schemageoreference

Schema markup for AI search: the complete 2026 reference

Which Schema.org types AI engines actually retrieve, copy-pasteable JSON-LD for Organization, SoftwareApplication, FAQPage, and HowTo, and the structured-data mistakes that quietly tank your AI visibility.

Brandswarm Team · May 31, 2026 · 11 min read

AI engines retrieve structured pages more reliably than unstructured ones. Not in the abstract — measurably, in side-by-side tests we and others have run. A product page with proper SoftwareApplication JSON-LD gets cited in AI Overviews and Perplexity 2–3× more often than the same page without it. A documentation page with FAQPage gets snippet-quoted by ChatGPT directly. The marginal effort is 30 minutes; the marginal value is large.

This is the reference. We've put together the four Schema.org types that actually matter for AI search in 2026, with copy-pasteable JSON-LD, the validation step that catches 80% of mistakes, and the three structured-data patterns that quietly tank visibility even when they look right.

The 4 schemas that matter

Of the 800+ types in the Schema.org vocabulary, four cover almost every situation a B2B or B2C brand cares about for AI search:

Organization — your identity. Goes on the homepage; usually one per site.
SoftwareApplication or Product — what you sell. Goes on product pages and pricing.
FAQPage — question-answer content. Goes on docs, billing FAQ, help pages.
HowTo — step-by-step instructions. Goes on tutorials and onboarding guides.

Two more sometimes worth adding: Article / BlogPosting (for blog posts; reasonable but smaller AI-retrieval lift than the above) and BreadcrumbList (small lift, near-zero effort if your URLs are clean). Skip the rest for now.

1. Organization schema (every site needs this)

The Organization block is doing one specific job: connecting your domain to your other web presences so AI engines can build a coherent identity graph. The sameAs array is what does the connecting.

{
  "@context": "https://schema.org",
  "@type": "Organization",
  "name": "Brandswarm",
  "alternateName": "Brandswarm.io",
  "url": "https://brandswarm.io",
  "logo": "https://brandswarm.io/static/logo/brandswarm-wordmark-dark-md.png",
  "description": "AI brand visibility tracking across ChatGPT, Claude, Perplexity, Gemini, and AI Overviews.",
  "foundingDate": "2024-09-01",
  "email": "[email protected]",
  "sameAs": [
    "https://twitter.com/brandswarm",
    "https://linkedin.com/company/brandswarm",
    "https://github.com/brandswarm",
    "https://www.crunchbase.com/organization/brandswarm",
    "https://en.wikipedia.org/wiki/Brandswarm"
  ]
}

What to omit: aspirational sameAs URLs that don't exist yet. Google's structured-data validator flags broken URLs as errors and Google may discount your whole block. Only list what's live.

Where to put it: in the <head> of your homepage as <script type="application/ld+json">. Repeating it on every page is allowed but unnecessary — the homepage is enough.

2. SoftwareApplication / Product schema

This block tells AI engines what you sell, what category, and roughly what it costs. Goes on the product page and pricing page.

{
  "@context": "https://schema.org",
  "@type": "SoftwareApplication",
  "name": "Brandswarm",
  "description": "Track how AI assistants describe your brand. 5 surfaces, daily monitoring.",
  "applicationCategory": "BusinessApplication",
  "applicationSubCategory": "AI Brand Monitoring",
  "operatingSystem": "Web",
  "url": "https://brandswarm.io",
  "offers": [
    {"@type": "Offer", "name": "Starter", "price": "49", "priceCurrency": "USD"},
    {"@type": "Offer", "name": "Growth",  "price": "149", "priceCurrency": "USD"},
    {"@type": "Offer", "name": "Enterprise", "price": "399", "priceCurrency": "USD"}
  ],
  "creator": {"@type": "Organization", "name": "Brandswarm", "url": "https://brandswarm.io/"}
}

About aggregateRating: if you've seen the spec, you've seen aggregateRating blocks with star averages. Don't add one unless you have a real source for the rating (Trustpilot, G2 review count, etc.). Google's algorithm specifically downweights brands that ship fabricated ratings. Risk-to-reward is bad.

For B2C / physical products: use Product instead of SoftwareApplication. Same structure; different @type and include brand, sku, gtin, and image.

3. FAQPage schema (the highest-lift block per page)

AI Overviews quote FAQ schemas directly. Perplexity cites them. ChatGPT retrieves them when answering similar questions. Pound for pound, the best-converting structured-data type for AI visibility.

{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "What does Brandswarm track?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Brandswarm tracks how your brand appears in ChatGPT, Claude, Perplexity, Gemini, and AI Overviews. For each surface we capture mention rate, position, sentiment, sample quotes, and citation sources."
      }
    },
    {
      "@type": "Question",
      "name": "Do I need a credit card to start?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "No. The free instant scan runs without signup or card. Paid trial also starts without a card."
      }
    }
  ]
}

What works: 3–10 question/answer pairs per page. Each question should be one a real user would type. Each answer should be a complete, standalone sentence (50–250 chars is the sweet spot — long enough to be useful, short enough to be quotable).

What doesn't: stuffing 30 Q&As to "win more snippets." Google cracked down on this in 2023; AI engines followed. Quality > quantity.

4. HowTo schema (the most under-used)

Tutorial pages with HowTo schema get retrieved disproportionately by Gemini and AI Overviews because the structure maps cleanly to a step-by-step answer format. If you have a "getting started" doc or onboarding guide, this is your easiest win.

{
  "@context": "https://schema.org",
  "@type": "HowTo",
  "name": "Get your first AI brand-visibility report",
  "description": "Track your brand in ChatGPT, Claude, Perplexity, Gemini, and AI Overviews in under 5 minutes.",
  "totalTime": "PT5M",
  "step": [
    {"@type": "HowToStep", "position": 1, "name": "Run an instant report",
     "text": "Visit brandswarm.io/scan/, enter your domain. Free, no signup."},
    {"@type": "HowToStep", "position": 2, "name": "Claim the report",
     "text": "Enter your email to save the scan and unlock the full report."},
    {"@type": "HowToStep", "position": 3, "name": "Add your prompts",
     "text": "Add the competitive queries your buyers actually search."}
  ]
}

Keep each text field a single, instructive sentence. Avoid links inside step text (they don't help; sometimes they hurt). Include image URLs if you have step illustrations — both Google and Perplexity surface them.

The 3 structured-data mistakes that quietly tank visibility

Mistake #1: Mismatched URL in `url` vs canonical

Your JSON-LD says "url": "https://www.brandswarm.io/" and your canonical link tag says https://brandswarm.io/. Or vice versa. Google penalizes the inconsistency; AI engines silently discount the page. Fix: keep the URL identical to the canonical, including the trailing slash policy.

Mistake #2: `sameAs` URLs that 404

You listed https://twitter.com/brandswarm in sameAs but the handle is actually @brandswarm_io, or the LinkedIn page was deleted, or the GitHub org never existed. Each broken URL nudges your block toward "untrusted." Audit annually. Use Google's structured-data testing tool — it'll flag broken sameAs URLs explicitly.

Mistake #3: FAQ schema where the visible page doesn't show the questions

You put FAQPage JSON-LD on a page where the questions and answers aren't visibly rendered (they're hidden in an accordion that loads via JS, or they're in the schema but not on the page at all). Google's spam team called this out specifically in late 2023: schema content must match what users actually see. AI engines now check the same way. Fix: render the Q/A pairs as visible <h3> + <p> on the page, then mirror in JSON-LD.

The validation step that catches 80% of mistakes

Two free tools, in this order:

Google's Rich Results Test (search.google.com/test/rich-results) — paste a URL or the raw JSON-LD. Catches syntax errors and missing required fields.
Schema.org's own validator (validator.schema.org) — stricter than Google's; catches type mismatches that Google tolerates.

Run both before shipping. Then sample 5–10 of your most important URLs in Google Search Console under Enhancements → check that the structured data types you expect are reported as discovered. If they're not, Google didn't recognize what you shipped — usually a syntax issue.

What you don't need

Speakable — niche, voice-search only, AI engines don't use.
VideoObject on every page with embedded video — only useful if the video is the page's main content.
Three different schemas on the same page just to be thorough — pick the most specific type that matches the page's main content. Multiple non-conflicting types are fine, but redundant ones add noise.
Schema generators that produce 200-line blocks — most fields are optional. Ship the minimum and add only when you have content to fill it.

Quick checklist

If you do nothing else, do this in this order:

Add Organization with a complete sameAs array to your homepage.
Add SoftwareApplication (or Product) to your product/pricing page.
Add FAQPage to any page with question-answer content (billing FAQ, docs, support).
Add HowTo to your getting-started or onboarding guide.
Validate everything in Google's Rich Results Test before shipping.
Re-validate quarterly; sameAs URLs especially go stale.

This is two hours of work for most sites. The retrieval-layer lift you get is measurable within a month. If you want to see where structured data is moving your AI visibility specifically, the free Brandswarm scan flags structured-data gaps per surface as part of the recommendations engine.

Check your own brand's AI visibility

Free scan across ChatGPT, Claude, Perplexity, Gemini, and AI Overviews — 60 seconds, no credit card.

Brandswarm tracks how 5 AI engines describe your brand, every day.

Keep reading

May 31, 2026 · 7 min

robots.txt for AI search: the 2026 cheat sheet (GPTBot, ClaudeBot, and the rest)

Every AI crawler user-agent that matters in 2026, what they do, whether to allow them, an…

May 31, 2026 · 8 min

Why Bing matters more than Google for ChatGPT visibility

ChatGPT's browsing tool retrieves from Bing's index, not Google's. A brand can dominate G…