Google has quietly dropped support for nine ItemTypes from its rich result gallery over the past two years, a window that lines up almost exactly with the mass adoption of ChatGPT and the shift toward AI-driven search. The most recent casualty, FAQ/FAQPage, has reignited a debate that keeps resurfacing in technical SEO circles: is schema markup dying?

The short answer is no, but the question itself is worth unpacking, because it points to a bigger shift in what technical SEO actually needs to protect now: not just crawlability, but the integrity of the data machines read from your site. See also: 3 CMS platforms control 73% of the market & shape technical.

What the FAQ schema deprecation actually means

Google’s deprecation notice for FAQ rich results was specific: the company said it would drop “the FAQ search appearance, rich result report, and support in the Rich results test” starting in June 2026. Notice what that notice does not say. It never states that FAQ schema itself is no longer worth implementing, only that the visual rich result built on top of it is going away.

That distinction matters. Schema.org markup is a comprehension layer, a way of identifying entities and the relationships between them, and a rich result is just one display feature built on top of that layer. Some properties are being deprecated while others, like Product, are actively being extended. SEO researcher Gianluca Fiorelli has pointed out that much of the testing used to judge whether schema affects AI citations runs on limited datasets, which should temper how confidently anyone declares schema irrelevant based on a handful of experiments.

There’s also a broader framing worth borrowing from SEO writer Suganthan Mohanadasan, who described schema as having three separate “lives”: Google’s index pipeline, indirect influence during LLM pretraining, and direct use during LLM runtime retrieval. Most SEO measurement has historically focused only on the first of those three, which means schema’s real value has likely been under-reported for years, not overstated.

Ambiguity, not deprecation, is the real threat

Schema functions as an ontology, a shared vocabulary that supports data integrity across a web standard. The real risk to that integrity isn’t a missing rich result. It’s ambiguity.

Ambiguous data leads directly to hallucinations. Hallucinations compound over time, and in the worst case, they can bake inaccurate associations into an LLM’s pretraining data itself, an error with effects that outlast any single search result. If a language model or agent can misread your content, at some point it will, and once it does, there’s a real risk of what researchers call semantic drift: a model gradually favoring narrative plausibility over verified fact. A study by Andrea Volpini and Chiara Carrozza, titled “Sangue e Grafi: Teaching a Small Model to Read the Bloodline,” found exactly this pattern: frontier models tended to favor narrative over facts, while a much smaller model equipped with knowledge-graph tools performed on par with them. The lesson is that structured, disambiguated data can outperform raw model scale when it comes to factual accuracy.

The five layers that make up data integrity

If the SEO’s job now is to maximize data integrity rather than chase a specific rich result, it helps to break that goal into distinct layers:

• Entities. What exists, and what that thing actually is: Thing, Organization, Person, stabilized with @ids and tied out via sameAs to authorities like Wikidata, GS1, ISNI, or ORCID, so a machine can tell your “Apple” from the fruit.
• Relationships. How those entities connect to each other, expressed through @id and sameAs links and RDF. Tools like Yoast SEO’s schema aggregation feature and Dixon Jones’s EntityMap exist specifically to help manage this layer.
• Format. How the structure gets serialized and served: JSON-LD, RDFa, and Microdata for markup, Markdown for newer standards like llms.txt and agents.md, and endpoints handling content negotiation, ARD, and MCP.
• Actions. What an agent is told it’s allowed to do, through schema.org Actions like BuyAction, and newer protocols including WebMCP, ACP, and UCP.
• Perception. Grounding, third-party perception, and sentiment, essentially how the web’s broader signal about you gets factored in alongside your own structured data.

Of those five, you have direct control over the first four. Perception is shaped by everyone else who writes about you, which is precisely why brand mentions and reputation have become so tightly linked to AI visibility. We’ve covered that connection in more depth in why international SEO needs a global knowledge integrity strategy.

Why there’s no single standard to chase

Schema.org itself exists because Google, Microsoft/Bing, and Yahoo sat down and jointly built a shared standard, with Yandex joining later. The XML sitemap protocol emerged the same way five years earlier. When search engines needed a common standard in the past, the major players simply agreed on one together.

Nothing comparable is happening now. The current landscape of emerging protocols, mostly launched within the past two years, generally serves one of three goals: aggregation, guidance, or consumption, all aimed at reducing the number of requests an agent needs to make while increasing token efficiency. But there is no consortium bringing these platforms together, and even foundational questions, like how markdown should be served to agents, remain genuinely contested. The companies building these tools are no longer just competing in search; they’re navigating how their decisions ripple through jobs, economies, and the broader shape of the internet. SEO-specific clarity is unlikely to be anyone’s top priority while that’s the case. Our recent look at Google’s Open Knowledge Format and its new trust signals is a good example of how quickly this landscape keeps shifting without any unifying body behind it.

A practical order of operations

Given the lack of consensus, the responsible move is not to bet everything on one protocol. It’s to build the four controllable layers of data integrity in a sequence that limits wasted effort:

1. Stabilize your @ids and add sameAs links out to Wikidata and other relevant authorities first. Every other layer depends on this foundation being solid.
2. Test how you’re actually being interpreted, using a tool like NLWeb, rather than assuming your knowledge graph reads the way you intended it to.
3. If you run an ecommerce site, audit your product feed before touching anything newer, and look closely at BuyAction while you’re there. Only ReadAction and SearchAction are deployed at any meaningful scale today, which means this part of the field is still genuinely open for early movers.
4. Attempt an implementation of WebMCP. It works on any site, not just ecommerce, and does not require a large engineering lift.
5. Deprioritize markdown serving and content negotiation until you have spare engineering capacity, unless you can lean on an existing solution such as Cloudflare’s Markdown for Agents.
6. Keep an eye on OKF and ARD. New protocol launches from major platforms are worth tracking closely, particularly anything that changes how an agent or LLM understands a site as a whole.

None of this is a bet that any single protocol wins out. Implementing pieces of this stack reduces the odds that a language model has to take “the long way round” to construct an answer about you, and hedges against whichever protocol ends up winning adoption. Smaller-scale experiments are a reasonable way to learn how these pieces behave in practice before committing a larger site to them; several practitioners have already rebuilt personal or small business sites with a full “agentic ready” stack, including content negotiation, a markdown alternate, llms.txt, and WebMCP, specifically to observe how agents interact with each piece.

Own the layers, not the protocol

There is no single winning standard in sight, and no consortium currently working to repeat what happened with the XML sitemap or the original schema.org launch. The stack of competing protocols is large and still growing, but every piece of it draws from the same underlying substrate: accurate entities, explicit relationships, and content a machine can parse without guessing.

Rankings were the success metric of the old web. Trust, integrity, accuracy, and validity are what earn that success now, and building them deliberately, layer by layer, is still fundamentally the SEO’s job. For a broader view of how this fits into a modern technical audit, see our piece on why the technical SEO audit needs a new layer, and if you’re deciding how much of your roadmap to devote to this now, the data behind the SERP FAQ removal and what it means for schema’s AI search value is a useful next read.

Frequently asked questions

Is schema markup dead now that Google dropped FAQ rich results?

No. Google’s deprecation applies only to the FAQ rich result display feature, not to FAQ schema itself or to schema markup broadly. Schema still functions as a comprehension layer that helps both search engines and AI models identify entities and their relationships, and several properties, like Product, are actively being extended.

What is data integrity in the context of SEO?

It refers to how accurately and unambiguously a machine, whether a search crawler or an AI agent, can interpret your site’s entities, their relationships, the format they’re served in, the actions an agent is allowed to take, and how third parties perceive your brand. Weak data integrity increases the risk of AI hallucinations and semantic drift about your business.

What should I prioritize first for data integrity?

Stabilizing your @ids and linking them out to established authorities like Wikidata via sameAs comes first, since every other layer, from relationships to actions, depends on that foundation being reliable.

Do I need to adopt every new agentic protocol like WebMCP or OKF?

Not immediately, and not all at once. Prioritize based on your technical capacity and business type: WebMCP is low-effort and works on any site, while markdown serving and content negotiation can wait until you have engineering capacity to spare, unless you can use an existing solution like Cloudflare’s Markdown for Agents.