Schema Markup That Gets You Into AI Answers

Schema markup is a block of JSON-LD code that tells machines exactly what your page is: an article, a product, an organization. In 2026 it still earns Google rich results and it keeps your brand entity unambiguous, which matters for AI engines too. What it does not do, based on the best current data, is directly buy you AI citations. This post covers what to implement, one working example, and where the honest limits are.

What JSON-LD is and why entity clarity matters

JSON-LD (JSON for Linked Data) is a small <script> block you add to a page. It describes the page’s content using the shared schema.org vocabulary, so a crawler does not have to guess. Google recommends JSON-LD over other formats because it lives in one self-contained block, separate from your visible HTML, which makes it easy to generate and maintain at scale.

Two audiences read this data. Google uses it to qualify pages for rich results: the star ratings, prices, breadcrumbs, and article cards you see in search. And the broader machine ecosystem uses it to resolve entities: to confirm that the “Gantra” on this page is a marketing software company, with this exact URL and logo and these social profiles.

That second job is bigger than it sounds. Our own name collides with a pharmaceutical product of the same name. Without a clear, consistent Organization entity everywhere, a machine deciding what “Gantra” means has to guess. Entity consistency is also a core theme of generative engine optimization: AI engines cross-check what they know about a brand before naming it, and ambiguity reads as uncertainty.

The schema types that pay off for a startup site

You do not need fifty types. You need the small set that matches pages you actually have, implemented completely. For a typical startup site that is six:

Two 2026 realities to plan around. First, Google stopped showing FAQ rich results for all sites on May 7, 2026, after restricting them to government and health sites back in 2023. The FAQPage markup is still valid and causes no errors, but it no longer buys a visible SERP feature. Keep writing real FAQ content, because question-and-answer blocks are exactly what AI engines extract; just do not expect dropdowns in Google anymore.

Second, Google has been pruning the whole gallery. In June 2025 it retired seven structured data types at once, including Course Info, Claim Review, and Special Announcement, saying they were rarely used and no longer added value. The lesson: invest in the boring core types above, not in exotic ones that may vanish next.

A working Organization example

Here is a complete Organization block, the single highest-value piece of markup for a startup fighting for entity clarity. It goes on your homepage inside a script tag:

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Organization",
  "name": "Gantra",
  "url": "https://gantra.so",
  "logo": "https://gantra.so/logo.png",
  "description": "AI marketing agents that run SEO, GEO, and content distribution for startups.",
  "foundingDate": "2026",
  "sameAs": [
    "https://www.linkedin.com/company/gantra",
    "https://x.com/gantra"
  ]
}
</script>

The field doing the quiet heavy lifting is sameAs. It links your entity to your profiles on other platforms, so machines can connect the dots and confirm it is the same company everywhere. Use the exact name and description you use on LinkedIn, Crunchbase, and your directories. The markup should confirm what the rest of the web already says about you, not invent a new version.

For blog posts, the same idea applies with BlogPosting: headline, author, datePublished, dateModified, and the publisher pointing back to your Organization. Most frameworks and CMS plugins generate this for you; the only real work is checking the output once.

What the Rich Results Test checks (and what it does not)

Google’s Rich Results Test is the official validator. Paste a live URL or a code snippet and it tells you three things: which rich result types it detected on the page, whether the syntax parses, and whether required fields are missing. It also flags warnings for optional fields that would strengthen the markup, and it can render the page as Googlebot sees it, which catches JavaScript problems.

Know its limits. The tool validates eligibility, not outcomes. Google is explicit that a passing test does not guarantee your page will appear as a rich result; that decision depends on quality signals and query context. Also, it only covers Google’s supported rich result types. Valid schema.org markup that Google has no feature for (like plain Organization data) can show as “no items detected” while still doing its entity job. For full vocabulary checks, the schema.org validator at validator.schema.org is the companion tool.

One practical note for startups on JavaScript-heavy stacks: put the JSON-LD in your server-rendered HTML. Many AI crawlers do not execute JavaScript, so markup injected client-side may exist for Googlebot and be invisible to everyone else.

The honest limit: schema and AI citations

Here is the part most schema guides skip. The direct causal effect of structured data on AI citations is unproven, and the best experiment so far found roughly nothing. Ahrefs tracked 1,885 pages that added JSON-LD between August 2025 and March 2026, against 4,000 control pages. The result: citations moved minus 4.6% in Google AI Overviews (a small but real decline), plus 2.4% in AI Mode, and plus 2.2% in ChatGPT, with the last two indistinguishable from random noise. Their conclusion was blunt: adding schema produced no major uplift on any platform.

The same study explains why schema keeps getting oversold. In their earlier correlation data, AI-cited pages were nearly three times more likely to have JSON-LD than non-cited pages. But correlation is not cause: sites with good schema tend to be technically well-run sites with strong content, and that is what earns the citations. This mirrors the framing from our GEO guide: structured data helps entity disambiguation and classic SERP features, is cheap, and is worth doing. It is not an AI citation lever, and anyone selling it as one is ahead of the evidence.

So sequence your effort accordingly. What moves AI visibility is answer-shaped content backed by real statistics, plus mentions on sites you do not control. Schema is the tidy plumbing underneath, in the same “ship it, expect little” bucket as llms.txt.

Mistakes that can cost you a manual action

Schema is low-risk only while it stays honest. Google’s structured data policies are clear on the lines, and crossing them can trigger a manual action that strips your rich result eligibility site-wide. The common failures:

• Marking up content that is not on the page. Google’s rule is blunt: do not mark up anything a reader cannot see. If your JSON-LD claims a FAQ or a review that a human cannot find in the rendered page, that is spam by definition.
• Fake or self-written reviews. Ratings have to come from actual users. Adding AggregateRating with an invented 4.9 score is the classic startup shortcut, and the classic penalty.
• Misrepresenting who you are. The policies ban using structured data to impersonate people or organizations, or to misstate your affiliation or purpose.
• Stale fields. Prices, availability, and dates that drift out of sync with the visible page can invalidate the markup. Generate it from the same data source that renders the page.

The good news inside the penalty: a structured data manual action removes rich results, but Google states it does not affect how the page ranks in web search. Still not a trade worth making.

Schema hygiene is a maintenance task, which is why it belongs in an automated loop rather than a launch checklist. The daily SEO audit in Gantra checks structured data along with the rest of your technical layer, and it sits next to a GEO audit that tracks whether AI engines actually mention you. Both audits run on the free plan, as part of a 12-agent marketing team. Auto-publishing covers LinkedIn and X; Reddit, Hacker News, Medium, and Dev.to come out as drafts with one-click share.

The takeaway

Implement the boring six: Organization and WebSite on the homepage, BlogPosting and BreadcrumbList on content, Product and Offer where you sell, FAQPage only if it costs you nothing. Make the markup mirror your visible page exactly, validate it in the Rich Results Test, and keep your entity identical across the web. Expect entity clarity and classic rich results in return, not AI citations; the Ahrefs data says that lever does not exist yet. Then spend the time you saved on extractable answers and third-party mentions, which do move AI visibility. If you want the audits running daily without thinking about it, Gantra starts free.