The schema tag has had a rough week. Google has eliminated rich FAQ results. Four days later, Ahrefs published a report, finding that adding JSON-LD did not produce a clear citation lift across Google AI Overviews, AI Mode, or ChatGPT.
This development weakens two common spaces for schema markup: increased SERP visibility and potential AI citation benefits. This article examines their implications and what the data shows about the future of schema.
Google Rewards Visual Schema Has Declined
Google has been rolling back visual search rewards associated with structured data types since 2023. Google limited rich FAQ results to authoritative government and health sites, and rich HowTo results were limited to the desktop and later deprecated.
In 2025, Google announced that it will retire several structured data features, including Course Information, Claim Review, and Estimated Earnings. Book Actions was initially included but later pulled after Google removed its deprecation banner. Google called leaving the rest as “infrequently used in Search” and no longer provides value to users.
In 2026, the planned Practice Problem data was withdrawn. John Mueller noted on Reddit that “tag types come and go, but there are precious few worth sticking with.”
The pattern is that the rewards for visual structured data have disappeared after becoming mainstream SEO tactics. The tag itself is always valid, but the rich result is not. Google doesn’t always explain these removals as responses to overuse, but the pattern provides little reason to treat any type of markup as a strict strategy.
These recent revisions differ because the evidence for the single proposed value of the exchange is also weak. The “GEO” advice space for the claims schema improves AI citations, and Ahrefs data is partially tested for that.
Ahrefs Report findings
Ahrefs tracked 1,885 web pages that added the JSON-LD schema. Each page is matched with control pages that have not added a schema. Citation changes are measured across Google AI Overviews, AI Mode, and ChatGPT.
The results were low. Google AI Mode is shown +2.4%ChatGPT is displayed +2.2%and Google AI Overviews shown -4.6%.
The first two were too small to be distinguished from random variables. The decline in AI Overviews was statistically significant, but Ahrefs said it would not confidently attribute that to the schema.
Each page in the dataset already has over 100 AI Overview citations before any schema is added. These pages were already clear and quotable.
Ahrefs agreed that for pages that have not yet been exposed to AI, schema can still help with clarity, classification, or indexing. But their data cannot confirm that.
Gianluca Fiorelli, a strategic SEO consultant, called the study “one of the most credible pieces of research to come out of the AI Search space in 2026.” But he said the scope was narrower than the title suggested. He compared it to “testing whether adding a label to a bottle already on the supermarket shelf makes customers pick it up more often.”
Ahrefs also cited a searchVIU test that found five AI systems relied on visible HTML during direct page retrieval and did not use encrypted JSON-LD, Microdata, or RDFa. That discovery includes one section of the pipeline. It does not extract the schema that plays a role in the prior identification or understanding of the entity.
Ryan Law, Ahrefs director of content marketing, summarized LinkedIn’s findings, saying:
“Does adding a schema tag help your pages get indexed in AI searches? Probably not,” he wrote. He added that schema is “probably not a magic setting to improve your AI quotes.”
The Career Debate
Both update the location between active contention about schema and GEO.
About 168,000 pages use the phrase “FAQ schema is relevant to GEO,” according to search results Lily Ray, VP of SEO and AI Search at Amsive, flagged on LinkedIn. He called this a common practice.
“Anything that can be spammed in SEO, will be spammed,” Ray wrote. He warned of this in a 2019 Moz article when the FAQ schema was first introduced, and described the removal of Google’s FAQ as a repetition of the same cycle.
Ray went through all of his posts, calling it “putting on my foil hat” and “just an opinion.” But the pattern he described is the same as that seen in the timeline above. A useful markup is rated as a trick, Google takes the prize, and the industry moves on to the next one.
Joost de Valk, founder of Yoast, made the connection clear in a blog post. “The GEO industry is replaying the original SEO, very quickly,” says de Valk. “And the withdrawal of the FAQ schema is the first concrete point of proof that the cycle has turned.”
He also submitted a proposal to Schema.org for a new type of FAQSection to address what he saw as a structural problem, separating “this page has an FAQ section” from “this page IS an FAQ.”
The frustration was especially acute for staff who watched GEO’s playbook tighten up on the schema as its physical recommendation. Mark Williams-Cook, director at Candor and co-founder of Asked, shared the Ahrefs report on LinkedIn.
“The GEO bros are selling snake oil with a schema to increase quotes, but people like Gianluca Fiorelli are talking about logic,” he wrote.
Marie Haynes, founder of Marie Haynes Consulting, commented on Ray’s post with a completely different theory.
“My theory is that Google needs our FAQs to train the AI so they gave us an incentive to add them (rich results.) And now they don’t need them,” he wrote. The theory is not confirmed by any primary source, but it shows how far speculation has gone.
Some doctors defer to dark readings. Google’s broad guidance introduces structured data as a way to make page information machine-readable, and at the Search Central Live 2025 event in Madrid, the Search Relations team told practitioners that structured data types are still worth using.
What the Data Can’t Answer Yet
Whether the schema is helping pages that aren’t currently being cited is a separate question that the data can’t answer, because each page already has over 100 AI Overview test citations before the schema is added.
The test also grouped all schema types together. Article, FAQ, Product, HowTo, and Association are all treated as one section. The results of a certain type are not isolated, and they can look different.
A 30-day measurement window can miss slow results, and on live websites, schema changes can spill over into other page changes, making it difficult to distinguish what the schema did from what changed on its own. The report only checked the schema in the page’s HTML, not the JavaScript-injected schema, which AI crawlers treat differently.
Ahrefs rated Google AI Overview, AI Mode, and ChatGPT. Whether Bing, Copilot, Perplexity, Claude, or other response systems treat schema differently than Ahrefs’ rated systems is an open question.
The FAQ’s notice of revocation of the FAQ states that the company will continue to use the edited FAQ data to “better understand” the pages. What it produces in measurable terms is unclear. The same uncertainty applies if the schema affects the citations indirectly, by relevance, business sense, or source selection, instead of the direct retrieval that was evaluated by searchVIU.
No one has published data distinguishing that method.
Why This Matters
Ahrefs data doesn’t provide a compelling reason to add JSON-LD, expecting temporary AI citation benefits from pages already visible in AI Overview. The more difficult question is what to do with schema techniques in general.
Product, Update, Event, Video, and other types of structured data still support active features for rich results. Organization, Person, and Article tags can still help describe businesses and content, even if the payment is less obvious.
The “schema doesn’t work” blanket statement emphasizes what the data showed, because the test included all variables and measured only one effect. What the data challenges is a specific sales pitch.
“Add schema to improve AI citations” was one of the most trusted recommendations in the GEO guide. For example, Phrase.io called schema markup “very important for AI, GEO, and AEO searches.”
Without data backing up that claim, it’s hard to justify the investment. The AI systems in searchVIU tests rely on visual HTML during retrieval, not JSON-LD. That suggests content structure, clear headings, and direct responses in prose may be more important to AI citation than markup design.
Looking Forward
A lingering question in the SEO industry is when schema creates measurable value. Adding JSON-LD did not measurably increase AI citations on pages that already appear in the AI Overview.
On those pages, the schema looks more like pipes that serve other systems than a lever that moves the quote count. That’s still a real value, but a different tone.
Featured Image: THE FACILITIES ARE BEAUTIFUL/Shutterstock
