什么是 Schema 标记？定义与指南

Schema 标记是帮助搜索引擎理解内容的结构化数据标记。

SEO and Search

Schema markup is structured data written with the Schema.org vocabulary, usually as JSON-LD, that describes a page's entities and properties. Search engines can use it to understand content and judge eligibility for some rich results, but valid markup never guarantees display.

As of May 7, 2026, Google no longer shows FAQ rich results at all, according to Google's documentation changelog, yet FAQPage is still a valid Schema.org type.

That gap between what the vocabulary allows and what a search engine actually uses is the whole story. Valid markup isn't the same as useful markup. This entry covers what Google and Bing say schema markup does today, which features have disappeared, and why the real work now is keeping your markup true to the page it describes.

Schema Markup, Structured Data, and JSON-LD Aren't the Same Thing

People use these three terms as if they were one. They aren't.

Vocabulary Versus Format

Structured data is the broad idea: information about a page written so machines can read it without guessing. Schema.org's shared vocabulary project, founded by Google, Microsoft, Yahoo and Yandex, supplies the types and properties, such as Product, Organization, offers and author. The format is a separate choice. You can write the same vocabulary as JSON-LD linked data, as Microdata, or as RDFa.

Google supports all three formats and treats them as equally fine as long as the markup is valid and follows the documentation for each feature. It still recommends JSON-LD when your setup allows it, since a separate script block is simpler to add and keep current across many pages, as explained in Google's structured data introduction. JSON-LD sits in a script block, away from the visible text. That separation is convenient. That is also where mismatches can start.

Schema.org Types Versus Google-Supported Features

Schema.org is much bigger than Google Search. Google says its own Search Central documentation, not Schema.org, is the definitive guide to how Google Search behaves, and many Schema.org properties aren't required by Google but may still be useful to other services, tools and platforms.

A glossary page shows the gap clearly. Schema.org has a DefinedTerm type for formally defined terms, yet Google's list of structured data features has no guide for it. You can describe a glossary entry accurately with DefinedTerm and see no visible change in Google at all. Check this before anyone promises a client a new search feature.

Markup also connects entities through identifiers, linking an organization to its people, products and profiles. That entity layer deserves its own explanation, and our entry on Google's Knowledge Graph covers it. This page stays with the markup itself.

What Search Engines Say Schema Markup Does Today

Eligibility, Not a Guarantee

Markup makes a page eligible. It doesn't make anything appear. Google's page of general structured data guidelines says correct markup does not guarantee display in search results. The HTTP Archive 2025 SEO chapter describes structured data as not a direct ranking factor, which is the authors' assessment rather than a Google statement. Either way, none of the Google documentation reviewed for this entry promises ranking gains from markup, so you shouldn't promise them either.

Google AI Overviews and AI Mode

Google is unusually direct here. Its guide to optimizing for generative AI says "Structured data isn't required for generative AI search," and it lists overfocusing on structured data among the myths it wants site owners to ignore. The same guide still recommends keeping structured data as part of an overall SEO strategy, because it supports eligibility for rich results.

The reason lies in how these features find pages. Google's generative features rely on retrieval-augmented generation, which uses the core Search ranking systems to pull pages from the index, so a page has to be indexed and eligible to show with a snippet before any AI feature can cite it. Google adds that there's no special Schema.org markup to add, and that llms.txt files and other AI-specific markup aren't used by Google Search at all. Its separate AI features page also asks site owners to make sure structured data matches the visible text.

Bing and AI Answers

Bing frames it more warmly. In a November 2025 post on the Bing Webmaster Blog, Principal Product Managers Fabrice Canel and Krishna Madhavan wrote that structured content, including schema-marked product pages, helps AI systems "interpret and summarize your content more effectively." That's a statement about Bing's own systems. It isn't a measured uplift, and it doesn't describe how other AI assistants behave.

What No First-Party Source Has Confirmed

Neither Google nor Bing publishes a general figure for how much schema markup raises AI citations or rankings. Google's published click figures come from individual case studies, covered below. Google's answer is "not required," Bing's answer is "it helps interpretation," and nobody with access to those systems has put a general number on the effect. If someone quotes you a fixed percentage, ask who measured it, on which engine, and against what baseline.

The Rich Result Menu Is Shrinking While Data-Focused Support Grows

The clearest evidence of where schema markup is heading is Google's own changelog, which records every feature added and retired.

Features Google Retired or Narrowed Between 2023 and 2026

HowTo: documentation removed on September 14, 2023, after the rich result stopped appearing on desktop and mobile.
Sitelinks search box: documentation removed on November 29, 2024.
Breadcrumbs: shown only on desktop results since January 22, 2025.
Course info, estimated salary, learning video, special announcement and vehicle listing: documentation removed on September 9, 2025.
Practice problem: deprecated on November 5, 2025, and removed on January 6, 2026.
Dataset: clarified on November 5, 2025 as used only by Dataset Search, not Google Search.
FAQ: stopped appearing on May 7, 2026, with documentation removed on June 15, 2026.

Where Documentation Keeps Growing

Merchant shipping policy markup: added on November 12, 2025.
Discussion Forum and Q&A properties: expanded on March 24, 2026 to clarify comment thread structure.
Merchant listing category and sale duration: added on July 7, 2026, explicitly aligned with Merchant Center feed attributes.
Review snippet guideline: added on July 24, 2026 to address fake and undisclosed incentivized reviews.
VideoObject creator property: documented on September 24, 2026.

Put together, the list shows that the features that decorated the results page are going, while the properties that mirror product feeds and real records keep growing. That's our reading of the changelog rather than a Google statement, but the direction is hard to miss. Schema markup is turning from a visual trick into a data job, and data jobs fail when the data goes stale.

Implementing JSON-LD Without Creating a Second, Wrong Version of the Page

Every JSON-LD block is a second copy of facts that already live on the page. Two copies can drift apart.

Start From Visible, Verified Content

Google's rule is simple. Don't add structured data about information users can't see, even if that information is accurate, and don't create empty pages just to hold markup. Pick the type that matches the page's real subject, fill in the properties Google requires for your target feature, and treat every value as a claim you'd defend in front of the customer reading the page.

Templates, Plugins, and JavaScript

Most sites don't hand-write markup. Google notes that on CMS platforms such as Wix, WordPress and Shopify, structured data is often added through a settings page or a plugin, and that Google can read JSON-LD injected into the page by JavaScript. For Product markup, though, Google recommends putting it in the initial HTML for best results. The catch is that generated markup inherits whatever the template assumes. Google itself warns that pages can break after deployment because of templating or serving issues, which is why it points site owners to its rich result status reports instead of a single launch-day test.

Author Nodes

XenForo, the forum software company, offers a public example of markup being reviewed, with at least one fix queued at the source. In a thread on the XenForo community forum, Ryan Levering, who identified himself as handling structured data ingestion at Google, reviewed the platform's forum markup and suggested including profile URLs in author Person nodes rather than names alone.

As Levering put it, "Raw names are not nearly as useful for disambiguation."

A XenForo developer replied that a related fix, annotating posts with mainEntityOfPage, would ship "for the next release." Keep in mind that this is advice posted in a forum thread, not a Google policy page, but the lesson travels well: identifiers that point somewhere beat bare strings.

When Schema Markup Backfires or Doesn't Apply

Schema markup isn't always worth the effort. Sometimes it actively misleads.

Markup for content nobody can see. Describing reviews, prices or answers that aren't visible breaks Google's guidelines, and Google says markup that ignores its guidelines may become ineligible for rich result display.
FAQPage added only to win dropdowns. That display ended on May 7, 2026. FAQPage remains valid vocabulary for pages that genuinely contain questions and answers, but it no longer buys extra space in Google results.
Special markup for AI. Google says Search doesn't use AI-specific files or markup, so effort spent there won't help Google visibility.
Types without a Google Search feature, sold as rich result wins. DefinedTerm has no Google feature guide, and Dataset markup now serves Dataset Search only.
Markup that outlives the facts. A product page shows a new price while an older plugin block still declares the previous one. The code stays valid. The information doesn't.

The last case can be hard to spot because a validator can still report a pass.

Validating and Monitoring Markup

Three Tools, Three Different Questions

Each tool answers a different question, so you'll usually need more than one:

The Schema Markup Validator checks whether your markup is well formed against the Schema.org vocabulary.
Google's Rich Results Test tool checks eligibility for Google features and can preview some of them.
After launch, rich result status reports and the URL Inspection tool in Google Search Console show what Google actually found on live pages.

Why Valid Markup Still Breaks After Launch

Passing a test on publish day proves very little six months later. Theme updates, plugin updates and new product data can all change what the markup says without anyone touching it on purpose. A workable habit is to name one owner for structured data, rerun the checks after any template or plugin change, and spot check the fields that change most often, such as price, availability, authorship and business details. If you'd rather have this audited as part of a technical review, the SEO New Generation service is one option to discuss.

Measuring Whether Schema Markup Was Worth It

Google's Before-and-After Method

Google suggests a modest test. Take a few pages without structured data that already have several months of Search Console data, preferably pages that aren't seasonal, add the markup, confirm with URL Inspection that Google found it, and then compare performance in the Performance report over the following months. Google notes the test is tricky because page views vary for many reasons, so treat a single before-and-after change with care. For AI features, Google began rolling out Search Generative AI performance reports in Search Console in June 2026, starting with a subset of sites, and they're worth tracking alongside the standard reports if generative engine optimization work is part of your plan.

Reading the Famous Case Studies Correctly

Google's introduction lists four Google-reported examples. Two measured click-through rate: 25% higher for Rotten Tomatoes' enhanced pages and 82% higher for Nestlé pages shown as rich results. Food Network reported 35% more visits after enabling search features on 80% of its pages, and Rakuten saw users spend 1.5 times as long on marked-up pages. Rotten Tomatoes' implementation covered 100,000 unique pages.

Handle these numbers carefully. They're Google-reported, the year of each experiment isn't stated, and the Rotten Tomatoes and Nestlé figures compare different groups of pages, which isn't the same as measuring markup alone. They show what was possible for those companies. They don't predict your result.

Video Markup in Indonesia: The Vidio Case

Vidio, an Indonesian streaming service, offers a local example. According to Google's Vidio case study, published in June 2024, the company added VideoObject markup to a catalog of more than one million videos without changing its M3U8 video format, gave each video file a stable URL that Googlebot could fetch, and validated the result with URL Inspection. Within a year, Vidio saw about three times more video impressions and close to twice as many video clicks on Google Search.

There's an honest caveat in the same study. The number of videos Vidio published grew by about 30% between the first quarter of 2022 and the first quarter of 2023, so not all of the growth can be credited to markup. The markup also worked only because the video files themselves were accessible, which is the data integrity point again.

Frequently Asked Questions

What's the difference between schema markup and structured data?

Structured data is the general practice of describing a page in machine-readable form. Schema markup is structured data that uses the Schema.org vocabulary. JSON-LD, Microdata and RDFa are the formats you can use to write it.

Which format should I use for schema markup?

Google supports JSON-LD, Microdata and RDFa equally when they're valid. It recommends JSON-LD where possible because it is simpler to add and maintain across many pages.

Does schema markup guarantee rich results or better rankings?

No. Valid markup makes a page eligible for some features, but Google decides what to show, and its documentation doesn't promise ranking gains from structured data.

Does FAQ schema still produce FAQ rich results in Google?

No. Google stopped showing FAQ rich results on May 7, 2026, and removed the documentation in June 2026. FAQPage is still a valid Schema.org type, so it can still describe genuine question and answer content.

Do I need schema markup for Google AI Overviews or AI Mode?

No. Google says structured data isn't required for its generative AI features and there's no special Schema.org markup to add. Pages still need to be indexed and eligible to appear with a snippet.

How do I test schema markup?

Use the Schema Markup Validator for vocabulary checks and Google's Rich Results Test for Google feature eligibility. After launch, monitor rich result status reports and use URL Inspection on live pages in Search Console.