O

The logic of classic SEO cracked in 2026. Google AI Overviews — powered by Gemini 3 since January 27 — now appears in roughly half to three-quarters of U.S. searches, and nearly half of B2B buyers already use AI tools to research suppliers before they even click on a website. The stakes have therefore changed: the question is no longer where you rank on the results page, but whether you are included in the single summarized answer the buyer actually reads.

And here comes the uncomfortable part. According to an industry analysis from early 2026 examining nearly four million AI Overview citations, only a fraction of the cited pages now appear in the classic top 10 — compared with roughly three-quarters a year earlier. In other words, ranking and citation have separated. A lower-ranking but clearly structured page tied to a credible author can now regularly outperform a high-domain-authority but anonymous and unstructured competitor.

This guide translates exactly these two levers — semantic HTML structure and schema markup — into concrete, immediately deployable steps for B2B software companies. The framework I will walk you through is Miklós Róth’s SICT Protocol, which breaks algorithmic authority building into four engineering layers.

In the age of AI Overviews, content is not ranked — entities are cited. Our task is not to make the machine “find” the company, but to make it recognize the company as a clear, machine-readable authority.

01 — THE REALITY: What Did Gemini 3 Change in AI Overviews?

Gemini 3 (live since January 27, 2026) provides synthesized answers for most searches and cites only a handful of sources for each answer. The standard for visibility is no longer ranking position, but whether your content consists of machine-readable, standalone answer blocks supported by clear entity signals (schema).

In the previous model, the goal was to be one of the ten blue links. In the new model, the user often never reaches the links: the answer is assembled at the top of the SERP inside AI Overviews. This is called zero-click search, and it is especially dangerous in B2B because the buyer becomes “trapped” inside the summary during the most valuable stage of decision preparation.

The good news: the system is predictable. AI prefers sources that are a) clearly structured, b) tied to a verifiable entity (company + author), and c) provide a complete answer on their own to a specific question. All three conditions are technical — which means they can be intentionally engineered.

02 — THE PARADIGM SHIFT: Why Is Domain Authority No Longer the Deciding Factor?

Because AI search engines do not rank pages; they retrieve and verify entities. According to 2026 analyses, the correlation between domain authority and AI citation has shrunk to a negligible level — the decisive factors have become author-entity credibility and the cleanliness of the schema chain.

This has a concrete — and uncomfortable — consequence for the B2B SaaS sector: the generic “guide” type of content written by an anonymous content marketer, which had been the industry’s default output for years, is now uncitable. AI reduces its risk when generating an answer; therefore, it reaches for sources backed by an identifiable, real expert.

In 2026, E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) is therefore not a marketing slogan, but a practical filter. The three technical signals that support it are:

• Verifiable authorship — content tied to a concrete Person entity, with external validating links (sameAs: LinkedIn, Wikidata, the author’s own professional site).
• Entity clarity — the company, author, and product are defined consistently and identically across every platform.
• Citation discipline — original data, real sources, precise claims. AI now treats “fluent but evidence-free” text with a punitive logic.

03 — STRUCTURE: How Should You Structure Content with Semantic HTML?

Build the content around question-based H2/H3 headings, and place a direct, independently meaningful answer under each heading. AI often extracts a single section — so every block must be a complete truth on its own.

This is the first layer of SICT: Structure. Semantic HTML is not decoration — it is the map Gemini uses to navigate. The practical rules are:

1. The heading is the question; the paragraph is the answer

Every H2/H3 should be phrased the way a buyer would type it into search (“How can X be integrated with Salesforce?”), and directly below it there should be a complete 40–60-word answer. This “definition-first” pattern is the shared entry point for featured snippets and AI Overviews.

2. Real semantic elements

Use an <article>, <section>, and <h1>–<h3> hierarchy with a single H1; use <ul>/<ol> for lists and a real <table> for comparisons. Numbered steps and tables are disproportionately often included in AI answers.

3. Standalone blocks

Run this test: if you copy a single paragraph out of context, does it stand on its own? If not, AI will not highlight it either. Avoid internal references such as “as we saw above” inside key answer blocks.

04 — INFORMATION: Which Schemas Should You Use — and in What Order?

Use only JSON-LD format, in priority order: Organization (the entity foundation) → Person (author authority) → SoftwareApplication / Article → FAQPage → HowTo. Microdata and RDFa are practically obsolete — do not use them.

This is the second layer of SICT: Information — explicit, machine-readable signaling about what is what. The importance of structured data has increased, but it is worth being honest: Google itself says that schema is not mandatory for AI Overviews, and there is no single “magic schema” either. Schema does not replace weak content or missing authority — it clarifies what already exists. It provides a real advantage where strong content is paired with strong entity signaling.

The FAQPage Trap — and Why You Should Still Use It

In August 2023, Google withdrew FAQ rich result visibility for most websites (the visual “accordion” now typically appears only for government and healthcare sites). Many SEOs therefore wrote off FAQPage schema. AI engines, however, did the opposite: they rely on it more heavily because question-answer pairs provide exactly the format from which direct answers can be generated. Two conditions, however, are set in stone: the questions must be visibly present on the page, and the answers must match the markup exactly. (The FAQ section of this article is built this way as well.)

January 2026 deprecations: Google deprecated several rarely used types (e.g., Practice Problem, Dataset, Sitelinks Search Box), but you can safely build your strategy around the core high-load types — Organization, Article, Person, Product, Review.

05 — COHESION: How Do You Build the Entity Chain?

Connect the entities in a single @graph array using unique @id references: Organization ↔ Person ↔ SoftwareApplication ↔ Article. This way, AI does not see disconnected text fragments, but a coherent knowledge graph — “this is the company, it employs this expert, who wrote this article about this software.”

This is the third layer of SICT: Cohesion. The most common mistake companies make is publishing separate, disconnected schema blocks without @id — and Google cannot connect the entities. The solution is @graph + @id anchoring, and externally, sameAs (LinkedIn, Wikidata, the author’s own professional subdomain), which connects the author to the external knowledge graph.

06 — TRANSFORMATION: What Does This Look Like When Assembled? (JSON-LD Template)

This is the fourth layer of SICT: Transformation — when structure and signaling become a single deployable asset. The template below shows a connected @graph for a B2B SaaS product page/article. Fill in the fields marked with // REPLACE using real data — never leave invented URLs or IDs inside it.

schema.jsonld · @graph

<!-- Place in the <head> or below the content, in a separate <script> block -->
<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@graph": [
    {
      "@type": "Organization",
      "@id": "https://example-saas.com/#org",        // REPLACE
      "name": "Example SaaS Ltd.",
      "url": "https://example-saas.com/",
      "sameAs": [
        "https://www.linkedin.com/company/..."     // REPLACE
      ]
    },
    {
      "@type": "Person",
      "@id": "https://example-saas.com/#author",
      "name": "Miklós Róth",
      "jobTitle": "Algorithmic Authority Building Expert",
      "worksFor": { "@id": "https://example-saas.com/#org" },
      "sameAs": [
        "https://rothaiconsulting.com",
        "https://rothcomplexity.org",
        "https://www.linkedin.com/in/..."          // REPLACE
      ]
    },
    {
      "@type": "SoftwareApplication",
      "@id": "https://example-saas.com/#product",
      "name": "Example Platform",
      "applicationCategory": "BusinessApplication",
      "operatingSystem": "Web, iOS, Android",
      "offers": {
        "@type": "Offer",
        "price": "29.00",                          // REPLACE with real price,
        "priceCurrency": "EUR"
      },
      "provider": { "@id": "https://example-saas.com/#org" }
    },
    {
      "@type": "Article",
      "@id": "https://example-saas.com/blog/aio#article",
      "headline": "How Can Your B2B Software Dominate Google AI Overviews?",
      "author": { "@id": "https://example-saas.com/#author" },
      "publisher": { "@id": "https://example-saas.com/#org" },
      "about": { "@id": "https://example-saas.com/#product" },
      "datePublished": "2026-05-24",
      "dateModified": "2026-05-24"
    },
    {
      "@type": "FAQPage",
      "@id": "https://example-saas.com/blog/aio#faq",
      "mainEntity": [{
        "@type": "Question",
        "name": "Is schema mandatory for AI Overviews?",
        "acceptedAnswer": {
          "@type": "Answer",
          "text": "It is not mandatory, but it is strongly recommended..."  // exactly = the visible text on the page
        }
      }]
    }
  ]
}
</script>

Validation before deployment — always: Google Rich Results Test, the Schema.org Markup Validator, and after publication, the Google Search Console > Enhancements report. The most common errors are: missing @context, @graph elements without @id, and incorrect date formatting.

07 — MEASUREMENT: What Should You Track, and When Does It Take Effect?

Indexing a new schema typically takes 2–6 weeks, and citation behavior usually begins to shift within 4–12 weeks after that (Perplexity responds faster because it reads at query time; purely training-based models respond more slowly). Patience and measurement are required — this is not an instant switch.

The right metrics for the 2026 AI era — not old-style position tracking:

• AI Overview impression share — the share of impressions coming from AI appearances (GSC).
• Citation frequency — how often your content is cited for target questions.
• Schema error rate — GSC > Enhancements report.
• Rich result CTR delta — the change in click-through rate after implementing schema.

08 — EXECUTION: 14-Day Tactical Checklist

1. Audit: run your 20 most important pages through the Rich Results Test and mark missing schemas.
2. Deploy Organization schema with sameAs references — this is the entity foundation.
3. Create real author pages, each with Person schema and sameAs validation.
4. Convert the headings of key pages into question form, with standalone answer blocks beneath them.
5. Connect the entities in a single @graph using @id references.
6. Add SoftwareApplication schema to product pages (category, OS, price).
7. Add visible FAQs to key pages with FAQPage schema — the text must match exactly.
8. Validate, publish, submit the sitemap, then measure in GSC for 4–12 weeks.

FAQ

Is structured data mandatory for Google AI Overviews?

No, it is not mandatory, and there is no single “magic schema” — Google confirms this as well. At the same time, it is strongly recommended: schema works as a translation layer between AI and your content, so the system does not have to guess, but can understand your entities from explicit signals. It provides the greatest advantage when paired with strong content and real author authority.

What is the most important schema type for B2B SaaS?

Organization is the foundation because it defines the brand as an entity. Person (author authority) and SoftwareApplication (the product as an entity) build on top of it. Connect these three in a single @graph — this creates the strongest leverage.

Is FAQPage schema still worth using after 2023?

Yes. In 2023, Google withdrew visual FAQ displays for most websites, but AI engines still rely heavily on question-answer pairs for direct answer generation. The condition: the questions must be visible on the page, and the answers must match the markup exactly.

How long does it take for schema to affect AI citation?

Indexing a new schema typically takes 2–6 weeks, and citation behavior usually begins to shift within 4–12 weeks after that. Perplexity reacts faster because it reads at query time; training-based systems react more slowly. Measure patiently in the Search Console Enhancements report.

Methodological note: the industry ratios mentioned in this guide (AI Overviews coverage, separation of ranking and citation, indexing time windows) are based on public SEO/AEO analyses from 2025–2026 and may change with platform updates. The authoritative source for schema specifications is Google Search Central. The JSON-LD template is exemplary — before deployment, replace all placeholder data with real information and validate it.