The DDS SEO Magnet System Masterclass

Why This Class Exists

This is the core class of the Academy. Every other masterclass — Antigravity Onramp, MCP Server Forge, Sovereign RAG — was built with the system documented here. So this page does something the others do not: it teaches the system and is simultaneously a working example of it. Every tactic described below is in use on this very page, and the final section points to the proof.

The Shift: From Ranking to Being Cited

Generative engine optimization optimizes content so AI answer engines cite it as a source, where traditional SEO optimizes for a ranking position in a list of links. The distinction is decisive: a page can rank first on Google and never be cited by Claude, ChatGPT, Gemini, or Perplexity, because answer engines reward different structural signals than ranking algorithms do.

For two decades the goal was a position on a results page. A user typed a query, saw ten links, and clicked two or three. In 2026 a growing share of those users ask an AI instead, and the AI returns one synthesized answer crediting a handful of sources. The user gets the answer and moves on. The traffic that once came from a rank now comes — or does not come — from a citation.

This is not a reason to abandon SEO. The fundamentals still hold: quality content, topical authority, fast pages, and E-E-A-T help both ranking and citation. But citation rewards a layer of structure that classic SEO never required, and a few classic tactics — keyword density above all — actively hurt it. The SEO Magnet System keeps every ranking fundamental and adds the citation layer on top.

How AI Engines Pick Sources

An answer engine does not paste your query into a search box. It breaks the question into sub-queries (query fan-out), retrieves candidate passages for each, and synthesizes one answer citing the sources that were clearest, most specific, and safest to repeat. It cites statements, not pages — so the unit you optimize is the quotable, self-contained claim, not the document.

Three properties make a passage citable. It is specific: concrete numbers and named entities give the model something defensible to attribute. "Reduces review cycles 40%" is citable; "improves efficiency" is not. It is self-contained: a passage that stands alone, with no unresolved pronouns or references to earlier text, can be lifted verbatim. It is safe to repeat: sourced, factual, non-promotional statements carry low risk for the engine, which is why citation correlates with sources and negatively with hype.

This is why the same study that found statistics and citations help also found that brand mentions correlate with AI citation far more strongly than backlinks do. Answer engines are not counting links into your page; they are assessing whether a specific statement on your page is trustworthy enough to repeat in front of a user. Structure your pages so the answer to each likely question is sitting there, specific and self-contained, ready to lift.

The System: Nine Layers

The SEO Magnet System has nine layers across three groups: the on-page citation stack (schema, meta, speakable boxes, GEO tactics, components, answer capsules, entity authority), the site-level discovery stack (robots.txt, llms.txt, llms-full.txt), and a maintenance cadence. Every layer is present on every Academy page; the discovery and maintenance layers operate site-wide.

The rest of this class walks the layers that carry the most weight: schema done right, the GEO tactics and their measured effects, answer capsules, entity authority, the discovery stack, and the maintenance discipline that keeps all of it from drifting. The page you are reading implements all nine.

Layer 1: Schema Done Right

In 2026 structured data is an AI-trust and entity-verification signal, not a SERP display trigger. After Google's March 2026 core update, generic schema with only its required fields gives almost no citation lift; the value comes from completing the optional properties so an answer engine can attribute a claim to your brand with confidence. The SEO Magnet System ships ten complete JSON-LD blocks per page.

The ten: a primary content type (TechArticle, Article, Course, or SoftwareApplication), plus HowTo, Course, FAQPage, BreadcrumbList, Person, Organization, WebSite, WebPage with speakable, and ImageObject. The exact set flexes by page type, but the discipline does not: every block fills its optional properties.

Two non-negotiable schema rules. First, FAQPage must match the on-page accordion exactly — same questions, same answers — and it belongs only on genuinely question-and-answer content, never on marketing copy disguised as questions, which the March 2026 update demoted at scale. Second, dateModified is a separate value from datePublished and gets updated on every content change; both Google and answer engines weight freshness, and a stale modified date is a silent signal that your content is old.

Validate every block. All JSON-LD must parse cleanly, the {% schema %} block name must stay under twenty-five characters, and the FAQ accordion item count must equal the FAQPage Question count. These are mechanical checks; run them before every publish.

The Five GEO Tactics That Work

The peer-reviewed Princeton GEO study (Aggarwal et al., presented at ACM KDD 2024) tested nine content tactics across ten thousand queries and validated the results on a live answer engine. Five measurably raised citation rates: adding statistics, citing sources, adding quotations, fluency optimization, and an authoritative voice. Combining them outperforms any single tactic, and the combination of statistics plus fluency produced the largest effect in the study's tests.

These are not folklore. They are the findings of the first large-scale academic study of AI citation, and they are the spine of Layer 4 of the system.

Notice this very section: it opens with an answer capsule, names the study and its venue, and carries specific percentages. It is applying its own advice. There is also a compounding effect beyond the table — entity density. More named entities per paragraph (people, products, standards, tools) correlates with higher citation, because engines use entity recognition to gauge how information-rich a passage is. Name things precisely: "nomic-embed-text," "GOTS," "Atelier 2.1.1" — not "the model," "our certification," "the theme."

The Four Anti-Patterns That Fail

The same study found four tactics that do nothing or actively hurt AI citation. Keyword stuffing underperforms the baseline by about 10 percent. Promotional tone correlates with citation at roughly minus 26 percent. Content padding and pure persuasive language add nothing. Knowing what fails matters as much as knowing what works, because three of these four are exactly the habits that classic conversion copywriting trains into you.

Answer Capsules: The Highest-Leverage Change

An answer capsule is a 40-60 word self-contained answer placed at the top of a content section, before the elaboration. In 2026 citation audits it was the single strongest structural correlate of being cited, present in roughly 72 percent of cited pages. It must stand alone — no unresolved pronouns, no reference to earlier text — so an engine can lift it verbatim as a complete answer.

This is distinct from the page-level Quick Answer box. The Quick Answer summarizes the whole page in fifty words; an answer capsule summarizes one section in fifty words. Every major section on this page opens with one — including the section you are reading. That is deliberate: the pattern is cheap to apply, works on any content type, and gives an answer engine a clean, quotable unit for each question your page addresses.

The discipline that makes a capsule work is self-containment. "As discussed above, it improves things" is useless to an engine that retrieved only this passage. "An answer capsule is a 40-60 word self-contained answer placed atop a section" can be cited with zero surrounding context. Write each capsule as if it is the only sentence the model will see, because for a given sub-query, it might be.

Entity Authority and Share of Model

Share of Model is the 2026 metric for how often and how authoritatively AI systems cite a brand — the answer-engine analogue of share of voice. It is driven less by backlinks, which correlate weakly with AI citation, and more by entity clarity: consistent, verifiable facts about the brand across schema, llms.txt, and on-page content, so a model can identify and trust the entity it is about to cite.

The practical work is consistency. The same brand name, address, founding date, certifications, and social profiles should appear identically in your Organization and Person schema, in your llms.txt, and on your About and Contact pages. Contradictions — a different phone number here, a stale certification there — reduce a model's confidence that it has correctly identified the entity, which lowers citation. One canonical set of facts, repeated everywhere, is the goal.

Author authority is the same principle applied to people. A Person schema block with jobTitle, a portfolio url, and knowsAbout fields establishes a credentialed, identifiable author — the E-E-A-T signal that answer engines weight when deciding whether a claim is trustworthy enough to repeat. This page's author block does exactly that, naming Robert McCullock's role, portfolio, and areas of expertise.

Layer 8: The Site-Level Discovery Stack

On-page optimization is wasted if AI crawlers never reach the page. Three site-level files govern that: robots.txt decides which AI bots may crawl at all, llms.txt gives them a curated reading list pointing to your best content, and llms-full.txt hands them that content in one fetch. Together they are the upstream half of discoverability — the plumbing that lets the on-page work matter.

These are not page templates; they are theme-level files that operate site-wide. The DDS implementation runs robots.txt at v3.0.0, llms.txt at v2.0.0, and llms-full.txt at v1.1.0, all current to May 2026. The sections below show the architecture and the decisions behind each.

robots.txt: The Crawler Gate

robots.txt controls whether AI crawlers can reach your pages, which is upstream of every on-page signal. A visibility-first policy allows reputable AI search, retrieval, and training bots so the brand can appear in their answers, while blocking aggressive scrapers and SEO-data resellers that consume bandwidth without returning visibility. On Shopify it is a Liquid template that iterates the platform's safe defaults and adds AI directives.

The strategy for a brand that wants citations is to allow generously. Every AI answer that mentions the brand is free visibility; the cost of allowing is negligible bandwidth, and the cost of blocking is missing from every Claude, ChatGPT, Perplexity, Gemini, and Apple Intelligence answer for years. So the DDS policy allows the full reputable AI bot set and reserves blocking for scrapers that sell data to competitors without sending any visibility back.

Get the vendor frameworks right. The major AI vendors run multiple bots with different jobs, and you want all of them: a training crawler, a search-index bot, and a user-fetch bot. For Anthropic that is ClaudeBot, Claude-SearchBot, and Claude-User; for OpenAI, GPTBot, OAI-SearchBot, and ChatGPT-User. Blocking the search-index bot specifically removes you from that engine's search answers, which is the opposite of the goal.

# AI crawler directive — allow the full vendor framework
User-agent: ClaudeBot
Allow: /
Allow: /products/*.json
Allow: /collections/*.json
Disallow: /admin
Disallow: /cart*
Disallow: /checkout

User-agent: Claude-SearchBot
Allow: /
# Blocking this would remove the brand from Claude search answers.

User-agent: OAI-SearchBot
Allow: /
# OpenAI: blocking this removes you from ChatGPT search answers.

One Shopify-specific move: iterate the platform's robots.default_groups to preserve its safe defaults, then selectively allow what helps discoverability — the /policies/ pages (refund, shipping, privacy are E-E-A-T trust signals) and the /products/*.json and /collections/*.json endpoints that AI agents parse for structured catalog data.

llms.txt and llms-full.txt

llms.txt is a Markdown file at the domain root giving AI systems a curated reading list that points past navigation and marketing to your canonical pages. llms-full.txt is its high-fidelity companion, containing the complete Markdown of your most-cited pages so an agent gets full context in one fetch. Research indicates agents fetch llms-full.txt at roughly twice the rate of llms.txt where both exist.

Be honest about adoption. Anthropic officially supports llms.txt; OpenAI, Google, and Perplexity have not committed, and crawler logs suggest most do not fetch it yet — multiple analyses found no measurable traffic change after implementation. The correct framing is forward-compatible infrastructure at near-zero cost, not a guaranteed win. Ship it because it is cheap insurance and it cannot hurt, not because it has proven results it has not yet earned.

On Shopify, the platform will not serve a file at the domain root, so the pattern is a page template with the theme wrapper stripped, plus a URL redirect. Use dynamic Liquid so the file self-updates as collections, products, and posts change:

{% layout none %}
# Design Delight Studio

> One-paragraph factual brand summary. No hyperbole, no pricing,
> no competitor references — the spec prohibits all three.

## Collections
{% for collection in collections -%}
- [{{ collection.title }}]({{ shop.url }}{{ collection.url }})
{% endfor %}

The maintenance trap is real here. A static llms.txt drifts the moment you ship a new page — the file silently keeps describing the old site. The fix is twofold: use dynamic Liquid for everything that can be (collections, products, blog) so those sections never drift, and put the static sections (brand facts, certifications, the masterclass node list) on the quarterly-audit checklist so they get reconciled on a schedule rather than forgotten.

What Got Deprecated in 2026

Several long-standing SEO features were removed in 2026, and building against them now signals a stale page. FAQ rich results ended May 7, 2026. Dataset and the Sitelinks Search Box were deprecated in January 2026. Google deprecated Crawl-delay in February 2026. The schema types stay valid where noted, but their SERP display is gone — so build them for AI citation value, not for visual treatments that no longer appear.

The most important one to internalize is FAQ. The accordion that used to appear beneath a search result is gone, and its Search Console reporting retires in June 2026. But FAQPage the schema type is not deprecated — Google confirmed unused structured data causes no problems, and Claude, ChatGPT, Gemini, and Perplexity still parse it during answer extraction. The move is to keep FAQPage as an AI-citation asset on genuinely Q&A-shaped content and stop expecting a SERP accordion.

Track the vocabulary too. Schema.org reached v30 stable on March 19, 2026; new properties appear and old ones get marked superseded on a regular cadence. The point is not to chase every release but to know that the structured-data landscape moves, which is exactly why the next section exists.

Keeping It Updated: The Maintenance Cadence

An SEO system goes stale unless it is maintained on a schedule. The cadence has two parts: a per-publish checklist that fires every time a page ships, and a quarterly audit that reconciles the whole site. The per-publish step sets dateModified and adds the new page to the hub, llms.txt, and llms-full.txt; the quarterly step re-validates schema, reconciles node lists across surfaces, and re-checks the crawler taxonomy against a dated deprecation-watch list.

This is not optional polish — it is the layer that prevents the silent drift every multi-page site accumulates. The DDS discovery files demonstrated the failure mode in practice: after several masterclasses shipped, the llms.txt node list still described an earlier version of the Academy and even named a theme version that was no longer live. Nothing was broken, but the file quietly stopped telling the truth. A cadence catches that before it compounds.

Keep a dated deprecation-watch list alongside the system so removed features are caught deliberately rather than discovered by accident. The list in the section above is the current one; it carries a date because it will be wrong eventually, and knowing when it was last true is part of maintaining it.

This Page As Proof

This masterclass is its own reference implementation: every tactic it teaches is in use on the page you are reading, and all of it is verifiable in view-source. The page ships ten completed JSON-LD schemas, an answer capsule atop every major section, statistics throughout, a sticky table of contents, a breadcrumb, and a sixteen-item FAQ accordion that matches its FAQPage schema exactly. Nothing here is theoretical.

Open the page source and you can confirm each claim. The ten application/ld+json blocks are in the head region: TechArticle, HowTo, Course, FAQPage, BreadcrumbList, ImageObject, SoftwareApplication, Organization, WebSite, and WebPage with a speakable selector list. The Organization block carries the full address, contact point, founding date, and sameAs profiles. The Person block names the author's role, portfolio URL, and expertise. That completeness is the entity-authority layer, not decoration.

Read the prose and you can see the GEO tactics. Each section opens with a 40-60 word answer capsule. Concrete numbers appear roughly every paragraph — +41%, 72%, −26%, May 7. Sources are named: the Princeton GEO study and its KDD 2024 venue, the dated Google deprecation. The voice is plain and the tone is non-promotional, because the research says hype suppresses citation. The page is not describing the system from the outside; it is the system, running.

The Per-Page Build Checklist

Every Academy page runs through one checklist before it ships. It groups into four passes: structured data, GEO content, on-page components, and the discovery and maintenance hooks. The checklist is mechanical on purpose — it turns the system from a set of principles into a repeatable gate that produces the same citation advantage on every page, regardless of who builds it.

Use this as the final audit before publishing any page built with the SEO Magnet System.

Frequently Asked Questions

These sixteen questions cover the system end to end: what GEO is, which tactics the research validates, the 2026 deprecations, the discovery files, and how to measure results. Each answer here matches this page's FAQPage structured data exactly, so an answer engine extracting either the visible accordion or the schema gets the same canonical response.