Schema.org for GEO in 2026: Should You Still Mark Up After Google’s FAQ Deprecation?

Google has deprecated FAQ rich results. But LLMs still parse Schema.org. Here’s the markup that matters for GEO in 2026, and the numbers to ignore.

Google Deprecated FAQ Rich Results: Should You Still Mark Up for GEO?

Yes, but for a different reason than before. On 7 May 2026, Google officially deprecated FAQ rich results on its results page. The Search Central documentation confirms the staged rollout: displays gone on 7 May, end of rich result report in June 2026, end of Search Console API support in August 2026. The FAQPage markup remains a valid Schema.org type, but it no longer generates a rich result. The same logic had already applied to HowTo, deprecated for rich results in September 2023.

Consequence for GEO: schema shifts purpose. For ten years, we implemented it to win real estate in the visual SERP. Today, we implement it to help AI engines parse, index and cite content. The markup exits Google’s display game. It enters the citation game played by ChatGPT, Perplexity, Bing Copilot and Google AI Overviews.

This shift is confirmed on Microsoft’s side. Fabrice Canel, Principal Product Manager at Bing, told SMX Munich in March 2025 that Bing uses Schema.org to help its LLMs understand content for Copilot. That is the only official confirmation from an AI platform to date. Google made a vaguer signal in April 2025 about structured data giving an advantage in AI search experiences. OpenAI, Anthropic and Perplexity have said nothing publicly.

This doesn’t mean schema serves no purpose for other engines. It means the public evidence is thin, and you should implement it as infrastructure without expecting a miracle. The rest of the article separates what is proven, what is likely, and what is disguised marketing.

Schema.org and LLMs: What’s Proven, What Isn’t

The topic is saturated with spectacular unsourced statistics. You’ll read everywhere that FAQPage lifts citations by 340%, that schema multiplies mention frequency by 8.2x, or that you should expect 60% less visibility without markup. None of these numbers traces back to a verifiable study.

Here is the honest picture of the evidence in May 2026, documented by Jeremy Beckett (Belmore Digital, 2026):

• One official platform confirmation: Microsoft Bing, via Fabrice Canel at SMX Munich, March 2025.
• One vague signal from another: Google, April 2025, on the advantage of structured data in AI experiences, without specifics.
• No confirmation from OpenAI, Anthropic or Perplexity.
• One empirical study with a null result: Search Atlas (December 2024) found no correlation between schema coverage and LLM citation rates across OpenAI, Gemini and Perplexity.
• Zero peer-reviewed studies on the impact of Schema.org on AI visibility, as of May 2026.
• The Princeton / Georgia Tech GEO study (2024), often cited in favour of schema, never actually tested it. It tested adding citations and statistics within content, which is different.

Why this fog? Three reasons. First: LLM training pipelines are proprietary, so schema handling is opaque. Second: most published work on the topic comes from publishers selling schema or SEO tools, hence a direct commercial interest. Third: scientifically testing a single signal in a non-deterministic system requires methodological rigour the sector still lacks.

Reasonable defensive position: Schema.org is low-cost infrastructure to implement by default, without expecting a competitive advantage on its own. The content the markup points to matters more than the markup itself.

The 4 Schema.org Markups That Actually Matter for GEO in 2026

Four markups concentrate the defensive value for GEO. You implement them all, and you stop there for the first pass.

Organization: The Foundation of Disambiguation

The Organization markup identifies your brand as an entity LLMs can recognise. It goes once on the homepage, with the essential properties: name, url, logo, address, contactPoint, and especially sameAs linking to Wikidata, LinkedIn, X, YouTube and any official profile. It’s the markup that stops LLMs confusing your brand with a namesake or a competitor with a similar name. Without it, your mentions in article bodies can be attributed to the wrong entity.

Article or BlogPosting: Editorial Traceability

The Article markup identifies the author, publication date, modification date, publisher and main image. LLMs use these signals to evaluate freshness, credibility and authorship. One often-missed point: dateModified must be updated on every real revision of the content, not just at first publication. On fast-moving topics, it’s a recency signal AI engines use.

FAQPage: Question-Answer Extraction

Yes, even without the Google rich result. A marked-up FAQPage gives LLMs an explicit, identifiable, unambiguously extractable question-answer pair. It’s the ideal format to become the cited passage. But beware: the main value comes from the visible format on the page (H2 as question, direct answer in 2-3 sentences), not from JSON-LD alone. Schema reinforces, it doesn’t replace. And it only serves for real FAQs. FAQPage markup on content that isn’t an FAQ is treated as spam by every engine.

HowTo: Procedural Content

The HowTo markup identifies the steps of a tutorial, their order, and the estimated total time. Like FAQPage, it no longer produces a Google rich result since 2023, but remains useful for AI citation of procedures. Useful specificity: specificity beats generality. A HowTo in 7 precise steps with referenced tools and times gets cited more than a HowTo in 12 generic steps.

The 4 Emerging Schemas Rising for GEO in 2026

Beyond the base, four markups are emerging in 2026 and deserve attention or implementation on your key pages.

Speakable tells AI engines which passage of your page best answers the query. Particularly useful on long articles: you avoid the LLM lifting an off-topic passage by explicitly signalling the right one. ClaimReview positions your pages as verification sources on controversial or technical topics. DefinedTerm helps turn a sector glossary into a terminology reference for your industry, especially useful in young fields where the vocabulary is being built. The nested @graph is more best practice than schema: it consists of grouping Article, Author and Organization in a single JSON-LD block instead of separate script tags, which clarifies entity relationships for engines.

How to Implement Schema.org for GEO Without Falling for the Bullshit

Three rules frame an honest implementation.

1. The markup must accurately describe the visible content. Every engine, Google as well as LLMs, treats as spam a schema that announces what isn’t on the page. No invented FAQs to push a FAQPage, no fictional author to inflate an Article, no fake reviews to trigger a Review.
2. Validate systematically. Schema Markup Validator from Schema.org and Google’s Rich Results Test catch 90% of syntax errors. An invalid JSON-LD is ignored, not penalised, but you lose the signal.
3. Prioritise by business value, not exhaustiveness. No point marking up 200 secondary pages. Focus on the homepage, your 10 strategic pages, your pillar articles and your commercial pages.

And don’t expect Schema.org to do the work of your content. According to Pew Research Center (March 2025), 58% of US users see an AI summary on at least one Google search in a month. What decides your citation in those summaries isn’t markup. It’s the quality of the passage the markup points to.

FAQ

Should you remove FAQPage markup after the Google deprecation?

No. The rich result is gone, but the markup stays useful for LLMs. Keeping an existing FAQPage costs nothing, and it continues to signal an extractable question-answer format to AI engines. You just stop creating new FAQs solely to win SERP visuals.

Which schemas are confirmed as helping LLMs?

Only one has had official confirmation: Schema.org in general, by Microsoft Bing for its Copilot (Fabrice Canel, March 2025). No specific markup has been publicly tested by a platform. All type-specific recommendations come by extrapolation from classic SEO, not from direct LLM evidence.

JSON-LD or Microdata: which syntax should you use?

JSON-LD, without hesitation. It’s the syntax recommended by Google, more readable, separated from the visible HTML, easier to maintain. Microdata and RDFa remain valid on the Schema.org side but are declining. Any new implementation goes JSON-LD.

Is Schema.org markup a ranking factor?

No, neither for Google nor for LLMs. Google has confirmed this repeatedly. The Search Atlas study (December 2024) found no correlation between schema coverage and citation rates. Markup helps content understanding, not direct ranking.

Can you combine several schemas on the same page?

Yes, and it’s recommended. An article page can carry Article, Author, Organization, BreadcrumbList and FAQPage. The 2026 best practice is to nest them in a single JSON-LD block via the @graph property, which clarifies entity relationships for engines.

Should you mark up AI-generated content?

Yes, like any other content, provided it’s published on your site and faithful to reality. The writing mode doesn’t change Schema.org rules. What matters is that the markup matches the visible content.