import { writeFileSync } from 'fs';

// The article body content from ContentOS draft, converted to HTML sections
const bodySections = `
        <section aria-labelledby="why-heading">
          <p class="eyebrow">The shift</p>
          <h2 id="why-heading">Why AI citation structure matters for B2B content in 2026</h2>
          <p>
            LinkedIn Marketing Solutions reports that 71% of respondents say B2B content marketing has grown more important to their organizations over the past year [1]. That importance now runs through a new gate. Buyers ask ChatGPT, Gemini, Perplexity, and Google AI Overviews before they open a vendor page — and the answer they read is assembled from whatever those systems can parse, verify, and quote.
          </p>
          <p>
            Citation is not a keyword problem. It is a structure problem. A page earns citations when a machine can find a direct answer, resolve the entities in it, confirm the source behind each claim, and read structured data that mirrors the visible text. This guide covers the full workflow: prompt-page maps, answer blocks, entity tables, Schema.org markup, source packs, proof gates, and citation gap repair after publication.
          </p>
          <p>
            Two acronyms cover the discipline. AEO (Answer Engine Optimization) targets the answer surfaces themselves. GEO (Generative Engine Optimization) targets how generative models assemble and attribute content. From here on, both appear as AEO/GEO — the practices overlap almost entirely at the page level.
          </p>
          <p>
            The <a href="/research/aeo-geo-operating-system-weekly-workflow/">AEO/GEO Operating System</a> is a six-part workflow, and each part feeds the next:
          </p>
          <ul>
            <li><strong>Prompt-page map</strong> — the 12–15 buyer prompts one page must answer.</li>
            <li><strong>Answer state taxonomy</strong> — the five kinds of answers a prompt can demand.</li>
            <li><strong>Entity map</strong> — the named things the page must define in stable wording.</li>
            <li><strong>Schema layer</strong> — Article, Organization, and FAQPage markup on Schema.org vocabulary [5].</li>
            <li><strong>Source packs</strong> — grouped evidence behind every factual claim.</li>
            <li><strong>Proof gates</strong> — four checks that run before publish.</li>
          </ul>
          <p>
            Skip any one and the others weaken. Schema without answer blocks marks up prose no model wants to quote. Answer blocks without source packs produce confident sentences with no verifiable backing. A CMO reviewing an underperforming pillar page should audit all six, not just the markup.
          </p>
        </section>

        <section aria-labelledby="prompt-map-heading">
          <p class="eyebrow">Planning</p>
          <h2 id="prompt-map-heading">Build a prompt-page map before writing the AI-citable article</h2>
          <p>
            A <a href="/research/prompt-page-map-ai-search-site-architecture/">prompt-page map</a> lists the exact questions a buyer types into an answer engine, then assigns each one to a section of the page. Build it before the outline. Twelve to fifteen entries is the working range for a 2,500-word article — fewer leaves gaps, more splits attention across topics that deserve their own URLs.
          </p>
          <p>
            For this topic, the map looks like this:
          </p>
          <div class="facts-table">
            <table>
              <thead>
                <tr><th>#</th><th>Prompt</th><th>Answer state</th><th>Section</th></tr>
              </thead>
              <tbody>
                <tr><td>1</td><td>What is AI citation in search?</td><td>Definition</td><td>Section 1</td></tr>
                <tr><td>2</td><td>What does AEO/GEO mean?</td><td>Definition</td><td>Section 1</td></tr>
                <tr><td>3</td><td>Why does content structure affect AI citations?</td><td>Evidence</td><td>Section 1</td></tr>
                <tr><td>4</td><td>What is a prompt-page map?</td><td>Definition</td><td>Section 2</td></tr>
                <tr><td>5</td><td>How do you plan sections for AI search?</td><td>Procedure</td><td>Section 2</td></tr>
                <tr><td>6</td><td>What is an answer block?</td><td>Definition</td><td>Section 3</td></tr>
                <tr><td>7</td><td>How long should an answer block be?</td><td>Procedure</td><td>Section 3</td></tr>
                <tr><td>8</td><td>What is semantic completeness?</td><td>Definition</td><td>Section 3</td></tr>
                <tr><td>9</td><td>What is an entity in SEO?</td><td>Definition</td><td>Section 4</td></tr>
                <tr><td>10</td><td>Which entities should a SaaS page define?</td><td>Decision-support</td><td>Section 4</td></tr>
                <tr><td>11</td><td>What is schema markup for AI citations?</td><td>Definition</td><td>Section 5</td></tr>
                <tr><td>12</td><td>FAQPage vs Article vs Organization schema — which to use?</td><td>Comparison</td><td>Section 5</td></tr>
                <tr><td>13</td><td>How do you verify a claim before publishing?</td><td>Procedure</td><td>Section 6</td></tr>
                <tr><td>14</td><td>What is citation gap repair?</td><td>Definition</td><td>Section 7</td></tr>
                <tr><td>15</td><td>What does a finished AI-citable page look like?</td><td>Decision-support</td><td>Section 8</td></tr>
              </tbody>
            </table>
          </div>
          <p>
            <strong>The answer state taxonomy</strong> sorts prompts into five kinds: definition, comparison, procedure, evidence, and decision-support. Assign one primary state per H2. Mixed-state sections read fine to humans and extract badly for machines, because the model cannot tell whether the paragraph is defining a term or recommending a choice.
          </p>
          <p>
            Word budgets keep the map honest. Across eight H2 sections plus one FAQ block, a 2,500-word article splits roughly 280 / 320 / 350 / 320 / 360 / 300 / 280 / 190, with the FAQ absorbing the remainder. Write the budget into the outline document. It stops the definition section from swallowing 800 words that belonged to implementation.
          </p>
          <p>
            Main implementation checklist:
          </p>
          <ul>
            <li>List 12–15 real buyer prompts from sales calls, support tickets, and search console queries.</li>
            <li>Tag each prompt with one of the five answer states.</li>
            <li>Assign one primary answer state to each H2.</li>
            <li>Set a word budget per H2 before drafting.</li>
            <li>Mark which prompts need a visible question-based H3.</li>
            <li>Flag prompts that belong on a different URL entirely.</li>
            <li>Record the target answer engine for each prompt cluster.</li>
          </ul>
        </section>

        <section aria-labelledby="answer-blocks-heading">
          <p class="eyebrow">Content structure</p>
          <h2 id="answer-blocks-heading">Write answer blocks that AI systems can extract without context loss</h2>
          <p>
            MarketEngine defines answer blocks as short, self-contained sections that answer a specific question directly, helping readers and AI systems find important information without searching an entire page [13]. Self-contained is the operative word. If a block depends on the paragraph above it, a model that lifts it in isolation produces a broken quote — and usually skips it instead.
          </p>
          <p>
            MarketEngine's method runs in four steps: identify the primary user question, write a direct answer, build semantic completeness, and organize the information with a clear structure [13]. Semantic completeness means covering the topic with enough context, related concepts, examples, and supporting detail that the answer stands on its own [13].
          </p>
          <h3>How many answer blocks should one article contain?</h3>
          <p>
            Eight to ten, each 40–70 words, each sitting under one visible question-based H3. That range covers the major prompts in the map without turning the article into a stack of disconnected snippets. Blocks shorter than 40 words rarely carry enough context; blocks over 70 words start competing with the section prose around them.
          </p>
          <h3>What is schema markup for AI citation?</h3>
          <p>
            Schema markup is structured data added to a page in JSON-LD that describes entities, content type, and publication details in a format machines read directly. AISO Hub identifies three ways it supports AI citations: entity clarity, evidence exposure, and freshness and accuracy [7]. Organization, Person, Product, and LocalBusiness types tie claims to stable entities models already recognize [7].
          </p>
          <p>
            That block is the pattern. Question as an H3, answer in the first sentence, supporting detail and a cited source underneath, no dependency on surrounding text.
          </p>
        </section>

        <section aria-labelledby="entities-heading">
          <p class="eyebrow">Entity optimization</p>
          <h2 id="entities-heading">Use entities to make the page machine-readable before schema markup</h2>
          <p>
            Clearscope defines an entity as a distinct, singular, well-defined thing or concept that search engines can recognize and understand [9]. Entities are not keywords. "Schema markup" as a keyword is a string; Schema.org as an entity is a specific collaborative project with a URL, a vocabulary, and a place in Google's Knowledge Graph.
          </p>
          <p>
            Search Engine Land frames entity-first SEO as optimizing content for the entities behind keywords rather than the keyword strings themselves, aligning pages with Google's Knowledge Graph [10]. The practical consequence: pick one canonical wording per entity and never drift. Writing "FAQ schema" in one section and "FAQPage schema" in another splits a single entity into two weak mentions.
          </p>
          <p>
            A finished page should define at least 12 named entities. Build them into a five-column table:
          </p>
          <div class="facts-table">
            <table>
              <thead>
                <tr><th>Entity</th><th>Type</th><th>Canonical wording</th><th>Supporting source</th><th>On-page placement</th></tr>
              </thead>
              <tbody>
                <tr><td>Schema.org</td><td>Organization / vocabulary</td><td>Schema.org</td><td>schema.org [5]</td><td>Schema section</td></tr>
                <tr><td>FAQPage schema</td><td>Structured data type</td><td>FAQPage schema</td><td>Averi.ai [8]</td><td>FAQ + schema section</td></tr>
                <tr><td>Article schema</td><td>Structured data type</td><td>Article schema</td><td>Averi.ai [8]</td><td>Page metadata</td></tr>
                <tr><td>Organization schema</td><td>Structured data type</td><td>Organization schema</td><td>Averi.ai [8]</td><td>About/brand block</td></tr>
                <tr><td>Google Knowledge Graph</td><td>Knowledge base</td><td>Google Knowledge Graph</td><td>Search Engine Land [10]</td><td>Entity section</td></tr>
                <tr><td>ChatGPT</td><td>AI answer platform</td><td>ChatGPT</td><td>—</td><td>Intro + repair section</td></tr>
                <tr><td>Gemini</td><td>AI answer platform</td><td>Gemini</td><td>—</td><td>Intro + repair section</td></tr>
                <tr><td>Perplexity</td><td>AI answer platform</td><td>Perplexity</td><td>—</td><td>Intro + repair section</td></tr>
                <tr><td>Google AI Overviews</td><td>AI search surface</td><td>Google AI Overviews</td><td>—</td><td>Intro + repair section</td></tr>
                <tr><td>Clearscope</td><td>Software company</td><td>Clearscope</td><td>clearscope.io [9]</td><td>Entity section</td></tr>
                <tr><td>AISO Hub</td><td>Publisher</td><td>AISO Hub</td><td>aiso-hub.com [7]</td><td>Schema section</td></tr>
                <tr><td>MarketEngine</td><td>Publisher</td><td>MarketEngine</td><td>marketengine.ai [13]</td><td>Answer block section</td></tr>
              </tbody>
            </table>
          </div>
          <p>
            Five entity classes matter most for a B2B SaaS page: brand, category, problem, product, and method. A payments company writing about reconciliation needs its own brand entity, the category entity ("accounts receivable software"), the problem entity ("unapplied cash"), the product entity, and the method entity behind its approach. Each gets one canonical phrasing, one definition sentence, and one placement in the schema layer.
          </p>
        </section>

        <section aria-labelledby="schema-heading">
          <p class="eyebrow">Structured data</p>
          <h2 id="schema-heading">Add schema markup that exposes entities, evidence, and freshness</h2>
          <p>
            Schema.org is a collaborative, community activity with a mission to create, maintain, and promote schemas for structured data on the internet [5]. It supplies the vocabulary. The three types that carry the most weight for AI citation are Article, Organization, and FAQPage, and Averi.ai's implementation guide provides JSON-LD examples for all three [8].
          </p>
          <p>
            AISO Hub sorts the value into three categories [7]:
          </p>
          <ul>
            <li><strong>Entity clarity</strong> — markup ties claims to stable entities models already resolve.</li>
            <li><strong>Evidence exposure</strong> — author, publisher, and citation fields surface who stands behind the content.</li>
            <li><strong>Freshness and accuracy</strong> — date fields tell a model how current the claim is.</li>
          </ul>
          <p>
            Stack them in three layers. Article at page level, Organization at brand level, FAQPage at question level.
          </p>
          <h3>Article schema — page level</h3>
          <pre><code>{
  "@context": "https://schema.org",
  "@type": "Article",
  "headline": "How to Structure Content for AI Citation",
  "datePublished": "2026-08-10",
  "dateModified": "2026-08-10",
  "author": { "@type": "Person", "name": "Author Name" },
  "publisher": {
    "@type": "Organization",
    "name": "Your Company",
    "logo": { "@type": "ImageObject", "url": "https://example.com/logo.png" }
  },
  "about": ["Schema markup", "Answer engine optimization", "Entity SEO"]
}</code></pre>
          <h3>Organization schema — brand level</h3>
          <pre><code>{
  "@context": "https://schema.org",
  "@type": "Organization",
  "name": "Your Company",
  "url": "https://example.com",
  "logo": "https://example.com/logo.png",
  "description": "B2B content and AI search visibility for SaaS teams.",
  "sameAs": [
    "https://www.linkedin.com/company/your-company",
    "https://github.com/your-company"
  ]
}</code></pre>
          <h3>FAQPage schema — question level</h3>
          <pre><code>{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [{
    "@type": "Question",
    "name": "What is schema markup for AI citation?",
    "acceptedAnswer": {
      "@type": "Answer",
      "text": "Schema markup is structured data in JSON-LD that describes entities, content type, and publication details. It supports AI citations through entity clarity, evidence exposure, and freshness."
    }
  }]
}</code></pre>
          <p>
            One warning. Markup must mirror what a reader sees. Schema.org vocabulary describes page content, and Averi.ai's guide builds its examples around content that exists on the page [5][8]. FAQPage entries for questions absent from the visible text, or dateModified fields bumped without an edit, damage the trust signal the markup was supposed to create.
          </p>
        </section>

        <section aria-labelledby="source-packs-heading">
          <p class="eyebrow">Evidence</p>
          <h2 id="source-packs-heading">Create source packs and proof gates for every factual claim</h2>
          <p>
            A <a href="/research/source-packs-are-the-new-briefs/">source pack</a> is a grouped evidence set: the claim, the source name, the exact URL, the quoted wording, and the date checked. Keep it in the draft document, not in a separate tracker nobody opens. This article runs on eight named sources — LinkedIn Marketing Solutions, Schema.org, AISO Hub, Averi.ai, Clearscope, Search Engine Land, MarketEngine, and Miragium.
          </p>
          <p>
            Four <a href="/research/marketing-agents-should-stop-workflows-when-proof-is-weak/">proof gates</a> run before publish:
          </p>
          <ol>
            <li><strong>Source exists</strong> — the URL resolves and the page still contains the claim.</li>
            <li><strong>Claim matches source wording</strong> — the sentence does not stretch beyond what the source says.</li>
            <li><strong>Entity name is consistent</strong> — canonical wording from the entity table appears everywhere.</li>
            <li><strong>Schema mirrors visible content</strong> — every FAQPage question appears in the body text.</li>
          </ol>
          <p>
            Three worked examples show how the gates behave in practice.
          </p>
          <p>
            <strong>Business claim.</strong> "71% of respondents say B2B content marketing has grown more important to their organizations over the past year" passes gate two because it repeats LinkedIn Marketing Solutions' framing without converting it into a claim about budgets or headcount [1]. Rewriting it as "71% of companies increased content spend" would fail.
          </p>
          <p>
            <strong>Technical claim.</strong> "Averi.ai's guide provides JSON-LD examples for FAQPage, Article, and Organization schema" passes gate one and two — three named types, three examples, no inflated performance promise attached [8].
          </p>
          <p>
            <strong>Content-structure claim.</strong> "Answer blocks follow four steps: identify the question, write the direct answer, build semantic completeness, organize clearly" maps directly onto MarketEngine's published process [13]. Gate three then checks that "answer block" and "semantic completeness" appear in that exact wording throughout.
          </p>
          <p>
            Run the gates as a review pass, not a mental check. A marketing lead can clear all four on a 2,500-word article in about twenty minutes.
          </p>
        </section>

        <section aria-labelledby="repair-heading">
          <p class="eyebrow">Post-publish</p>
          <h2 id="repair-heading">Run citation gap repair after publication</h2>
          <p>
            <a href="/research/ai-search-citation-gap-repair-workflow/">Citation gap repair</a> is a 30–60 day review that tests whether target prompts cite your page, a competitor, or no source at all. Publishing is the midpoint of the work, not the end. Models re-crawl, re-rank, and re-summarize on their own schedule, and a page that earns no citations in month one often earns them in month three after two structural fixes.
          </p>
          <p>
            Run a 10-prompt audit. Pull the highest-value entries from the prompt-page map, then query each one across ChatGPT, Gemini, Perplexity, and Google AI Overviews. Record three things per prompt: whether your page appeared, which sources were cited instead, and what the engine actually said. Screenshots beat notes here — answers shift week to week.
          </p>
          <p>
            Repair actions:
          </p>
          <ul>
            <li>Add a missing answer block for any prompt the page failed to answer directly.</li>
            <li>Strengthen an entity definition where the engine confused your brand or category.</li>
            <li>Add a source pack where the engine cited a competitor with stronger evidence.</li>
            <li>Update FAQPage or Article schema, including dateModified, after every substantive edit.</li>
          </ul>
          <p>
            Four metrics keep the work reviewable by marketing and PR teams: LLM Citation Rate (share of target prompts citing the page), AI Short-list Rate (share of prompts naming the brand among recommended options), Entity Match Score (consistency between your canonical wording and how engines describe you), and Regional Attribution Accuracy (whether engines attribute the brand to the right markets). Track them monthly. Quarterly is too slow for a surface that changes weekly.
          </p>
        </section>

        <section aria-labelledby="template-heading">
          <p class="eyebrow">Template</p>
          <h2 id="template-heading">Show the finished AI-citable content template</h2>
          <p>
            A page built for citation has nine parts in a fixed order. Reuse the structure across every pillar article and the review pass gets faster each time.
          </p>
          <ol>
            <li><strong>TL;DR</strong> — three to five sentences summarizing the whole page.</li>
            <li><strong>Definition answer block</strong> — 40–70 words under a question-based H3.</li>
            <li><strong>Entity table</strong> — Entity, Type, Canonical wording, Supporting source, On-page placement.</li>
            <li><strong>Prompt-page map</strong> — 12–15 prompts, tagged by answer state.</li>
            <li><strong>Comparison block</strong> — a table or list resolving the main "X vs Y" prompt.</li>
            <li><strong>Implementation checklist</strong> — at least six action steps.</li>
            <li><strong>Schema examples</strong> — Article, Organization, FAQPage in JSON-LD [8].</li>
            <li><strong>Source pack</strong> — every claim with a named source and URL.</li>
            <li><strong>FAQ</strong> — five questions matching the article's core prompts.</li>
          </ol>
          <p>
            Schema sits in three places: Article in the page metadata, Organization near the brand or about block, FAQPage adjacent to the visible FAQ [8].
          </p>
          <p>
            Publishing checklist:
          </p>
          <ul>
            <li>All four proof gates cleared.</li>
            <li>Schema validated and error-free.</li>
            <li>Canonical entity wording consistent across body, headings, and markup.</li>
            <li>Every source link resolves to the claimed content.</li>
            <li>Eight to ten answer blocks present, each 40–70 words.</li>
            <li>FAQPage questions match visible page questions word for word.</li>
            <li>Citation gap audit scheduled for day 30.</li>
          </ul>
        </section>`;

const faqHtml = `            <div class="faq-item">
              <h3>What is schema markup for AI citations?</h3>
              <p>Schema markup is structured data written in JSON-LD using Schema.org vocabulary, describing entities, content type, author, and dates [5]. AISO Hub attributes its effect on AI citations to three factors: entity clarity, evidence exposure, and freshness and accuracy [7]. The markup must describe content that appears on the visible page.</p>
            </div>
            <div class="faq-item">
              <h3>Which entities should a B2B SaaS page define?</h3>
              <p>Five classes: brand, category, problem, product, and method. Clearscope defines an entity as a distinct, singular, well-defined thing or concept that search engines can recognize and understand [9]. Search Engine Land's entity-first approach aligns those definitions with Google's Knowledge Graph so engines resolve your brand to the right category [10].</p>
            </div>
            <div class="faq-item">
              <h3>How long should an answer block be?</h3>
              <p>Forty to seventy words, sitting under a visible question-based H3, answering that question in the first sentence. MarketEngine's four-step method — identify the question, write the direct answer, build semantic completeness, organize clearly — produces blocks a model can lift without surrounding context [13].</p>
            </div>
            <div class="faq-item">
              <h3>What goes into a source pack?</h3>
              <p>The claim, the source name, the exact URL, the quoted wording, and the date checked. This article's pack holds eight sources: LinkedIn Marketing Solutions [1], Schema.org [5], AISO Hub [7], Averi.ai [8], Clearscope [9], Search Engine Land [10], MarketEngine [13], and Miragium [15]. Store it in the draft, not a separate file.</p>
            </div>
            <div class="faq-item">
              <h3>How often should citation gap repair run?</h3>
              <p>Every 30 to 60 days per page, using a 10-prompt audit across ChatGPT, Gemini, Perplexity, and Google AI Overviews. Log which sources each engine cites instead of yours. Then apply the four repair actions: add an answer block, strengthen an entity definition, add a source pack, update the schema.</p>
            </div>`;

const sourcesHtml = `            <div class="source-item" id="source-1">
              <div class="source-number">1</div>
              <div>
                <h3><a href="https://www.linkedin.com/business/marketing/blog/content-marketing/the-ultimate-guide-to-improve-your-b2b-content-marketing-strategy" rel="nofollow noopener noreferrer">LinkedIn Marketing Solutions: The Ultimate Guide to Improve Your B2B Content Marketing Strategy</a></h3>
                <p>LinkedIn Marketing Solutions — B2B content marketing importance statistics and strategic framing.</p>
              </div>
            </div>
            <div class="source-item" id="source-2">
              <div class="source-number">2</div>
              <div>
                <h3><a href="https://schema.org/" rel="nofollow noopener noreferrer">Schema.org</a></h3>
                <p>Schema.org — collaborative community activity creating, maintaining, and promoting schemas for structured data on the internet.</p>
              </div>
            </div>
            <div class="source-item" id="source-3">
              <div class="source-number">3</div>
              <div>
                <h3><a href="https://aiso-hub.com/insights/schema-markup-ai-citations/" rel="nofollow noopener noreferrer">AISO Hub: Schema Markup For AI Citations 2026</a></h3>
                <p>AISO Hub — guide with templates for schema markup types that support AI citations through entity clarity, evidence exposure, and freshness.</p>
              </div>
            </div>
            <div class="source-item" id="source-4">
              <div class="source-number">4</div>
              <div>
                <h3><a href="https://www.averi.ai/blog/schema-markup-for-ai-citations-the-technical-implementation-guide" rel="nofollow noopener noreferrer">Averi.ai: Schema Markup for AI Citations — Technical Implementation Guide</a></h3>
                <p>Averi.ai — JSON-LD implementation examples for Article, Organization, and FAQPage schema types.</p>
              </div>
            </div>
            <div class="source-item" id="source-5">
              <div class="source-number">5</div>
              <div>
                <h3><a href="https://www.clearscope.io/blog/what-is-an-entity-in-SEO" rel="nofollow noopener noreferrer">Clearscope: What Is an Entity and Why Does It Matter for SEO</a></h3>
                <p>Clearscope — entity definition and why entities matter for search engine understanding beyond keywords.</p>
              </div>
            </div>
            <div class="source-item" id="source-6">
              <div class="source-number">6</div>
              <div>
                <h3><a href="https://searchengineland.com/guide/entity-first-content-optimization" rel="nofollow noopener noreferrer">Search Engine Land: Entity-first SEO</a></h3>
                <p>Search Engine Land — how to align content with Google's Knowledge Graph through entity-first optimization.</p>
              </div>
            </div>
            <div class="source-item" id="source-7">
              <div class="source-number">7</div>
              <div>
                <h3><a href="https://marketengine.ai/blogs/index.php/2026/07/13/complete-guide-answer-blocks-semantic-completeness/" rel="nofollow noopener noreferrer">MarketEngine: AI Search Optimization with Answer Blocks</a></h3>
                <p>MarketEngine — complete guide to answer blocks, semantic completeness, and four-step method for AI-extractable content.</p>
              </div>
            </div>
            <div class="source-item" id="source-8">
              <div class="source-number">8</div>
              <div>
                <h3><a href="https://www.miragium.com/tools/ai-answer-block-generator" rel="nofollow noopener noreferrer">Miragium: AI Answer Block Generator</a></h3>
                <p>Miragium — free tool for generating first-draft answer blocks for AI Overviews and answer-engine pages.</p>
              </div>
            </div>`;

const relatedHtml = `            <a class="related-item" href="/research/aeo-geo-operating-system-weekly-workflow/">
              <span>Operating system</span>
              <strong>The AEO/GEO Operating System: A Weekly Workflow from Signal to Proof</strong>
            </a>
            <a class="related-item" href="/research/from-seo-sprints-to-aeo-loops/">
              <span>Migration</span>
              <strong>From SEO Sprints to AEO Loops: How to Migrate Your Content Team</strong>
            </a>
            <a class="related-item" href="/research/source-packs-are-the-new-briefs/">
              <span>Source packs</span>
              <strong>Source packs are the new briefs</strong>
            </a>
            <a class="related-item" href="/research/prompt-page-map-ai-search-site-architecture/">
              <span>Architecture</span>
              <strong>Prompt-page map for AI search site architecture</strong>
            </a>
            <a class="related-item" href="/research/ai-search-citation-gap-repair-workflow/">
              <span>Repair</span>
              <strong>AI Search citation gap repair workflow</strong>
            </a>
            <a class="related-item" href="/research/marketing-agents-should-stop-workflows-when-proof-is-weak/">
              <span>Proof gates</span>
              <strong>Marketing agents should stop workflows when proof is weak</strong>
            </a>`;

const jsonld = `    <script type="application/ld+json">
    {
      "@context": "https://schema.org",
      "@graph": [
        {
          "@type": "Article",
          "@id": "https://gregshevchenko.com/research/how-to-structure-content-for-ai-citation/#article",
          "headline": "How to Structure Content for AI Citation: Schema, Entities, and Answer Blocks",
          "datePublished": "2026-08-10",
          "dateModified": "2026-08-10",
          "author": { "@id": "https://gregshevchenko.com/#person" },
          "publisher": { "@id": "https://gregshevchenko.com/#organization" },
          "mainEntityOfPage": {
            "@type": "WebPage",
            "@id": "https://gregshevchenko.com/research/how-to-structure-content-for-ai-citation/"
          },
          "citation": [
            "https://www.linkedin.com/business/marketing/blog/content-marketing/the-ultimate-guide-to-improve-your-b2b-content-marketing-strategy",
            "https://schema.org/",
            "https://aiso-hub.com/insights/schema-markup-ai-citations/",
            "https://www.averi.ai/blog/schema-markup-for-ai-citations-the-technical-implementation-guide",
            "https://www.clearscope.io/blog/what-is-an-entity-in-SEO",
            "https://searchengineland.com/guide/entity-first-content-optimization",
            "https://marketengine.ai/blogs/index.php/2026/07/13/complete-guide-answer-blocks-semantic-completeness/",
            "https://www.miragium.com/tools/ai-answer-block-generator"
          ]
        },
        {
          "@type": "BreadcrumbList",
          "@id": "https://gregshevchenko.com/research/how-to-structure-content-for-ai-citation/#breadcrumb",
          "itemListElement": [
            { "@type": "ListItem", "position": 1, "name": "Home", "item": "https://gregshevchenko.com/" },
            { "@type": "ListItem", "position": 2, "name": "Research", "item": "https://gregshevchenko.com/research/" },
            { "@type": "ListItem", "position": 3, "name": "How to Structure Content for AI Citation", "item": "https://gregshevchenko.com/research/how-to-structure-content-for-ai-citation/" }
          ]
        },
        {
          "@type": "FAQPage",
          "@id": "https://gregshevchenko.com/research/how-to-structure-content-for-ai-citation/#faqpage",
          "mainEntity": [
            { "@type": "Question", "name": "What is schema markup for AI citations?", "acceptedAnswer": { "@type": "Answer", "text": "Schema markup is structured data written in JSON-LD using Schema.org vocabulary, describing entities, content type, author, and dates. It supports AI citations through entity clarity, evidence exposure, and freshness and accuracy." } },
            { "@type": "Question", "name": "Which entities should a B2B SaaS page define?", "acceptedAnswer": { "@type": "Answer", "text": "Five classes: brand, category, problem, product, and method. Each gets one canonical phrasing, one definition sentence, and one placement in the schema layer." } },
            { "@type": "Question", "name": "How long should an answer block be?", "acceptedAnswer": { "@type": "Answer", "text": "Forty to seventy words, sitting under a visible question-based H3, answering that question in the first sentence. Eight to ten per article is the working range." } },
            { "@type": "Question", "name": "What goes into a source pack?", "acceptedAnswer": { "@type": "Answer", "text": "The claim, the source name, the exact URL, the quoted wording, and the date checked. Store it in the draft document, not a separate tracker." } },
            { "@type": "Question", "name": "How often should citation gap repair run?", "acceptedAnswer": { "@type": "Answer", "text": "Every 30 to 60 days per page, using a 10-prompt audit across ChatGPT, Gemini, Perplexity, and Google AI Overviews. Track LLM Citation Rate, AI Short-list Rate, Entity Match Score, and Regional Attribution Accuracy monthly." } }
          ]
        },
        {
          "@type": "WebSite",
          "@id": "https://gregshevchenko.com/#website",
          "url": "https://gregshevchenko.com/",
          "name": "Gregory Shevchenko",
          "description": "First-party research and notes on AI Search visibility, AEO/GEO, ContentOS, marketing agents, and agentic engineering.",
          "publisher": { "@id": "https://gregshevchenko.com/#organization" }
        },
        {
          "@type": "Person",
          "@id": "https://gregshevchenko.com/#person",
          "name": "Gregory Shevchenko",
          "url": "https://gregshevchenko.com/",
          "jobTitle": "Founder and CEO/CTO",
          "worksFor": { "@id": "https://gregshevchenko.com/#organization" },
          "image": "https://gregshevchenko.com/assets/greg-portrait.jpg",
          "sameAs": [
            "https://humanswith.ai/team/gregory-shevchenko/",
            "https://www.linkedin.com/in/gregory-shevchenko-77567070/",
            "https://github.com/g-shevchenko",
            "https://x.com/GregShevchenko"
          ]
        },
        {
          "@type": "Organization",
          "@id": "https://gregshevchenko.com/#organization",
          "name": "Humanswith.ai",
          "url": "https://humanswith.ai/",
          "foundingDate": "2019-07-15",
          "founder": { "@id": "https://gregshevchenko.com/#person" },
          "description": "Humanswith.ai builds AI-native marketing infrastructure for SMBs that need to be cited by AI answers.",
          "sameAs": [
            "https://www.linkedin.com/company/humanswith-ai/",
            "https://github.com/humanswith-ai",
            "https://medium.com/@humanswith.ai"
          ]
        }
      ]
    }
    </script>`;

const html = `<!doctype html>
<html lang="en">
  <head>
    <meta charset="utf-8" />
    <meta name="viewport" content="width=device-width, initial-scale=1" />
    <title>How to Structure Content for AI Citation: Schema, Entities, and Answer Blocks | Gregory Shevchenko</title>
    <meta name="description" content="A page earns citations when a machine can find a direct answer, resolve entities, confirm sources, and read structured data. The full workflow: prompt-page maps, answer blocks, entity tables, Schema.org markup, source packs, proof gates, and citation gap repair." />
    <meta name="google-site-verification" content="lUo2slSHA_AlyZSlTLxl-nHw9LVV1cpvxijpngcv3U0" />
    <meta name="robots" content="index, follow, max-snippet:-1, max-image-preview:large" />
    <link rel="canonical" href="https://gregshevchenko.com/research/how-to-structure-content-for-ai-citation/" />
    <link rel="alternate" hreflang="en" href="https://gregshevchenko.com/research/how-to-structure-content-for-ai-citation/" />
    <link rel="alternate" hreflang="x-default" href="https://gregshevchenko.com/research/how-to-structure-content-for-ai-citation/" />
    <link rel="icon" href="/assets/favicon-32x32.png" sizes="32x32" type="image/png" />
    <link rel="icon" href="/assets/favicon-16x16.png" sizes="16x16" type="image/png" />
    <link rel="shortcut icon" href="/favicon.ico" />
    <link rel="apple-touch-icon" href="/assets/favicon-180x180.png" />
    <link rel="alternate" type="application/rss+xml" title="Gregory Shevchenko Feed" href="https://gregshevchenko.com/feed.xml" />
    <link rel="me" href="https://humanswith.ai/team/gregory-shevchenko/" />
    <link rel="me" href="https://www.linkedin.com/in/gregory-shevchenko-77567070/" />
    <link rel="me" href="https://github.com/g-shevchenko" />
    <link rel="me" href="https://x.com/GregShevchenko" />
    <meta property="og:type" content="article" />
    <meta property="og:locale" content="en_US" />
    <meta property="og:title" content="How to Structure Content for AI Citation: Schema, Entities, and Answer Blocks | Gregory Shevchenko" />
    <meta property="og:description" content="A page earns citations when a machine can find a direct answer, resolve entities, confirm sources, and read structured data. The full workflow from prompt-page maps to citation gap repair." />
    <meta property="og:url" content="https://gregshevchenko.com/research/how-to-structure-content-for-ai-citation/" />
    <meta property="og:image" content="https://gregshevchenko.com/og/research.png" />
    <meta property="og:image:alt" content="Gregory Shevchenko research cover for how to structure content for AI citation" />
    <meta property="og:image:width" content="1200" />
    <meta property="og:image:height" content="630" />
    <meta property="og:image:type" content="image/png" />
    <meta name="twitter:card" content="summary_large_image" />
    <meta name="twitter:title" content="How to Structure Content for AI Citation: Schema, Entities, and Answer Blocks | Gregory Shevchenko" />
    <meta name="twitter:description" content="Schema, entities, answer blocks, source packs, proof gates, and citation gap repair — the full page-level workflow for AI citability." />
    <meta name="twitter:image" content="https://gregshevchenko.com/og/research.png" />
    <meta name="twitter:image:alt" content="Gregory Shevchenko research cover for how to structure content for AI citation" />
    <link rel="preload" href="/assets/fonts/humans/HumansSerifDisplay-Regular.woff2" as="font" type="font/woff2" crossorigin />
    <link rel="preload" href="/assets/fonts/humans/HumansSerifDisplay-Italic.woff2" as="font" type="font/woff2" crossorigin />
    <link rel="preload" href="/assets/fonts/humans/HumansText-Regular.woff2" as="font" type="font/woff2" crossorigin />
    <link rel="preload" href="/assets/fonts/humans/HumansMono-Regular.woff2" as="font" type="font/woff2" crossorigin />
    <link rel="stylesheet" href="/assets/fonts.css" />
    <!-- Google tag (gtag.js): G-PVJSB1YZ31 -->
    <link rel="stylesheet" href="/assets/consent-v1.css" />
    <script>
      window.dataLayer = window.dataLayer || [];
      function gtag(){dataLayer.push(arguments);}
      window.hwaiGoogleTagLoaded = window.hwaiGoogleTagLoaded || false;
      window.hwaiLoadGoogleTag = window.hwaiLoadGoogleTag || function(){
        if (window.hwaiGoogleTagLoaded) return;
        window.hwaiGoogleTagLoaded = true;
        const script = document.createElement("script");
        script.async = true;
        script.src = "https://www.googletagmanager.com/gtag/js?id=G-PVJSB1YZ31";
        document.head.appendChild(script);
        gtag("js", new Date());
        gtag("config", "G-PVJSB1YZ31");
      };
      (function(){
        const denied = {
          ad_storage: "denied",
          ad_user_data: "denied",
          ad_personalization: "denied",
          analytics_storage: "denied",
          functionality_storage: "granted",
          security_storage: "granted",
          wait_for_update: 500
        };
        let hasAnalyticsConsent = false;
        try {
          const stored = JSON.parse(localStorage.getItem("gregshevchenko_com_consent_v1") || "null");
          if (stored && stored.analytics === true) {
            denied.analytics_storage = "granted";
            hasAnalyticsConsent = true;
          }
        } catch {}
        gtag("consent", "default", denied);
        if (hasAnalyticsConsent) {
          const load = () => window.hwaiLoadGoogleTag();
          if ("requestIdleCallback" in window) {
            requestIdleCallback(load, { timeout: 3000 });
          } else {
            window.addEventListener("load", () => setTimeout(load, 1200), { once: true });
          }
        }
      })();
      gtag("set", "ads_data_redaction", true);
    </script>
    <script defer src="/assets/consent-v1.js"></script>
    <!-- end Google tag -->
    ${jsonld}
    <link rel="stylesheet" href="/assets/site-chrome-v1.css" />
    <link rel="stylesheet" href="/assets/greg-research-article-v1.css" />
  </head>
  <body class="note-source-page">
    <header class="site-header">
      <nav class="nav" aria-label="Primary navigation">
        <a class="brand" href="/">
          <span class="brand-avatar" aria-hidden="true">
            <span class="brand-avatar-inner">
              <img src="/assets/derived/gregory-shevchenko-headshot-68.jpg" srcset="/assets/derived/gregory-shevchenko-headshot-68.jpg 1x, /assets/derived/gregory-shevchenko-headshot-136.jpg 2x" alt="Gregory Shevchenko headshot in the site navigation" width="68" height="68" decoding="async" fetchpriority="high" />
              <img class="brand-mark" src="/assets/derived/greg-shevchenko-mark-circle-34.png" srcset="/assets/derived/greg-shevchenko-mark-circle-34.png 1x, /assets/derived/greg-shevchenko-mark-circle-68.png 2x" alt="Gregory Shevchenko monogram logo" width="34" height="34" />
            </span>
          </span>
          <span>Gregory Shevchenko</span>
        </a>
        <button class="nav-backdrop" type="button" aria-label="Close navigation" tabindex="-1"></button>
        <div class="nav-links" id="primary-nav-links">
          <a href="/">Main</a>
          <a href="/research/" aria-current="page">Research</a>
          <a href="/writing/">Writing</a>
          <a href="/speaking/">Speaking</a>
          <a class="nav-email-item" href="mailto:g@humanswith.ai">Email me</a>
        </div>
        <span class="nav-separator" aria-hidden="true">|</span>
        <a class="nav-email" href="mailto:g@humanswith.ai" title="Email Gregory Shevchenko">Email me</a>
        <button class="nav-toggle" type="button" aria-label="Open navigation" aria-expanded="false" aria-controls="primary-nav-links">
          <span class="nav-menu-icon" aria-hidden="true"></span>
        </button>
      </nav>
    </header>

    <main>
      <section class="hero" aria-label="Article introduction">
        <nav class="breadcrumbs" aria-label="Breadcrumb">
          <a href="/">Main</a>
          <span>/</span>
          <a href="/research/">Research</a>
          <span>/</span>
          <span>How to Structure Content for AI Citation</span>
        </nav>
        <p class="kicker">Research essay · Published 10 August 2026</p>
        <h1 id="article-top">How to Structure Content for AI <em>Citation</em>: Schema, Entities, and Answer Blocks</h1>
        <p class="lede">
          A page earns citations when a machine can find a direct answer, resolve the entities in it, confirm the source behind each claim, and read structured data that mirrors the visible text. This guide covers the full workflow: prompt-page maps, answer blocks, entity tables, Schema.org markup, source packs, proof gates, and citation gap repair.
        </p>

        <div class="hero-meta" aria-label="Article metadata">
          <div class="meta-card">
            <div class="meta-label">Author</div>
            <div class="meta-value">Gregory Shevchenko</div>
          </div>
          <div class="meta-card">
            <div class="meta-label">Primary intent</div>
            <div class="meta-value">Show content teams how to structure a page for AI citability with schema, entities, answer blocks, source packs, and proof gates</div>
          </div>
          <div class="meta-card">
            <div class="meta-label">Source base</div>
            <div class="meta-value">LinkedIn Marketing Solutions, Schema.org, AISO Hub, Averi.ai, Clearscope, Search Engine Land, MarketEngine, Miragium</div>
          </div>
          <div class="meta-card">
            <div class="meta-label">Best use</div>
            <div class="meta-value">Practical implementation companion to the AEO/GEO Operating System — page-level citability template with nine-part structure</div>
          </div>
        </div>

        <div class="summary-card">
          <h2>What to cite from this page</h2>
          <p>
            Citation is a structure problem, not a keyword problem. A page earns citations when a machine can extract, verify, and quote its content.
          </p>
          <ul>
            <li>Six-part workflow: prompt-page map, answer state taxonomy, entity map, schema layer, source packs, proof gates.</li>
            <li>Answer blocks: 40–70 words, self-contained, under question-based H3, eight to ten per article.</li>
            <li>Entity optimization: 12+ named entities with canonical wording, five classes (brand, category, problem, product, method).</li>
            <li>Schema stack: Article (page), Organization (brand), FAQPage (questions) — markup must mirror visible content.</li>
            <li>Four proof gates: source exists, claim matches source, entity name consistent, schema mirrors visible content.</li>
            <li>Citation gap repair: 30–60 day audit across ChatGPT, Gemini, Perplexity, Google AI Overviews.</li>
          </ul>
        </div>
      </section>

      <article>
${bodySections}

        <section aria-labelledby="faq-heading">
          <p class="eyebrow">FAQ</p>
          <h2 id="faq-heading">Questions this page should answer</h2>
          <div class="faq">
${faqHtml}
          </div>
        </section>

        <section aria-labelledby="sources-heading">
          <p class="eyebrow">Source trail</p>
          <h2 id="sources-heading">Sources and related canonicals</h2>
          <div class="source-list">
${sourcesHtml}
          </div>
        </section>

        <section aria-labelledby="related-heading">
          <p class="eyebrow">Read next</p>
          <h2 id="related-heading">Continue the thread</h2>
          <div class="related-list">
${relatedHtml}
          </div>
        </section>
      </article>
    </main>

    <footer class="site-footer">
      <span class="footer-identity"><img class="footer-mark" src="/assets/derived/greg-shevchenko-mark-24.png" srcset="/assets/derived/greg-shevchenko-mark-24.png 1x, /assets/derived/greg-shevchenko-mark-48.png 2x" alt="Gregory Shevchenko monogram logo" width="24" height="24" /><span class="footer-identity-text"><span class="footer-name">© 2026 Gregory Shevchenko.</span><span class="footer-role">Founder of&nbsp;<a class="footer-company-link" href="https://humanswith.ai/">Humanswith.ai</a></span><a class="footer-email" href="mailto:g@humanswith.ai">Email&nbsp;me</a></span></span>
      <span class="footer-links" aria-label="Footer links"><div class="footer-row footer-row--site"><span class="footer-label">Site</span><nav class="footer-internal" aria-label="Site links"><a href="/start/">Start here</a><a href="/notes/">Notes</a><a href="/writing/">Writing</a><a href="/research/">Research</a><a href="/speaking/">Speaking</a></nav></div><div class="footer-row footer-row--elsewhere"><span class="footer-label">Elsewhere</span><nav class="footer-external" aria-label="External links"><a href="https://www.linkedin.com/in/gregory-shevchenko-77567070/">LinkedIn.com</a><a href="https://medium.com/@GregShevchenko">Medium.com</a><a href="https://github.com/g-shevchenko">GitHub.com</a><a href="https://x.com/GregShevchenko">X.com</a><a href="https://vc.ru/gshevchenko">VC.ru</a><a href="https://t.me/gshevchenko">t.me</a></nav></div></span>
    </footer>

    <script>
      (() => {
        const toggle = document.querySelector(".nav-toggle");
        const links = document.getElementById("primary-nav-links");
        const backdrop = document.querySelector(".nav-backdrop");
        if (!toggle || !links || !backdrop) return;
        const setOpen = (open) => {
          document.body.classList.toggle("nav-open", open);
          toggle.setAttribute("aria-expanded", String(open));
        };
        toggle.addEventListener("click", () => setOpen(!document.body.classList.contains("nav-open")));
        backdrop.addEventListener("click", () => setOpen(false));
        links.querySelectorAll("a").forEach((link) => link.addEventListener("click", () => setOpen(false)));
        document.addEventListener("keydown", (event) => {
          if (event.key === "Escape") setOpen(false);
        });
      })();
    </script>
  </body>
</html>`;

writeFileSync('projects/gregshevchenko-com/research/how-to-structure-content-for-ai-citation/index.html', html);
console.log('Written', html.length, 'chars');