Schema Markup for AI Search: What Still Works in 2026

Structured data has become the most oversold item on the average SEO proposal. It is cheap to add, it looks technical, and for years it produced a visible result you could screenshot. That last part has quietly stopped being true for two of the most popular types.

This guide covers what Google actually still renders, what changed and when, what schema realistically does for AI search, and the specific markup worth implementing. If you want the wider context on how AI search changed the job, start with our pillar guide on GEO vs AEO vs SEO.

What Google Actually Deprecated, and When

Two dates matter, and both come from Google’s own changelog rather than from commentary about it.

FAQ rich results: Google added a deprecation notice to the FAQ structured data documentation, stating the feature would no longer appear in Google Search starting 7 May 2026. In June it removed the documentation entirely, along with the FAQ search appearance filter, the rich result report, and Rich Results Test support.

HowTo rich results: gone considerably longer. Google reduced HowTo visibility on mobile in August 2023, then confirmed that as of 13 September 2023, Google Search no longer shows How-to rich results on desktop, making the type deprecated outright.

If your agency has shown you a proposal line item for FAQ schema in the last few months, it is worth asking what result they expect it to produce.

Should You Remove FAQ and HowTo Markup?

No, and this is where a lot of teams overcorrect. Google’s own guidance on unused structured data is unambiguous: it does not impact your visibility or rankings, and it is fine to maintain the markup for other services or systems that use it.

That last clause is the interesting one. Google’s rich result is one consumer of your structured data. It is not the only one. Other search engines, aggregators, and the crawlers behind AI assistants all read the same markup. Removing valid, accurate schema because one consumer stopped rendering it is throwing away information for everybody else.

What should change is your expectation, and your reporting. If someone is still showing you FAQ rich result impressions as a win, those numbers stopped existing.

We keep FAQPage markup on our own articles for exactly this reason, including the one you are reading. It buys us nothing in Google Search today. It still describes the questions this page answers, in a format any machine can parse without guessing.

Does Schema Get You Cited by AI?

Here is the honest answer, and it is less satisfying than what you will be told elsewhere: there is no published evidence that structured data is a direct ranking or citation factor for AI engines. Nobody has demonstrated that adding JSON-LD causes a model to cite a page it would otherwise have skipped.

What schema plausibly does is help with entity resolution — connecting “Digital Rhodium” the string to a specific organization with a specific location, phone number and service set. When a model has to decide whether a claim can be attributed to a known entity, being unambiguous about who you are is not nothing. But that is a mechanism argument, not a measured effect, and we would rather say so than sell you a number.

If an agency tells you schema markup will get you into ChatGPT answers, ask what evidence they have. The honest version is that clear, well-evidenced, well-structured content is what gets cited, and schema is a small supporting part of being unambiguous.

The Schema a Business Actually Needs

For the overwhelming majority of businesses, the useful set is small. Implement these properly before worrying about anything exotic.

Organization — once, site-wide. Your name, URL, logo, contact details and sameAs links to the profiles you actually control. This is the single most valuable block for entity resolution, and it is the one most often missing.

LocalBusiness (or a subtype like ProfessionalService) — only where you have a real place of business a customer can be served at. This is worth stating clearly because it is regularly abused: marking up locations you do not physically occupy is a fast route to a suspended Google Business Profile.

Service with areaServed — the correct type for a service-area business. If you serve a city you have no address in, this is what you want instead of LocalBusiness.

BreadcrumbList — still renders, still cheap, still helps describe site structure.

Article — on editorial content, with a real author and dates. One article entity per page. If your SEO plugin already emits BlogPosting, do not hand-add a second Article block; two article entities for one URL is a contradiction, not a bonus.

Blocks That Validate

These are complete and correct. Replace the values, keep the structure.

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Organization",
  "name": "Your Company",
  "url": "https://example.com/",
  "logo": "https://example.com/logo.png",
  "telephone": "+1-000-000-0000",
  "email": "hello@example.com",
  "sameAs": [
    "https://www.linkedin.com/company/yourcompany",
    "https://www.facebook.com/yourcompany"
  ]
}
</script>

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Service",
  "serviceType": "Search Engine Optimization",
  "name": "SEO Services in Springfield, IL",
  "provider": {
    "@type": "Organization",
    "name": "Your Company",
    "url": "https://example.com/"
  },
  "areaServed": {
    "@type": "City",
    "name": "Springfield",
    "containedInPlace": { "@type": "State", "name": "IL" }
  }
}
</script>

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "ProfessionalService",
  "name": "Your Company",
  "url": "https://example.com/",
  "telephone": "+1-000-000-0000",
  "address": {
    "@type": "PostalAddress",
    "streetAddress": "123 Real Street",
    "addressLocality": "Springfield",
    "addressRegion": "IL",
    "postalCode": "62701",
    "addressCountry": "US"
  }
}
</script>

Note what the second and third blocks do differently. The Service block claims a service area and no address. The ProfessionalService block claims a physical location. Use the first for cities you cover and the second only for places you genuinely occupy.

How to Check Yours in Ten Minutes

1. View source and search for ld+json. Count the blocks. If you find two article-type entities on one page, that is your first fix.
2. Paste each block into a JSON validator. A block that does not parse does nothing at all, and this fails silently — nothing in Search Console tells you.
3. Run the page through Google’s Rich Results Test for the types it still supports, and Schema.org’s validator for everything else.
4. Check the values are true. An address you do not occupy, an author who does not exist, a rating you invented — these are the ones that cause real problems.
5. Check Search Console’s Enhancements reports for the types Google still reports on. FAQ is no longer among them.

Most schema problems we find are not sophisticated. They are a plugin and a theme both emitting an organization block with different phone numbers, or a JSON block broken by a stray quote from a page builder.

The Mistakes We See Most

Marking up locations you do not occupy. The most damaging one, because it risks the Google Business Profile that local visibility depends on.

Two article entities on one page. Usually an SEO plugin emitting BlogPosting while someone hand-added Article. Pick one.

Schema that contradicts the visible page. An author in the markup who is not the byline on screen, or a business name that does not match the footer. Structured data is a description of the page, not an alternative version of it.

Invented aggregate ratings. Self-serving review markup is explicitly against Google’s guidelines, and it is one of the easier things for them to detect.

Treating markup as a substitute for content. Marking up a vague answer produces a vague answer that is well described. Google has confirmed its spam policies apply to generative AI responses in Search as well, so structured thin content is still thin content.

Frequently Asked Questions