What Structured Data for AI Actually Still Does

Summarise this blog

ChatGPT Perplexity Claude Grok

Key Takeaways

• Structured data for SEO isn’t the same play it was a year ago. Google fully deprecated FAQ rich results in May, HowTo rich results have been gone since 2023, and Sitelinks Search Box was retired back in January, part of a consistent pattern of pruning underused types.
• Roughly 31 schema types still carry active rich result support in Google Search, with Organisation, Article, Product, and LocalBusiness performing strongest, particularly where the schema ties to a real intent signal like product availability, event timing, or recipe detail.
• A controlled 2025 Search Engine Land experiment found a page with well-implemented schema appeared in an AI Overview, while an otherwise identical page with no schema failed to even get indexed, real evidence that structured data still earns its place for AI visibility specifically.
• Bare-minimum schema barely moves the needle. A generic Article or Organization block with only the required fields gives close to zero AI citation advantage. The real lift comes from completing the optional, content-specific properties most sites skip.
• FAQPage and HowTo no longer produce a Google rich result, but the markup still gets crawled by Bing, PerplexityBot, and RAG-based AI systems, so it still supports structured data for AI comprehension even without the SERP payoff.
• Schema still has to match visible content exactly, JSON-LD is still the preferred format, and a quarterly audit catches the stale dates and deprecated types that quietly undermine everything else.

Schema advice from even a year ago is already partly wrong, and most sites haven’t caught up. Structured data for SEO used to have a simple pitch: mark up your FAQs and your how-to steps, and Google would reward you with extra SERP real estate. That pitch stopped being true, and Google’s been quietly pruning underused rich result types for a couple of years running now. Here’s what actually still works, what only matters for AI crawlers now, and where structured data for AI search picks up where Google’s classic rich results left off.

What Changed With FAQ and HowTo Rich Results

Google removed HowTo rich results entirely back in September 2023, first restricting them to desktop, then dropping them altogether within weeks. FAQ rich results followed the same pattern more slowly. Google narrowed FAQ display to government and health sites in August 2023, then finished the job in May this year, adding a deprecation notice to its own developer documentation confirming FAQ rich results no longer appear in Google Search at all, for any site. Sitelinks Search Box was quietly retired back in January, another entry in the same pattern.

Neither FAQ nor HowTo markup is broken or invalid. Both remain valid Schema.org types, Google has said plainly that unused structured data doesn’t cause problems for Search, and there’s no need to strip the code out. What’s gone is the visible payoff, the expandable question dropdown and the step-by-step display that used to justify adding the markup in the first place.

Here’s the part worth understanding properly. The schema was never what did the ranking work. The content underneath it did. Removing the rich result doesn’t remove the value of writing genuinely well-structured, question-led content. It just removes Google’s willingness to give that structure extra visual space in the SERP.

Which Schema Types Still Earn Real Results

Not every schema type took the same hit. Roughly 31 schema types still carry active rich result support in Google Search. The strongest performers among them are tied to a concrete, specific intent signal rather than a generic content label.

• Organisation. The entity foundation. Ties a site to a known, disambiguated entity in the Knowledge Graph, opening the door to knowledge panels and stronger brand-specific search features.
• Article and BlogPosting. Still supports thought leadership and editorial content, helping both Google and AI summary engines display author, date, and context correctly.
• Product. Still drives pricing, availability, and rating display directly in search results, one of the strongest remaining investments for e-commerce.
• LocalBusiness. Strong where a physical location genuinely matters, address, hours, and location signals feed both local rankings and AI-driven local answers.
• FAQPage and HowTo. No longer produces a Google rich result, but the markup still gets crawled by Bing, PerplexityBot, and various AI retrieval systems, so it still contributes to structured data for AI comprehension even without the SERP payoff.

The practical shift is simple. Stop applying FAQPage or HowTo schema purely to chase a rich result that no longer exists. Keep using them where the content genuinely is a set of questions or a genuine process, because the underlying comprehension value for AI systems hasn’t gone anywhere, only the Google display incentive has.

Why Completeness Beats Coverage for AI Citation

Here’s the nuance most schema checklists skip entirely. Adding a generic Article or Organization block with only the required properties filled in gives close to zero advantage when it comes to AI citation. The actual lift comes from completing the optional, content-specific properties most implementations leave blank, author credentials, detailed offer data on Product schema, full review counts and ratings, precise event timing.

There’s real evidence behind this. A controlled Search Engine Land experiment in 2025 tested two otherwise identical pages, one with well-implemented schema, one with none. The schema page appeared inside an AI Overview. The page with no structured data at all didn’t even get indexed. That’s a strong practical argument for treating schema completeness as core infrastructure, not a box-ticking exercise squeezed in after the content ships.

The common failure modes worth watching for: Product schema missing offer data, Article schema missing an author, LocalBusiness schema missing an address, any of which suppresses rich result eligibility outright. Beyond that, stale dateModified values and lingering deprecated types (Sitelinks Search Box among them) quietly signal an unmaintained site to both Google and AI crawlers alike.

Implementation Rules That Still Matter

A handful of foundational rules haven’t changed, and they matter more than any single schema type.

Use JSON-LD, Not Microdata or RDFa

Google and every major AI engine, Google itself, Bing, Perplexity, and ChatGPT, consistently favour JSON-LD over older Microdata or RDFa formats, since it sits separately from the visible HTML rather than tangled inside it. Place it in the page head or immediately before the closing body tag, and generate it from the same data source that powers the visible content, hardcoded schema next to dynamic content is exactly how the two drift out of sync.

Mark Up Only What’s Visible and Consistent

Schema data has to match what’s actually on the page, word for word in substance. Hidden or misleading markup risks a manual penalty and actively breaks AI extraction rather than helping it, since a mismatch between what the schema claims and what a crawler actually reads undermines trust in the whole page.

Audit on a Quarterly Schedule, Not Just at Launch

Run pages through Google’s Rich Results Test and the Schema Markup Validator on a recurring cadence, not only when the schema first goes live. Stale dates, missing required properties, and deprecated types accumulate quietly, and a quarterly audit catches them before they cost eligibility.

Enrich With Entity Data Where It’s Genuine

Where it’s accurate, embed entity signals inside schema, sameAs links to verified social profiles and a Wikipedia entry where one exists, author bios, links to supporting datasets. This feeds the Knowledge Graph directly and helps both Google and AI systems establish real authority and context around a page, rather than treating it as an anonymous block of text.

Building a Structured Data Strategy That Holds Up

The sites getting this right in the current environment aren’t the ones stacking schema onto every page. They’re the ones applying the right type to the right page, filling in the properties most competitors leave blank, keeping markup honestly matched to visible content, and treating structured data for AI comprehension as separate from chasing a Google rich result that, for FAQ and HowTo at least, no longer exists.

Getting this consistent across an entire site takes ongoing technical discipline most in-house teams don’t have the bandwidth to sustain. Working with an SEO agency that treats schema as infrastructure rather than an afterthought closes that gap properly.

At Primal, our AI search work builds structured data around what actually earns visibility today, not what worked two years ago. As an AI SEO agency in Bangkok brands rely on for the long haul, we keep schema, content, and technical SEO aligned so pages perform under Google’s current rules and get cited across AI Overviews, ChatGPT, and Perplexity alike.

Explore with AI

ChatGPT Key takeaways Explain principles Create action plan Future perspectives

Perplexity Key takeaways Explain principles Create action plan Future perspectives

Claude Key takeaways Explain principles Create action plan Future perspectives

Grok Key takeaways Explain principles Create action plan Future perspectives