owned case studies	Reported 3.2x citation rate in Perplexity analysis
Contrarian position	Evidence-backed views against consensus	Triggers Perplexity's entropy diversity signal
Experience gap	First-person specifics a model cannot fabricate	Required for E-E-A-T "Experience" (QRG 4.6.6)
Novel framing	Unique terminology, frameworks, mental models	Creates semantic distance from competitors
Second-click content	Answers the follow-up query after the #1 result	Captures recursive fan-out

Score each opportunity High (data or experience ready to publish), Medium (angle exists, research needed first), or Low (theoretical, no evidence).

Content nodes. Entity clusters become Hubs, QDP sub-queries become Spokes, QDH items become sections, QDS items become sentences. Each node declares URL, title, node type (Hub / Spoke / Entity Home), primary entity, fan-out queries answered, IG angle, and its citation-signal targets:

• ≥3 sourced data points per page when the topic benefits from data
• ≥1 permissioned quote or attributed expert source when claims need authority
• ≥5 primary or high-quality secondary sources for research-heavy pages
• 2–3 verifiable atomic facts per paragraph
• 60–100 word paragraphs where that improves scannability and passage retrieval
• 15–20 word sentence ceiling
• Direct answer in the first 30–50 words after the H2

Schema prescription (eligibility and clarity, not guaranteed citation lift):

Node type	Schema	Purpose
FAQ / Q&A	FAQPage where eligible	Clarifies question-answer structure
How-to / process	HowTo where eligible	Clarifies steps, tools, prerequisites
Data / research / stats	Dataset where eligible	Clarifies dataset ownership and fields
All informational	Article or BlogPosting	Clarifies authorship, dates, subject
Entity Home	Organization or Person + sameAs	Clarifies canonical identity

Incomplete or generic schema carries a reported 18% citation penalty versus no schema at all (Growth Marshal, Feb 2026) — populate every relevant attribute.

Citation Impact Score sets publishing order:

CIS (1-10) = (Fan-Out Coverage × 0.3) + (IG Novelty × 0.3) + (Entity Authority × 0.2) + (GEO Signal Density × 0.2)

Fan-Out Coverage  = (sub-queries answered / total for pillar) × 10
IG Novelty        = Low 3 · Medium 6 · High 10
Entity Authority  = Entity Home 10 · Hub with schema 7 · Spoke 4 · QDH section 2
GEO Signal Density= planned stats + quotes + citations per page, normalized to 10

Sequencing. Entity Home pages first (weeks 1–2, with Schema.org and Wikidata submissions together — entity authority has no shortcut). Hubs before their spokes. High-IG pages inside month 1, where freshness compounds novelty. Cross-cluster pages spread through the calendar to keep building entity connections.

Distribution. Each node gets a plan; publishing only on the site leaves most of the reachable surface untouched.

Surface	Format	Timing
Own site	Hub / Spoke page	Day 0
LinkedIn	Article adaptation, 500–2,000 words	Day 1–2
Reddit	Discussion or comparison thread	Day 3–5
Industry directories	Listing or review (G2, Capterra, Yelp per vertical)	Week 1
Guest post / PR	Adapted article	Week 2–3

Engine	What it leans on	The move
Gemini / Google AI surfaces	Google Search crawl and index	Helpful, crawlable, snippet-eligible pages; schema; consistent subdomains
ChatGPT	Mixed retrieval, browsing, third-party consensus	Directory and review-site presence, "best of" inclusion, retrievable pages
Perplexity	Industry experts and niche directories	Vertical-specific: Zocdoc (health), Avvo (legal), G2 (SaaS)
Google AI Overviews	Search ranking, retrieval, query fan-out	Search fundamentals, source quality, page-level answers, entity presence, freshness
Community / social	Public discussion, professional context	Participate transparently

What each citation signal is actually good for — use the left column for the middle column only, and carry the caveat:

Signal	Use as	Caveat
External citations	Provenance and source-quality support	Not a guaranteed AI visibility lift
Expert quotes	Attributable authority and customer language	Requires permission and context
Statistics / data	Specificity and answer usefulness	Must be sourced and current
Proprietary research	Information gain and PR asset	Needs methodology disclosure
Topic clusters	Coverage and internal linking	Measure by queries and conversions, not vanity citation counts
Freshness	Accuracy maintenance	Update only when facts or examples change
Multi-platform distribution	Legitimate audience discovery	No astroturfing, hidden ownership, or fake consensus
Schema	Entity and content clarity	No substitute for useful visible content

Briefs for the first 4 weeks follow harness/brief-schema.md plus AEO extensions: information_gain_angle, fan_out_queries[], citation_signals (statistics/quotes/external citations required), entity_targets[], schema_type, a schema-completeness note, and a per-channel distribution_plan.

Output goes to workspace/topical-map/: _discovery.md, _entity-map.md, _fan-out-matrix.md, _information-gain-audit.md, _content-nodes.md, _schema-blueprint.json, _90-day-calendar.md, _distribution-plan.md, briefs/w[N]-[slug].json, _quality-report.md. The calendar is directly consumable by /kai-content-calendar.

Sources for the figures above: HubSpot AEO case studies (blog.hubspot.com/marketing/answer-engine-optimization-case-studies) · Superlines AI search statistics (superlines.io/articles/ai-search-statistics) · Yext AI visibility study (yext.com/blog/ai-visibility-in-2025-how-gemini-chatgpt-perplexity-cite-brands) · Semrush/ALM LinkedIn citations (almcorp.com/blog/linkedin-ai-search-citations-2026) · Schema App entity linking case study · Wellows AI Overview ranking factors · ALM AI Overview citation shifts · Discovered Labs on Reddit as an AEO signal source · GEO paper, Aggarwal et al. 2024 (Princeton/Georgia Tech) · Information Gain patent US12013887B2 (Carbune & Gonnet, 2024) · Query Fan-Out, Liz Reid, Google I/O 2025.

Escalate when

• The brand has no proprietary data or first-hand experience for a pillar — that pillar needs research before it earns a content plan.
• The topic space is regulated and the obvious content angles carry compliance risk.
• Competitor citation positions cannot be observed and the gap analysis would be guesswork.
• The user wants a citation, traffic, or ranking guarantee — that promise cannot be made.
• Entity Home pages would require site changes the user has not authorized.
• The plan's volume exceeds what the team can produce, and the priority cut needs a human decision.