Topical Authority for AI Search: How to Get Cited by ChatGPT, Perplexity, and Google AI Overviews

Topical authority has mattered in SEO since Google's Panda update in 2011. Write comprehensively about a topic, earn trust on it, rank for its keywords. The core logic has been stable for fifteen years.

In the AI search era, the stakes are different. Topical authority in 2025 does not determine how you rank. It determines whether an AI system knows you exist and considers you worth citing. ChatGPT, Perplexity, Claude, and Google's AI Overviews are all performing some version of the same operation: identify authoritative sources on a topic and synthesize their content into an answer. If your content is not considered authoritative on a topic, you are invisible in this synthesis.

The mechanism is not visible. AI systems do not publish their source weighting algorithms. But the patterns are consistent enough to work with. Sites with deep, comprehensive, interlinked coverage of a topic appear as citations significantly more often than sites with thin coverage of many topics. That is the topical authority signal translating directly into AI citation frequency.

The framework covers content cluster architecture, coverage depth requirements, entity-based optimization, measurement methodology, and the specific differences between traditional Google ranking signals and AI citation signals. Whether you are building a new content program or auditing an existing one, the principles here apply.

What Topical Authority Actually Means (and Why the Definition Changed)

Topical authority in traditional SEO meant: your site is recognized as a primary source on a topic based on link equity, content depth, and usage signals. The primary mechanism was PageRank flowing through topic-relevant links, an external signal from other sites vouching for your authority.

In the AI search era, topical authority means something related but distinct: your content's claims, definitions, and frameworks are represented in the training data and retrieval indexes used by AI systems, weighted by how consistently and comprehensively your site covers the topic.

The shift has two implications:

First: AI systems do not primarily use link signals to assess authority. When an AI generates an answer citing a source, it is using retrieval-augmented generation (for systems like Perplexity) or sampling from training data (for systems like ChatGPT). Neither mechanism directly measures inbound links. What they measure is closer to: does this source have high-quality, specific, accurate content on this exact topic? Does it appear frequently in the index when queries about this topic are issued?

Second: Comprehensiveness matters more at the topic level, not the page level. A single well-written page about LLM inference gets you into Google. Comprehensive coverage of LLM inference (covering tokenization, attention, KV cache, quantization, speculative decoding, serving infrastructure) gets you established as a topic authority that AI systems associate with the domain.

The practical distinction: you can rank in Google for one keyword with one great article. You need a coherent topic cluster to appear consistently in AI-generated answers about a domain.

E-E-A-T: Experience, Expertise, Authoritativeness, Trustworthiness. The first-hand experience signal is new and directly relevant to AI citation patterns. AI systems trained on web data appear to weight experiential, first-person accounts differently from aggregated information.

Source: Google Search Quality Evaluator Guidelines, 2024

The KDD 2024 GEO (Generative Engine Optimization) paper found that sources with greater topical coverage (defined as the breadth of subtopics covered within a domain) appear in AI-generated answers up to 40% more frequently than sources with comparable depth on individual topics but narrower breadth. Topical comprehensiveness is a direct input to AI citation likelihood.

Traditional Search vs AI Search: Different Signal Architecture

To build content strategy correctly for both channels, you need to understand how their ranking signals differ:

The key insight from this comparison: backlinks matter much less for AI visibility than they do for Google ranking. This is significant for new sites and smaller publishers. The primary investment required for AI topical authority is content quality and comprehensiveness, not link acquisition.

Backlinks remain the primary signal for Google ranking, and Google ranking remains the primary driver of URLs that end up in Perplexity's retrieval index. The relationship is indirect: backlinks → Google ranking → Perplexity index inclusion → AI citation. But the relationship between backlinks and direct AI visibility (ChatGPT training representation, Perplexity preference weighting) is weaker.

What this means for content investment priorities:

For traditional Google ranking: One deep, well-linked pillar article beats a topic cluster of thin pages.

For AI topical authority: A coherent topic cluster where each article is genuinely comprehensive and all articles are interlinked beats a single monolithic article, even at the same total word count.

Building for both requires the cluster approach. Individual articles remain deep and linkable; the cluster structure signals topical comprehensiveness to AI systems.

Content Cluster Architecture: The HubandSpoke Model for 2025

The hub-and-spoke content cluster model (one pillar article covering a broad topic, multiple spoke articles covering specific subtopics) is the right architectural foundation for both traditional SEO and AI topical authority. But the 2025 version has specific requirements that the 2018 version did not.

The pillar article (hub):

• Defines the topic and its key subtopics with original, specific definitions
• Covers the full scope at a high level with enough depth to be self-contained
• Provides the authoritative definitional sentences that AI systems can quote directly
• Links to all spoke articles explicitly
• Contains the primary statistics and claims with citation sources
• Target: 4,000-6,000 words for competitive topics

The spoke articles:

• Cover one specific subtopic in full technical depth
• Are self-contained (can be read without the pillar)
• Contain specific, quotable claims with evidence
• Link back to the pillar and to adjacent spokes
• Target: 2,500-4,000 words each

The coverage map: the set of subtopics the cluster must cover to establish topical authority:

For a technical domain like LLM inference:

Pillar: LLM Inference Optimization (overview + taxonomy)
├── Spoke: Memory Bandwidth and the Roofline Model
├── Spoke: KV Cache Architecture and Optimization
├── Spoke: Quantization Methods (int8, int4, GPTQ, AWQ)
├── Spoke: Speculative Decoding
├── Spoke: Continuous Batching and Serving Infrastructure
├── Spoke: Model Architecture for Inference (GQA, MLA, etc.)
├── Spoke: Edge Inference Patterns
└── Spoke: Benchmarking and Measurement

The minimum viable cluster for topical authority signals to AI systems is 6-8 spokes covering the most-queried subtopics of the pillar topic. A cluster with 3-4 spokes will rank individual pages but will not establish the topical authority that drives consistent AI citation.

The coverage gap analysis: Before writing, audit what the cluster needs to cover. The method:

1. List the top 20-30 questions users ask about your topic (use forums, Reddit, answer-the-public, your own search query data)
2. Map each question to a subtopic
3. Identify which subtopics you have no content for
4. Prioritize coverage gaps by question volume and competitive gap (how bad is the existing content?)

The questions that most need answered (with specificity and evidence) are the ones that will drive AI citation when covered well.

Coverage Depth: What "Comprehensive" Means for AI Citation

"Comprehensive" is easy to say and hard to define. In the context of AI citation patterns, the specific properties that characterize content AI systems prefer to cite are:

1. Operational specificity: Specific numbers, formulas, thresholds, and measurements rather than vague claims.

Bad: "KV cache quantization can significantly reduce memory requirements." Good: "Int8 KV cache quantization reduces memory by approximately 2x compared to bfloat16 with less than 0.5 perplexity points of quality degradation."

The specific version is quotable. It contains a claim with a number that an AI can cite with attribution. The vague version is not.

2. Definitional completeness: Every major term used in the content is defined within the content. AI systems treat pages that define their own terminology as higher-authority sources than pages that assume term knowledge.

A page about GQA that defines what GQA stands for, what problem it solves, how it differs from MHA, and when to use it will be cited more often than a page that uses GQA as an assumed term.

3. Logical structure that stands alone: Each section should answer a complete question. A section titled "When to Use StreamingLLM" that actually answers when to use StreamingLLM (specifically and completely) is a candidate for AI citation. A section that says "StreamingLLM has many use cases" is not.

4. First-hand evidence and examples: The E-E-A-T signal for experience is real in AI systems. Content that includes "in production deployment X, we observed Y" or "when I tested this with model Z, the result was W" is treated differently from pure aggregation of external sources. This is consistent with the GEO paper's finding that statistics with attribution outperform unattributed statistics for AI citation frequency.

5. Claim density: AI systems retrieve content based on semantic similarity to queries. Content with higher density of specific, accurate claims about a topic matches a wider range of related queries. A 3,000-word article that makes 15 specific, citable claims will be retrieved more often than a 5,000-word article that makes 5 specific claims surrounded by padding.

Practical word count guidance by content type:

These are minimums for content that will be cited, not just ranked. Below these thresholds, the content may rank for long-tail keywords but is unlikely to appear consistently as an AI citation source.

Entity Optimization: Becoming Recognizable Across Training Data

Entity-based SEO has been discussed since Google's Knowledge Graph launched in 2012. In the AI era, entity recognition is more important and more tractable than before.

An entity in this context is a recognizable concept, person, organization, product, or concept that appears consistently across your content and across external references to your content. AI systems build an internal representation of entities from training data. The more consistent and specific your content's treatment of the entities you own, the stronger your entity association.

What entities to establish and how:

Proprietary frameworks and concepts: Coin specific terminology for concepts you introduce. The "roofline model" in ML inference is a concept with a specific owner (Williams et al., 2009). Content about ML inference now reliably cites it. If you introduce a useful taxonomy or framework, name it explicitly, define it consistently, and reference it across multiple pieces. This creates an entity that AI systems can associate with your domain.

Definitional sentences: Every major concept should have a definitional sentence that works as a standalone quotable claim. The format: "[Term] is [what it is] that [what it does/why it matters]." Write one for every major term in your content. AI systems preferentially cite content with clear, quotable definitional sentences.

Named approaches: When describing implementation patterns, give them names. "The dual-indexed episodic store pattern" is more citable than "the way we implemented memory." Names create entities. Entities get cited.

Schema markup for entity recognition: Use Schema.org markup to signal entities explicitly to crawlers that feed AI training and retrieval:

{
  "@context": "https://schema.org",
  "@type": "TechArticle",
  "headline": "KV Cache Internals: Compression, Quantization, and Eviction",
  "author": {
    "@type": "Person",
    "name": "Chait",
    "knowsAbout": ["LLM inference", "edge AI", "transformer architectures"]
  },
  "about": [
    {"@type": "DefinedTerm", "name": "KV cache", "description": "Key-Value cache storing intermediate attention tensors in transformer inference"},
    {"@type": "DefinedTerm", "name": "KV cache quantization", "description": "Reducing precision of cached attention tensors to int8 or int4 to reduce memory consumption"}
  ],
  "mentions": ["Llama 3.1", "vLLM", "PagedAttention", "StreamingLLM", "H2O"]
}

Cross-article entity consistency: Use the same terminology across all articles in a cluster. If you call it "grouped query attention" in one article and "GQA" in another without cross-referencing, you weaken entity association. Establish canonical names and use them consistently.

Internal Linking Strategy for Topical Coherence

Internal linking serves two distinct purposes in the AI era: PageRank flow (traditional SEO) and topical coherence signaling (AI systems).

For topical coherence signaling, the linking pattern that matters is bidirectional completeness within the cluster. Every spoke article should link to the pillar. Every spoke article should link to the adjacent spokes it is conceptually related to. The pillar should link explicitly to every spoke.

The reason this matters for AI systems: crawlers building retrieval indexes treat a cluster of highly interconnected pages about a topic as evidence of topical authority on that topic. A site with 10 articles about LLM inference, all interlinked, signals different authority than a site with 10 disconnected articles on 10 different topics, even at identical total word count.

Anchor text specificity: For AI retrieval signals, descriptive anchor text is more important than keyword-rich anchor text. "See our detailed guide to KV cache quantization for memory reduction formulas" is more valuable than "KV cache quantization [link]". The description tells the crawler what the target article contains.

The three-link minimum: Every spoke article should have at minimum three internal links in the body (not counting navigation/sidebar): link to the pillar, link to the most adjacent spoke, and one contextual link to any cluster article where it adds genuine value. This is the minimum for topical coherence signaling.

Avoid orphan articles: Any article with zero internal links pointing to it is invisible to topical authority signals. Before publishing a new piece, ensure it will be linked from the pillar and at least one adjacent spoke. Build the links before publishing, not after.

Freshness, Accuracy, and Claim Specificity

For AI retrieval systems (Perplexity and Google AI Overviews), freshness matters directly. They are querying live indexes and can see publication dates. For AI training-based systems (ChatGPT), freshness only matters up to the training cutoff.

Freshness signals for AI retrieval:

• Publication date visible in the page
• Last-updated date visible in the page (more important than publication date for evergreen content)
• Date in the URL (year at minimum, ideally YYYY/MM)
• Content that explicitly references the current year or recent developments

Accuracy signals that AI systems respond to:

• Claims backed by citations to primary sources (papers, official documentation, official announcements)
• Specific version numbers and dates for software/model references ("vLLM 0.3.0, released March 2024")
• Corrections and updates clearly marked ("Updated April 2025: The following section reflects the new PagedAttention v2 implementation")
• Numbers that can be independently verified (hardware benchmarks, official specs)

The correction signal: Sites that visibly correct errors when they are identified (with explicit correction notices in the content) signal higher accuracy standards. AI systems trained to prefer accurate sources appear to weight correction culture positively. A site that has made errors and corrected them visibly may be considered more trustworthy than a site that has never visibly corrected an error.

Claim specificity in competitive contexts: In AI-generated answers, the sources cited for specific quantitative claims are disproportionately likely to be the origin of those claims (either the original research that established the number, or the first accessible popularization of it). If you publish a specific, citable claim with a number before anyone else, you establish priority. "Int8 KV cache quantization reduces memory by approximately 2x compared to bfloat16" with verification is more citable than citing someone else's statement of the same thing.

Measuring Topical Authority in the AI Era

Topical authority is more abstract than keyword ranking. Here is how to measure it practically:

Traditional SEO measurement (still necessary):

• Keyword coverage across cluster: track ranking positions for your target keywords and the long-tail variations around your cluster topic
• Topical coverage score: count the distinct subtopics you have ranking content for vs. total subtopics in the domain
• Internal link equity: monitor that your spoke articles are passing PageRank back to the pillar effectively

AI citation measurement:

Direct measurement is manual but tractable. For each major topic you want to own, run a monthly audit:

Query set for LLM inference cluster:
1. "how does KV cache work in LLMs"
2. "what is speculative decoding"
3. "KV cache quantization tradeoffs"
4. "how to optimize LLM inference memory"
5. "what is grouped query attention"

For each query, check:

• Perplexity: is your content cited as a source?
• ChatGPT (GPT-4): does your content appear in generated answers when browsing is enabled?
• Google AI Overview: is your content shown in the AI Overview panel?
• Claude (web): does Claude cite your content in factual answers?

Track citation frequency over time. A rising citation rate in AI answers for your cluster topic is the strongest available signal of improving AI topical authority.

Proxy metrics that correlate with AI authority:

• Featured snippet win rate: Google featured snippets are essentially the same AI authority signal applied to traditional search: structured, standalone answers to specific questions. A site that wins featured snippets for cluster keywords is also a candidate for AI citation.

• Knowledge panel association: If your entity (company, person, product) has a Knowledge Panel on Google associated with your cluster topics, this entity association carries over to AI systems.

• Share of voice in AI-generated category definitions: Search "what is [your primary topic]?" in three AI systems. Which sources are cited? Which frameworks are used? If your framework is reflected without citation (your concept represented but not attributed), you have entity presence without citation capture. The next step is ensuring your attribution is maintained.

Competitive benchmarking:

Audit your top three competitors' citation rates for the same query set. If they appear in 3/5 queries and you appear in 1/5, you need 2-3x more cluster depth to close the gap. This is typically achievable with 4-6 additional well-structured spoke articles targeting the gaps.

Cluster Templates: Topic Coverage Maps for Common Domains

Template 1: AI/ML Technical Domain

Pillar: [Technology]: Overview, Taxonomy, and Why It Matters
├── [Technology] Fundamentals: How [Core Mechanism] Works
├── [Technology] Performance: Benchmarks and Measurement
├── [Technology] at Scale: Production Architecture
├── [Technology] vs [Alternative]: Technical Comparison
├── [Technology] Optimization: Advanced Techniques
├── [Technology] for [Specific Use Case]: Practical Guide
├── [Technology] Tradeoffs: When Not to Use It
└── [Technology] Ecosystem: Libraries, Frameworks, Tools

Template 2: Business/Strategy Domain

Pillar: [Strategy]: Complete Framework for [Audience]
├── [Strategy] Foundations: Core Principles and Mental Models
├── [Strategy] for Early-Stage: Getting Started Guide
├── [Strategy] for Scale: Enterprise Patterns
├── [Strategy] Metrics: What to Measure and How
├── [Strategy] Tools: Software and Platform Comparison
├── [Strategy] Case Studies: [Domain] in Practice
├── [Strategy] Mistakes: What Fails and Why
└── [Strategy] Future: Trends and Emerging Patterns

Template 3: How-To/Tutorial Domain

Pillar: How to [Accomplish Goal]: Complete 2025 Guide
├── [Goal] Explained: What It Is and How It Works
├── [Step 1]: Full Technical Guide
├── [Step 2]: Full Technical Guide
├── [Common Problem 1]: Diagnosis and Fix
├── [Common Problem 2]: Diagnosis and Fix
├── [Tool/Library]: Deep Dive
├── [Alternative Approach]: When to Use It
└── [Goal] Best Practices: Production-Ready Patterns

The minimum cluster size to establish topical authority: 7-8 total pieces (1 pillar + 6-7 spokes). Below 6 spokes, the coverage breadth signal is insufficient for AI systems to associate you with the topic cluster rather than individual keywords.

Common Topical Authority Mistakes in the AI Era

Mistake 1: Building for breadth of topics instead of depth on one topic

Many publishers respond to "you need more content" by spreading across more topics. This produces a site that ranks for 50 topics at low authority instead of ranking at high authority for 5 topics. AI systems cite sources that own a topic, not sources that touch many topics. Build full authority on one cluster, then add an adjacent cluster. Do not add thin coverage across many unrelated topics.

Mistake 2: Writing for the topic without writing for the questions

Topic authority comes from answering the specific questions users ask about a topic, not from covering the topic abstractly. A post titled "KV Cache Architecture" that thoroughly explains the architecture but does not directly answer "how much memory does a KV cache use at 128K context?" will rank on broad terms but miss the specific question queries where AI citation happens. Map your content to specific questions, not to topics.

Mistake 3: Treating cluster building as a one-time project

Topical authority degrades when a domain evolves and your content does not. New research, new models, new tools: all create freshness gaps. A cluster that was authoritative in 2023 and has not been updated loses citation frequency to newer content on the same topics. Plan quarterly cluster audits: identify which articles need freshness updates, which new subtopics have emerged, which existing articles have accuracy issues.

Mistake 4: Ignoring the spoke-to-spoke linking structure

The most common implementation error is building a pillar and spokes but only linking spokes to the pillar, not to each other. The topical coherence signal requires lateral links between related spokes. A cluster about LLM inference where the KV cache article and the speculative decoding article do not cross-reference each other signals weaker topical cohesion than a cluster where every related pair of articles links to each other.

Mistake 5: Publishing definitional content without operational content

"What is X" articles are valuable but insufficient for topical authority. AI systems need evidence that you understand X at the operational level: how to implement it, what goes wrong, what the specific performance characteristics are. For every definitional article, the cluster needs at least one operational article (tutorial, case study, troubleshooting guide) that demonstrates practical knowledge of the same topic.

Mistake 6: Not owning your statistics and attributions

Publishing a statistic without citing its source is a citation risk: AI systems may cite the original source rather than your content. Publishing a statistic with an explicit citation ("According to [Source], X%...") establishes your content as the synthesizer rather than the origin. Better: generate your own measurements and data. Original measurements are citable in a way that cited-from-elsewhere statistics never fully are.

Key Takeaways

• Topical authority in the AI era means your content's claims, definitions, and frameworks are represented in AI training data and retrieval indexes with enough consistency and comprehensiveness that AI systems associate your site with the topic. This is different from, and partially independent of, backlink-based Google authority.

• The minimum viable cluster for AI topical authority is 7-8 total pieces: one pillar article (4,000-6,000 words covering the topic comprehensively) plus 6-7 spoke articles (2,500-4,000 words each) covering the primary subtopics. Below this threshold, coverage breadth is insufficient for consistent AI citation.

• AI systems preferentially cite content with operational specificity (specific numbers and formulas), definitional completeness (every major term defined inline), logical structure that answers questions directly, and first-hand evidence. These properties make content quotable, which is the prerequisite for being quoted.

• Internal linking within a cluster serves two purposes: PageRank flow (traditional SEO) and topical coherence signaling (AI systems). Both require bidirectional completeness: every spoke links to the pillar, every spoke links to adjacent spokes, and the pillar links to every spoke. Spoke-to-spoke linking is the most commonly omitted element.

• Measure AI topical authority by running a monthly query audit: 5-10 specific questions about your cluster topic across Perplexity, ChatGPT (with browsing), Google AI Overviews, and Claude. Track citation frequency over time. A rising citation rate is the primary available signal of improving AI authority.

• Topical authority degrades without maintenance. Quarterly cluster audits should identify freshness gaps, new subtopics that need coverage, and accuracy issues in existing articles. A cluster authoritative in 2023 that has not been updated will lose AI citation frequency to newer content on the same topics within 12-18 months.

FAQ

What is topical authority in SEO?

Topical authority in SEO is the degree to which a website is recognized as a primary, comprehensive source on a specific topic or topic cluster. A site with high topical authority on a topic has deep, interlinked coverage across the topic's subtopics, is cited and referenced by other sources in the domain, and consistently ranks for the full range of queries related to the topic. Topical authority is distinct from domain authority (overall site strength based on backlinks). A site can have high topical authority in a narrow domain with few total backlinks if its coverage of that domain is sufficiently comprehensive and accurate.

How does topical authority affect AI search visibility?

Topical authority directly affects AI search visibility because AI systems like Perplexity, ChatGPT, and Google AI Overviews preferentially cite sources that are authoritative on the topic of the query. For retrieval-augmented AI systems (Perplexity), topical authority determines which sources appear most frequently in the retrieval results for a topic cluster. For training-based AI systems (ChatGPT), topical authority affects how strongly a source is represented in training data for a given topic. The KDD 2024 GEO paper found that sources with greater topical coverage (broader breadth across subtopics) appear in AI-generated answers up to 40% more frequently than sources with comparable depth on individual topics but narrower breadth.

How many articles do you need for a content cluster?

A minimum viable content cluster for topical authority signals consists of one pillar article plus six to seven spoke articles: eight pieces total. Below six spokes, the coverage breadth is typically insufficient for AI systems to associate the site with the topic cluster rather than with individual keywords. The pillar article should cover the topic comprehensively at high level (4,000-6,000 words for competitive topics), while spoke articles cover individual subtopics in full depth (2,500-4,000 words each). The total word count of the cluster matters less than the completeness of subtopic coverage: a cluster that covers 8 subtopics shallowly will establish weaker topical authority than a cluster that covers 6 subtopics comprehensively.

What is the difference between GEO and SEO for content strategy?

GEO (Generative Engine Optimization) focuses on optimizing content to appear in AI-generated answers from systems like ChatGPT, Perplexity, and Google AI Overviews. SEO focuses on optimizing for traditional search result rankings. The key differences in content strategy: GEO prioritizes claim specificity (exact numbers, quotable statements) over keyword density; GEO values topical cluster comprehensiveness over individual page optimization; GEO rewards first-hand experience signals and source citations within content; GEO benefits from definitional sentences (clear "X is Y" formulations) that AI systems can quote directly. In practice, a well-executed SEO strategy that emphasizes E-E-A-T quality signals overlaps substantially with GEO strategy. The content properties that make AI systems cite you are largely the same properties that make Google rank you highly.

How do you build topical authority for a new site?

Building topical authority on a new site requires choosing one specific topic cluster and going comprehensively deep before expanding. The process: (1) choose one topic cluster narrow enough to cover completely with 8-12 articles; (2) audit what questions users ask about the topic and what existing content fails to answer well; (3) write the pillar article first, establishing the taxonomy and definitional framework; (4) write spoke articles covering each primary subtopic in full depth, linking each back to the pillar and to adjacent spokes; (5) publish with explicit publication dates, author credentials, and Schema markup; (6) build initial links to the pillar from relevant sources; (7) run monthly AI citation audits to measure visibility progress. Resist expanding to new topic clusters until you can demonstrate AI citation for the first cluster. Attempting breadth before establishing depth produces mediocre authority across many topics rather than strong authority in one.

Topical authority has never been achieved by publishing one great article. It has always required demonstrating that you understand a domain comprehensively: knowing the whole landscape, not just one peak on it.

In the AI search era, the mechanism is the same but the signal detection is different. Google measures comprehensiveness through link equity and user signals. AI systems measure it through the coverage and coherence of your content cluster. The content properties that drive AI citation (specificity, definitional completeness, operational depth, first-hand evidence) are different from keyword optimization but overlapping with genuine quality.

The practical implication: the content strategy that builds AI topical authority is not a shortcut strategy. It is the same strategy that builds genuine subject matter authority: covering a domain completely, at depth, with accurate and specific claims, updated when the domain changes. The reward in the AI search era is direct. AI systems that cite your content bring traffic and credibility in an environment where zero-click AI answers are replacing traditional search clicks. Topical authority is what determines whether the AI gives away your content or tells users to go read it.

Build the cluster. Own the topic. The citations follow.