Schema Markup, AI Citations, and the SEO Rules That Just Changed

Two updates landed within 48 hours of each other last week, and together they quietly redraw the rules of structured data strategy. Pay attention — this is one of those moments where the right read saves you six months of wasted implementation work.

Google’s FAQ Cleanup Is a Signal, Not Just a Housekeeping Move

Search Engine Journal reports that Google has fully deprecated FAQ rich results, completing a rollback that began in 2023 when the feature was restricted to government and health sites. For the vast majority of brands, FAQ schema had already stopped generating the accordion-style SERP enhancements it once did. The formal removal simply closes the loop.

But the strategic read here goes deeper than “stop tagging your FAQs.” Google’s deprecation of FAQ rich results is part of a broader contraction of traditional SERP features as AI Overviews absorb the answer-layer real estate that FAQ snippets once occupied. The machine is consolidating. What used to take a dedicated SERP feature — surfacing direct answers — is now handled upstream by generative AI. That shift doesn’t eliminate the value of FAQ-style content; it relocates where that content needs to perform. The question is no longer “will this generate a rich result?” but “will this get cited by an AI model?”

For Southeast Asian brands publishing across Bahasa Indonesia, Thai, Filipino, and English simultaneously, this also surfaces a practical issue: FAQ schema was one of the cleaner ways to signal structured Q&A content across multilingual pages. Teams will need to reassess how they signal answer-intent content without that scaffold.

Schema and AI Citations: The Correlation That Doesn’t Mean What You Think

This is where it gets interesting — and a little uncomfortable for anyone who’s been selling schema as an AI visibility play.

Ahrefs tracked 1,885 pages that added schema markup and measured the impact on AI citation rates. The headline stat making the rounds: AI-cited pages were almost three times more likely to have JSON-LD than non-cited pages. Sounds like a green light to schema everything, right?

Not quite. Adding schema to pages that weren’t already being cited by AI models produced almost no movement in citation rates. The correlation between JSON-LD and AI citations reflects the quality of pages that tend to implement schema — authoritative, well-structured, technically mature — not a causal relationship between the markup itself and AI selection behaviour. Schema appears to be a hygiene signal, not a ranking lever. AI models are selecting for topical authority, content depth, and source credibility. Schema helps them parse what you’ve already built; it doesn’t compensate for content that isn’t citation-worthy in the first place.

The practical implication: if your AI citation strategy starts with schema implementation, you’re starting in the wrong place.

---

What Actually Earns AI Citations in 2026

So if schema is necessary but not sufficient, what is the actual work? The Ahrefs data points toward a familiar but newly urgent conclusion: AI models reward content that demonstrates clear expertise, answers questions with specificity, and is structured in ways that make extraction easy.

For teams in Southeast Asia, this means a few concrete things. First, depth over breadth — a single comprehensive page on, say, cross-border e-commerce logistics in the Philippines will outperform ten thin category pages targeting the same cluster. Second, named authorship and expertise signals matter more than they did two years ago. A bylined piece from a recognised practitioner in a field reads differently to a language model than anonymous brand content. Third, internal linking architecture that creates clear topical clusters helps AI systems understand the scope of your authority on a subject — not just the existence of a single page.

Brands on high-competition platforms like Lazada or Shopee face a separate challenge: their most valuable content often lives inside marketplace ecosystems with limited structured data control. Building an owned content hub that links outward to those product pages — and earns citations in its own right — is the only real hedge against marketplace platform risk.

Rebuilding Your Structured Data Strategy From the Right Starting Point

With FAQ schema gone and the schema-citation relationship clarified, here’s how to reprioritise. Article, Product, BreadcrumbList, and Review schema remain genuinely useful for helping search systems parse page context and relationships — keep those. HowTo schema still shows engagement in certain verticals. But the implementation priority should follow content quality, not precede it.

Semrush’s recent overview of AI-assisted SEO workflows highlights an underused application here: using AI tools to audit existing content for answer-completeness before adding structured markup. The logic is sound — if a language model can’t extract a clean, specific answer from your prose, wrapping it in JSON-LD won’t change that. Run your top 20 pages through a prompt that asks “what question does this page definitively answer, and how specifically?” If the answer is vague, that’s your editorial brief, not your schema brief.

For multilingual teams managing pages in three or more languages — common across regional brands in SEA — structured data also needs to be audited per locale, not just implemented globally and mirrored. Language models don’t treat a Thai-language page and its English equivalent as interchangeable authority signals.

---

Key Takeaways

• FAQ schema is fully deprecated — redirect that implementation effort toward content depth and topical authority signals that AI models actually respond to.
• Schema correlates with AI citations because high-quality pages tend to have both, not because markup causes citations — sequence your strategy accordingly.
• For Southeast Asian multilingual sites, structured data needs locale-specific auditing; a global implementation copied across language versions won’t carry equal authority signals.

The deeper question worth sitting with: as AI models become the primary arbiters of what content gets surfaced — not just ranked — does the entire SEO framework of “optimise the page for the crawler” need to be rebuilt around “make the content genuinely worth extracting”? The technical and the editorial are converging. The teams that figure out how to run those two disciplines in the same room will have a structural advantage that no schema implementation can replicate.

At grzzly, this is exactly the territory we work in with regional brands — mapping where structured data, content architecture, and AI visibility strategy intersect, and building search programmes that hold up as the ground keeps shifting. If your team is rethinking its approach to search in 2026, Let’s talk