Schema markup is code that tells a search engine what a page is about: that this number is a price, that this string is an author, that these two lines are opening hours. It adds no content. It translates the content already on the page into a format a machine can read.
Everyone agrees on that much. The disagreement starts with the return. Guides on the subject regularly promise a click-through rate increase of 20% to 30%, and none of the ones I opened names the study the figure comes from. Meanwhile, on 7 May 2026 Google stopped showing one of the most widely installed rich results, and four days later Ahrefs published an experiment on 1,885 pages that takes apart the other promise of the moment, the one about artificial intelligence.
What Google does with structured data, and what it doesn't
The official documentation is drier than the people who summarise it. The page Introduction to structured data markup explains that markup gives explicit clues about the meaning of a page and that the recommended format is JSON-LD. Then it adds the line almost nobody quotes: "You must include all the required properties for an object to be eligible for appearance in Google Search with enhanced display". Eligibility, not a guarantee.
Two practical consequences follow. First: structured data does not move rankings. It changes how a result looks once the position is already there, and position is something you work on with other tools, from keyword research to on-page optimisation. Second: if the rich result tied to a markup type no longer exists, that markup produces nothing visible. That is exactly what has just happened to FAQs.
On 7 May 2026 FAQ rich results disappeared
Google made no announcement: it added a note to the FAQPage documentation. The text reads: "As of May 7, 2026, FAQ rich results are no longer appearing in Google Search. We will be dropping the FAQ search appearance, rich result report, and support in the Rich results test in June 2026". Support in the Search Console API is removed in August 2026.
| Date | What changes |
|---|---|
| August 2023 | FAQ rich results limited to authoritative government and health websites |
| September 2023 | HowTo documentation removed: the rich result no longer appears in results |
| 7 May 2026 | FAQ rich results stop appearing in Google Search altogether |
| June 2026 | The FAQ search appearance, the Search Console report and Rich Results Test support are removed |
| August 2026 | FAQ support removed from the Search Console API |
You don't need to do anything with markup that is already installed. Google says unused structured data causes no problems for Search, and the FAQPage type remains valid in the Schema.org vocabulary: it simply no longer generates an accordion in the SERP. What FAQs are worth for SEO today is something we measured separately.
One check I'd recommend before the reports disappear: open Search Console and look at which search appearances your site has actually recorded. On the visilay.com property, from 1 January to 6 May 2026, the only appearance recorded is TRANSLATED_RESULT, with 14 impressions and zero clicks. No FAQs, ever, despite the markup being on dozens of articles. Since August 2023 that rich result had already been reserved for others, and for three years the FAQ block did nothing on the SERP side.
Schema markup is not the lever for getting cited by AI
The promise that replaced the rich snippet one is that marking pages up in JSON-LD makes them easier for language models to understand, and therefore cited more often. It sounds reasonable. The two sources that matter say something else.
The first is Google. On the page AI features and your website: "You don't need to create new machine readable files, AI text files, or markup to appear in these features. There's also no special schema.org structured data that you need to add". This applies to AI Overviews and AI Mode.
The second is an experiment. On 11 May 2026 Louise Linehan and Xibeijia Guan published a study on Ahrefs that followed 1,885 pages which added JSON-LD between August 2025 and March 2026, comparing them with 4,000 control pages with similar citation levels, measured 30 days before and 30 days after using four different statistical methods.
| Platform | Change in citations after adding schema | Statistically significant |
|---|---|---|
| Google AI Mode | +2.4% | No |
| ChatGPT | +2.2% | No |
| Google AI Overviews | −4.6% | Yes (p ≈ 1/2500) |
The authors' own sentence is "adding schema produced no major uplift in citations on any platform". The drop on AI Overviews is measurable but small, and even they are not prepared to attribute it to the markup.
That doesn't mean structure is irrelevant to AI. It means JSON-LD is not the shortcut it is sold as. Another Ahrefs study, from 2 March 2026 on 863,000 SERPs and 4 million cited URLs, shows that only 38% of AI Overview citations come from top 10 pages, against 76% a year earlier: selection is moving outside the original SERP. If the goal is to appear there, the work that moves the needle is different, and I have set it out in the guide to AI Overviews ranking factors and in the one on how to appear in ChatGPT answers.
What rich results are really worth: four data series
The CTR percentages that get passed around have no source. These do: they come from the Search Console properties of two sites we manage, an Italian outdoor equipment e-commerce business and its shop on a subdomain. Period 8 May to 18 August 2026, aggregated by page.
| Property | Search appearance | Impressions | Clicks | CTR | Average position |
|---|---|---|---|---|---|
| Main site | Property total | 214,096 | 4,234 | 1.98% | 6.52 |
| Main site | Product snippet | 27,907 | 519 | 1.86% | 12.06 |
| Shop (subdomain) | Merchant listing | 1,307 | 35 | 2.68% | 3.59 |
| Shop (subdomain) | Product snippet | 1,675 | 21 | 1.25% | 32.50 |
Two readings. First: on the main site, pages that get the product snippet have a CTR of 1.86%, which is below the property average of 1.98%. Not because the rich result does harm, but because those pages sit at an average position of 12.06 against 6.52. Position weighs more than how the result looks, and a generic figure along the lines of "30% more clicks" doesn't tell you that.
The second is more useful if you run an e-commerce site. On the shop, the same Product markup generates two different appearances. The merchant listing, which according to the Product documentation needs price, availability and shipping and returns information, has an average position of 3.59 and a CTR of 2.68%. The product snippet, the reduced version, sits at position 32.50 with 1.25%. The markup is the same: what changes is how complete it is. On these projects, the half day spent adding shippingDetails and hasMerchantReturnPolicy pays back more than any other change to the JSON-LD.
The types that still produce something visible in 2026
| Type | What it produces today | Who it's for |
|---|---|---|
| Product, Offer | Product snippet and merchant listing: price, availability, reviews, shipping | E-commerce |
| LocalBusiness | Business information in local results | Businesses with a physical location |
| Organization | Logo, company details, entity linking through sameAs | Everyone |
| BreadcrumbList | Navigation path in place of the URL in the snippet | Sites with a deep hierarchy |
| Article, NewsArticle | Eligibility for Google's editorial surfaces | Publishers and blogs |
| VideoObject | Previews and key moments | Sites with video |
| Event | Events carousel | Event organisers |
| JobPosting | Google Jobs | Anyone publishing job ads |
| Recipe | Recipe carousel | Food |
| FAQPage | Nothing, since 7 May 2026 | Nobody |
| HowTo | Nothing, since September 2023 | Nobody |
The full list is in the search gallery, and it is shorter than you might expect: the Schema.org vocabulary contains hundreds of types, and the ones Google turns into something visible number a few dozen. Marking up the rest is work nobody reads, apart from the language models passing by. If you have video on your site, VideoObject is the only type on this list worth handling by hand, and I cover it in the guide to video SEO.
JSON-LD, microdata, RDFa: why only one is left
Microdata is written inside the HTML, attribute by attribute, and every theme redesign breaks it. RDFa has the same problem with worse syntax. JSON-LD sits in a separate block, can be injected via JavaScript or a tag manager, and nothing happens to it when the design changes. It is the format Google recommends, and the only one worth writing today.
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "Organization",
"name": "Visilay",
"url": "https://www.visilay.com/",
"logo": "https://www.visilay.com/logo.png",
"sameAs": [
"https://www.linkedin.com/company/visilay"
]
}
</script>The property that matters most in this block is sameAs. It states that the company on the site and the LinkedIn profile, the Google Business Profile listing and the entry in a company register are the same entity. It is the kind of statement a search engine cannot work out on its own with certainty, and the one place where markup adds information instead of repeating what is already visible.
How to implement it without rebuilding the site
- On WordPress, Yoast SEO and Rank Math already generate the basic graph: Organization, WebSite, WebPage, Article and BreadcrumbList. For most company websites that is all you need, and it just has to be filled in properly in the settings.
- On WooCommerce, Product markup comes from the theme or a plugin. The properties that are almost always missing are the shipping and returns ones, and they are what makes the difference between a product snippet and a merchant listing.
- For cases that aren't covered, JSON-LD is added by hand in the template or via Google Tag Manager. This is the route for LocalBusiness with several locations, Event and JobPosting.
- The most common risk is duplicate markup: two plugins declaring two different Organizations on the same page. Before adding anything, check what is already there.
How to test it, and what will stop working
Google's Rich Results Test tells you whether a page is eligible for a supported rich result, and since June 2026 it no longer reports on FAQs. The Schema Markup Validator checks syntactic validity against the vocabulary, regardless of what Google displays: it is the right tool for types that exist only to be read by machines. The Enhancements report in Search Console is the only one of the three that works on the live site rather than on a single URL.
If you have dashboards or reports that pull FAQ metrics from the Search Console API, August 2026 is the deadline: after that, the call returns nothing.
The mistakes that cost you a manual action
The structured data general guidelines are short, and they are nearly always broken in the same three ways.
- Marking up content users can't see. The rule is literal: "Don't mark up content that is not visible to readers of the page". FAQs declared in JSON-LD but missing from the page fall under this.
- Reviews not written by real users. Google says ratings or reviews not produced by real users can lead to a manual action. The aggregateRating a company awards itself is the most frequent case on service business websites.
- Markup that doesn't represent the content. Declaring a page as Recipe when it contains no recipe, or a category page as Event.
The consequence is limited, and it is worth knowing because you read all sorts of claims: "A structured data manual action means that a page loses eligibility for appearance as a rich result; it doesn't affect how the page ranks in Google web search". You lose rich result eligibility, not rankings, which also disposes of one of the most widespread SEO myths. The distinction also matters when you assess the credibility of a site as a whole: fake reviews are a trust problem long before they are a markup problem.
Where half a day on it still pays off
Three situations, in order of return. An e-commerce site selling products with changing prices and availability: here complete markup changes how the result looks, and the figures in the table above give the order of magnitude. A business with one or more physical locations, where LocalBusiness works alongside the Google Business Profile listing. An editorial site with video, events or job ads, meaning content Google still shows in dedicated formats.
Outside these three cases, for the average company website the useful work is done in half a day: Organization filled in properly with sameAs, BreadcrumbList, Article on blog posts. The rest of the time pays back more when spent on content, page speed or the overall SEO strategy. Most of the sites I see have neither a catalogue to mark up nor an audience that arrives through carousels.
There is one thing the industry rarely says, and it explains why the two promises taken apart above were wrong from the start. Markup doesn't make anything happen: it makes something that was already on the page unambiguous. That is why it works well on data that changes, such as price, availability, opening hours or the date of an event, where a machine that misreads causes real trouble. And it does almost nothing for ideas, where there is no ambiguity to remove. When someone suggests marking up an opinion piece to get cited by a language model, they are asking a translator to improve the original text.
Frequently asked questions about structured data
No. Google uses it to understand the content of a page and to decide eligibility for rich results. Indirect confirmation is in the documentation on manual actions: a structured data manual action removes rich result eligibility and does not affect how the page ranks in web search.
No, there is no need. Google says unused structured data causes no problems for Search, and the FAQPage type remains valid in the Schema.org vocabulary: it simply no longer generates a rich result. If those FAQs were written only to get the accordion in the SERP, though, you may as well rewrite them starting from the questions you actually receive.
The available data says no. The Ahrefs study of 11 May 2026 followed 1,885 pages that added JSON-LD and measured +2.4% citations on Google AI Mode and +2.2% on ChatGPT, neither statistically significant, and a 4.6% drop on AI Overviews. Google states there is no special schema.org markup you need to add to appear in its AI features.
JSON-LD. It is the format Google recommends, it sits in a block separate from the HTML, it can be injected via JavaScript or a tag manager, and it doesn't break when the site theme changes. Microdata and RDFa live inside the page markup and have to be redone with every redesign.
It depends on how often the page is crawled and on whether all the required properties of the declared type are present. Markup makes a page eligible, it doesn't guarantee display: the final decision is Google's, and it can show a standard result even when the markup is correct.
