/* VALDEX — Citation policy */

const UPDATED_C = "May 18, 2026";

function CitationPage() {
  return (
    <LegalShell
      title="Citation policy"
      updated={UPDATED_C}
      intro={
        <React.Fragment>
          <p>
            This document explains how <strong>VALDEX LLC</strong> measures, reports, and
            sources the citation, ranking, and revenue claims we make about our work — on
            this website, in our audit documents, in client reports, and in any case study
            we publish.
          </p>
          <p>
            We publish it because the easiest place to lie in an advertising agency is in
            the numbers, and we want it to be obvious how ours are produced.
          </p>
        </React.Fragment>
      }
      sections={[
        { id: "definitions", h: "1. Definitions we use", body: (
          <React.Fragment>
            <p>Wherever you see these terms — here, in a report, or in marketing material — they mean exactly this:</p>
            <h3>Citation share</h3>
            <p>
              The percentage of times a brand is named in the citation list of a tracked
              answer, across a fixed prompt set, across a fixed list of engines, over a
              defined window. <span className="term">"+482% citation share"</span> means the absolute
              percentage grew from <code>X</code> to <code>5.82·X</code>, not that the brand grew "by 482" of anything.
            </p>
            <h3>Cited first</h3>
            <p>
              The brand appears as the first reference in the citation list of an answer.
              Different engines order citations differently; we record the order as the
              engine returns it.
            </p>
            <h3>Retrieval-corpus inclusion</h3>
            <p>
              A binary measure: is a page indexed in an engine's live retrieval corpus
              (yes/no), as determined by direct probes against the engine.
            </p>
            <h3>Tracked prompt set</h3>
            <p>
              A finite, written-down list of prompts agreed with each client at the start of
              the engagement. We don't change the list secretly — adding or removing a prompt
              requires a one-line note in the next Tuesday status email and the client's
              consent.
            </p>
          </React.Fragment>
        )},
        { id: "sampling", h: "2. How we sample", body: (
          <React.Fragment>
            <ul>
              <li><strong>Engines.</strong> Our default tracked set is ChatGPT, Perplexity, Google AI Overviews, Claude, and Gemini. Copilot is sampled on request. We add new engines when their U.S. share crosses 1%.</li>
              <li><strong>Cadence.</strong> Twice weekly — typically Tuesday and Friday, between 09:00 and 17:00 MT — through an automated harness running with fresh sessions each time.</li>
              <li><strong>Region.</strong> All probes originate from a U.S. residential IP unless the engagement specifies otherwise.</li>
              <li><strong>Personalization.</strong> We disable account-level personalization where the engine offers a setting; we use signed-out sessions otherwise. We document which prompts use which mode.</li>
              <li><strong>Variance.</strong> Generative engines are stochastic. We report citation share over a rolling 14-day window to smooth single-sample noise.</li>
            </ul>
          </React.Fragment>
        )},
        { id: "reporting", h: "3. How we report", body: (
          <React.Fragment>
            <p>
              Every claim about a client's performance — in their dashboard, in the Tuesday
              note, in the Friday metrics drop, or in a case study — is sourced from the
              automated harness, the analytics platform we set up, or the ad-platform export.
              We will, on request, hand over the raw data behind any number we publish.
            </p>
            <p>
              When we round, we round to two significant figures. When we annualize, we say
              so. When we exclude an outlier we say so and why. When a number changes between
              one report and the next, the change is noted at the top of the new report.
            </p>
          </React.Fragment>
        )},
        { id: "claims", h: "4. Claims on this website", body: (
          <React.Fragment>
            <p>
              Numbers shown on this website (case studies, headline stats, percentages on
              the home page) reflect actual engagements — not benchmarks, simulations, or
              composites. Where a logo is shown, we have written permission from the client
              to do so. Where a logo is withheld, the engagement is either active or the
              client asked us not to name them publicly; the numbers are unchanged.
            </p>
            <p>
              <strong>Median lift figures.</strong> When we describe medians across our portfolio
              (e.g. <span className="term">"median citation share growth +312%"</span>) we mean the
              median across every client engagement that has been running for at least 90
              days at the time of publication. We do not exclude clients that performed
              poorly to inflate the median. The full underlying dataset is available to
              prospective clients on request, under NDA.
            </p>
          </React.Fragment>
        )},
        { id: "third", h: "5. Third-party citations", body: (
          <React.Fragment>
            <p>
              Part of GEO work is earning citations from independent third-party sources
              (publications, reviewer sites, comparison pages). Our outreach is on these
              principles:
            </p>
            <ul>
              <li>We never pay a publisher for editorial coverage and call it earned. Paid placements are disclosed in the report as paid.</li>
              <li>We don't impersonate clients or fabricate reviewer accounts.</li>
              <li>We don't generate fake reviews, fake testimonials, or fake third-party endorsements.</li>
              <li>We don't engage in private blog networks, link farms, or any tactic that violates the published guidelines of the engines we work with.</li>
            </ul>
            <p>
              If a published citation later turns out to violate a guideline, we remove it
              and report the removal in the next status note. We have done this twice.
            </p>
          </React.Fragment>
        )},
        { id: "ai", h: "6. How we use AI in the work", body: (
          <React.Fragment>
            <p>
              We are an AI-native agency. Models draft long-form pages, summarize research,
              and power our internal agents. We follow three rules:
            </p>
            <ul>
              <li><strong>Every published claim is verified by a human against a primary source</strong> before the page ships. We do not publish unverified model output.</li>
              <li><strong>We do not generate content that misrepresents itself</strong>. Synthetic case studies, made-up customer quotes, and fictional testimonials are out of scope, and we will not produce them on request.</li>
              <li><strong>We disclose AI involvement where it matters.</strong> Articles that use models for drafting are still authored, edited, and stand behind a named human. We will note model-assisted production in a methodology section when a piece is heavily synthesized from research.</li>
            </ul>
          </React.Fragment>
        )},
        { id: "errors", h: "7. Corrections", body: (
          <p>
            If we publish a number on this website or in a public report that turns out to
            be wrong, we correct it within five business days, note the change at the top
            of the document, and — if the error materially overstated our results — notify
            anyone we directly communicated the wrong number to.
          </p>
        )},
        { id: "questions", h: "8. Questions about a specific claim", body: (
          <React.Fragment>
            <p>
              If you see a number on this website and want to know how we got it, write to
              us. We'll show you the raw data and the harness configuration that produced it,
              redacting only client identifiers we don't have permission to share.
            </p>
            <p>
              <strong>VALDEX LLC</strong><br />
              1309 Coffeen Ave, Ste 1200<br />
              Sheridan, WY 82801<br />
              <a href="mailto:methodology@valdexai.com">methodology@valdexai.com</a>
            </p>
          </React.Fragment>
        )},
      ]}
    />
  );
}

ReactDOM.createRoot(document.getElementById("root")).render(<CitationPage />);