Schema, FAQ and AI Search: What Changes After the New Data

For several years, structured data was presented as one of the most direct ways to turn content into additional visibility in search results. For many SEO teams, the logic seemed simple: mark up information more clearly, help search engines interpret the page and, in return, earn more attractive visual elements in the SERP. The problem is that this relationship between technical implementation and visible reward is no longer as predictable.

Google’s removal of FAQ rich results and the new data published by Ahrefs on citations in AI-powered search systems force the industry to rethink the way it talks about schema markup. The issue is not whether schema is dead. It is not. The more demanding question is this: what measurable value does it still create, in which contexts, and with what realistic expectations?

The recent enthusiasm around so-called Generative Engine Optimization, or GEO, brought a very appealing promise to the market: adding structured data could increase the likelihood of a page being cited in AI-generated answers. That promise became especially attractive because it is concrete, easy to sell and relatively simple to execute. But the simplicity of a recommendation does not, by itself, turn it into a sustainable competitive advantage.

The end of FAQ in the SERP is more than a visual change

The discontinuation of FAQ rich results should not be seen merely as the loss of a format. For a long time, FAQs allowed websites to occupy more space in search results, answer user questions directly and increase click-through rates for certain queries. Implementation was relatively accessible and, for that reason, quickly moved from technical best practice to mass-market tactic.

When a format stops distinguishing useful content and starts being applied systematically to almost every page, the visual reward loses value for the user. That is what many professionals had already observed: FAQs added to pages without a real editorial need, artificial questions created only to gain space, and short answers with little real contribution. The technology was technically correct, but the incentive began to generate noise.

This development fits into a broader pattern. Google has been reducing visible rewards associated with certain types of structured data. Some formats become limited to specific sectors, others are removed, and others remain valid without guaranteeing any visual element. The result is clear: the existence of a schema-supported property should not be confused with a promise of permanent visibility.

Structured data remains valid, but its role has changed

It is important to separate two ideas that are often treated as if they were the same. The first is the technical validity of structured data. The second is the existence of a visible reward in the SERP. A page can have correct markup in JSON-LD, Microdata or RDFa and still receive no visual enhancement in search results.

This distinction changes the way companies should look at investment. Implementing structured data only because there is an expectation of an immediate rich result has become a fragile approach. Implementing it as part of a clearer information architecture, consistent with entities, products, organisations, articles or events, still makes sense when there is a technical and editorial reason to do so.

Schema should be seen as an additional semantic layer. It helps declare what a page represents, what relationships exist between elements and what type of content is present. However, it does not replace visible content, answer quality, domain authority, user experience or brand consistency across digital touchpoints.

What the Ahrefs data adds to the AI debate

The Ahrefs report referenced in the source text analysed pages that added JSON-LD and compared citation changes in Google AI Overviews, AI Mode and ChatGPT against control pages that did not receive that change. The results did not show a clear and consistent increase in citations after the implementation of schema.

The figures presented were limited in two systems and negative in a third. Google AI Mode reportedly showed a positive variation of 2.4%, ChatGPT a positive variation of 2.2%, and Google AI Overviews a decline of 4.6%. Even where growth existed, the framing of the data itself points to a difference too small to be treated as robust evidence of impact. In the case of the decline in AI Overviews, Ahrefs also did not directly attribute that change to schema.

The most relevant detail lies in the profile of the analysed pages. All of them already had more than 100 citations in AI Overviews before structured data was added. In other words, these were pages that were already visible, accessible and recognised by AI-powered search systems. In this context, adding a new layer of markup can be compared to placing an extra label on a product that is already on a busy shelf: the label may organise information better, but it does not guarantee that the product will be chosen more often.

The study does not prove that schema is useless

A rushed interpretation could conclude that structured data no longer serves any purpose. That conclusion would be just as exaggerated as the promise that schema automatically increases AI citations. The study described evaluates a specific result: citation variation in certain systems, over a specific period, for pages that were already being cited before the change.

Several questions remain open. Would the impact be different on new pages that do not yet have visibility in AI answers? Would some types of schema, such as Product, Review, Event, Video, Organization or Article, behave differently if analysed separately? Would a longer observation period reveal indirect effects, for example in entity interpretation or eligibility for certain formats? These questions still do not have a definitive answer.

It is also relevant to note that different systems may handle structured data in different ways. One model may prioritise visible HTML during direct page retrieval, while another may use signals collected at earlier stages of indexing, entity extraction or knowledge consolidation. The fact that an experiment does not detect direct use of JSON-LD at one stage of the process does not eliminate every possible role structured data may play in the ecosystem.

The problem lies in an overly simplified sales pitch

The main weakness revealed by these developments is not in the technology, but in the way it is sold. The phrase “add schema to increase AI citations” has become a comfortable promise for consultants, tools and optimisation guides. It is specific, sounds technical and gives clients the feeling that there is an objective action to solve a new problem.

But AI search does not work like a switch activated by a JSON-LD property. Answer systems combine crawling, indexing, information retrieval, relevance assessment, content quality, perceived authority, freshness, consistency and suitability to the user’s intent. Schema may help clarify some elements, but it does not turn a weak page into a citable source.

This is where many GEO approaches start repeating old SEO mistakes. When a tactic appears to work, or at least seems plausible, it is quickly packaged as a universal solution. It is then replicated at scale, loses differentiation and ends up generating low-quality signals. The history of FAQs is a good example: a useful feature became an overused practice and, over time, the reward disappeared.

Visible content may matter more than invisible markup

One of the most interesting points in this debate is the relationship between structured data hidden in the code and content that is actually visible on the page. If certain AI systems rely mainly on visible HTML during direct retrieval, then editorial structure becomes even more important. Clear titles, useful subheadings, direct answers, objective definitions, well-built lists and concrete examples may become more relevant for AI citation than invisible markup that is not reflected in the content.

This does not mean abandoning technical work. It means aligning technology and content. A page that declares itself as an article through structured data should contain a strong article. A product page with Product schema should present complete information, clear prices where applicable, availability, variants, useful descriptions and trust signals. An FAQ area should answer real questions, not simply fill a markup block.

In practice, the best optimisation for answer systems often starts by making the page more understandable to humans. When content is well organised, answers concisely and goes deep enough to demonstrate authority, it also becomes easier for machines to interpret. Schema should confirm and structure that reality, not try to compensate for its absence.

How brands should review their structured data strategy

The first step is to audit what already exists. Many companies accumulate old markup, unused properties, FAQ blocks that no longer have visual impact and implementations created by automatic modules without strategic validation. Keeping that history without review can create a false sense of optimisation.

A useful audit should answer simple questions: which types of structured data are active? Are they aligned with visible content? Are they supported by features that are still relevant? Are there technical errors? Is there inconsistent information between HTML and JSON-LD? Is the same content marked up in contradictory ways? These questions are more important than adding new layers of code simply because a trend recommends them.

Then, it is necessary to prioritise. In many cases, Product, Review, Event, Video, Organization and Breadcrumb still have practical value when they are applicable and correctly used. Article and Organization can contribute to a clearer representation of publications, authors, entities and brands. FAQ, on the other hand, should no longer be treated as a tool for visual expansion and should only be used when there is a real question-and-answer section that is useful to the user.

What to do with FAQ after the removal of rich results

The removal of the visual enhancement does not mean that every FAQ section should disappear from websites. Many remain useful for reducing friction, anticipating commercial doubts, explaining conditions, clarifying processes and supporting purchase decisions. The difference is that their existence should be justified by editorial value and not by the expectation of occupying more space in the SERP.

A good FAQ section should be based on real questions: doubts raised with sales teams, internal site searches, social media interactions, conversations with customers, customer support data and search intent analysis. When answers come from this knowledge, the section becomes an experience asset, not merely a technical device.

It is also worth rethinking the format. Some questions deserve short answers. Others justify dedicated pages, detailed guides, comparisons or demonstrations. The decision should not depend only on the existence of a schema type, but on the depth required to resolve the user’s intent.

AI search requires authority, not just formatting

Citations in AI-generated answers tend to favour sources that appear trustworthy, complete and suitable for the question. Technical markup can help remove ambiguity, but authority is built through consistency. This includes updated content, identifiable authors when relevant, signs of expertise, clear references, thematic coherence and a digital presence that confirms the credibility of the brand.

For companies, this means thinking beyond isolated pages. A single page optimised with JSON-LD will hardly compensate for a weak ecosystem. On the contrary, a consistent set of content, well-structured service pages, case studies, proprietary data, clear institutional information and useful answers to market questions can create a stronger context for search engines and AI systems.

That is why the conversation about structured data should be integrated into a broader digital marketing strategy. Technical work remains important, but it only creates advantage when connected to positioning, content, architecture, experience and business goals.

Less automation, more editorial intent

One of today’s temptations is to automate markup and content production using the same logic: generate more, faster and with less human intervention. This approach may seem efficient, but it increases the risk of creating redundant pages, artificial sections and structured data that simply repeats information without adding clarity.

Search is becoming more demanding in its assessment of usefulness. This does not mean that automation has no place. It does, especially in large catalogues, online stores, knowledge bases and websites with a lot of repeatable information. However, automation should respect editorial rules, correct data and continuous review. Automatically marking up every possible element is not a strategy; it is merely technical accumulation.

Editorial intent should lead. Before deciding which schema to apply, the team should understand what problem the page solves, which entity it represents, what information needs to be unambiguous and what action is expected from the user. Only then does it make sense to choose the markup that best translates that structure.

The impact on ecommerce and transactional pages

In online stores, the conversation is especially sensitive. Product schema may remain relevant because it helps declare essential attributes such as price, availability, reviews, brand, reference and variants. When applied correctly, it can support rich results and make complex catalogues easier to interpret.

But here too there is a difference between markup and experience quality. A product page with complete structured data but poor descriptions, generic images, missing logistics information or unconvincing reviews will hardly maximise its potential. Markup should reflect a strong page, not disguise an incomplete one.

For brands with a presence on platforms such as Shopify, technical implementation should be accompanied by a review of themes, applications, product data and consistency between catalogue, HTML and JSON-LD. Often, the problem is not the total absence of schema, but duplicated, incomplete or contradictory markup generated by different modules.

How to measure value without falling into easy promises

Measuring the value of structured data requires choosing the right metrics. For some types, it makes sense to track impressions and clicks associated with rich results in Google Search Console. For others, evaluation involves technical errors, coverage, eligibility, entity consistency and organic visibility development. In the case of AI citations, measurement is even more unstable because systems vary, answers change and the available data remains limited.

A good methodology should compare periods, isolate changes whenever possible and avoid drawing conclusions from individual cases. If a page gains citations after adding JSON-LD, that does not prove causality. Perhaps the content was updated, perhaps demand for the topic increased, perhaps the page received links, perhaps the AI system changed its behaviour. The opposite is also true: the absence of growth in a specific sample does not prove universal uselessness.

The central point is to abandon shortcut thinking. Schema should enter the plan as a technical hypothesis with defined goals, not as a guarantee. When an active rich result exists, that result should be measured. When the value is semantic, coherence and the ability to reduce ambiguity should be assessed. When the ambition is to appear in AI answers, content quality, thematic authority and the visible structure of the page must also be measured.

The future of schema will be less spectacular and more structural

Structured data is likely to remain part of search, but with fewer moments of spectacular visual reward. The trend points to a more discreet role: helping systems understand entities, relationships, content types and important attributes. That value may be real even when it does not appear as stars, expandable boxes or highlighted cards.

For SEO professionals, this requires maturity. Not everything that matters can be seen in the SERP. But not everything that is invisible should be sold as a guaranteed advantage either. The balance lies in implementing what makes sense, removing what has lost usefulness, validating what is active and explaining to clients that structured data is a piece of infrastructure, not a magic formula.

The removal of FAQs and the Ahrefs data do not close the topic. On the contrary, they make the conversation more serious. The market needs fewer absolute claims and more well-designed tests, segmented by type of schema, page type, sector, AI system and domain maturity. Only then will it be possible to understand where markup continues to create measurable value.

Conclusion: schema is not dead, but the promise has changed

Schema markup remains useful when it translates real content, supports active features and improves the semantic clarity of a page. What has become weaker is the pitch that presents it as a direct shortcut to gain space in the SERP or secure citations in AI systems. After the removal of FAQs and the data presented by Ahrefs, that promise requires much more caution.

Brands should continue using structured data, but with revised priorities. First, visible and useful content. Then, clear architecture. Next, markup aligned with what the page actually contains. Finally, honest measurement. This path may not be as seductive as a quick solution for AI citations, but it is much more resistant to platform changes.

At BYDAS, we combine strategy, content and technical implementation to turn structured data into part of a stronger organic presence. If your brand needs to review priorities, structure and search opportunities, discover our work in SEO and organic traffic.