Schema Markup for SEO: What Actually Gets You Cited by AI

A marketing lead we spoke with had spent most of 2023 and 2024 wiring FAQ schema into every blog post and service page on the site, chasing the little accordion rich result that used to show up under a handful of lucky listings. Then, without a blog post or an announcement, Google stopped showing FAQ rich results altogether in 2026. The feature the team had been optimizing for simply stopped existing. When someone on the team asked ChatGPT the same question their FAQ block answered, the brand was not mentioned at all, despite the schema being technically flawless.

That gap between having correct schema and actually showing up in an AI-generated answer is the real question most businesses are asking right now, whether they phrase it that way or not. This article goes past the generic checklist and covers what schema markup can and cannot do, which platforms have actually confirmed using it, which schema types are worth the engineering time in 2026, and how to measure whether any of this is doing anything at all.

What Is Schema Markup, and What Job Is It Actually Doing on Your Site?

Schema markup is a shared vocabulary, maintained through schema.org, for describing what a piece of content actually is in terms a machine can parse without guessing. A paragraph of text tells a human reader that a business is open Tuesday through Saturday. Schema markup tells a crawler, in an explicit, structured format, that this specific string is an openingHours property attached to a LocalBusiness entity. Most sites implement it as JSON-LD, a block of code that sits in the page’s HTML but is invisible to a visitor, since it is written for machines rather than for people scrolling the page.

The job schema does has not really changed since it was introduced. It removes ambiguity. Search engines and, increasingly, the retrieval systems behind generative AI tools have always had to infer meaning from unstructured prose, and inference is where mistakes happen. Schema is the shortcut that tells a system definitively who published something, what type of entity is being described, and how its parts relate to each other, rather than leaving that judgment to a language model’s best guess.

Does Adding Schema Markup Directly Move Your Google Rankings?

No, and Google’s own documentation is direct about this. Structured data is not treated as a general ranking factor. What it earns instead is eligibility, a page becomes a candidate for certain enhanced search features, provided the content backing the markup actually meets the quality bar for that feature. A product page with flawless Review schema still will not outrank a thin, poorly written competitor purely because the markup is technically correct.

This distinction matters because it is where a lot of schema implementation goes wrong from the start. Teams treat markup as a lever that pulls rankings up on its own, get frustrated when a page with perfect schema still sits on page two, and conclude schema does not work. Schema was never supposed to do that job. Its actual value shows up downstream, in whether a system can confidently extract and reuse what is already on the page, which is a different problem than whether the page ranks in the first place.

Why Did Google Just Kill FAQ Rich Results, and What Does It Mean for Your Schema Plan?

Google had already been narrowing FAQ rich results since 2023, restricting the accordion-style display to a small set of well-known government and health sites. In 2026, Google finished the job. FAQ rich results stopped appearing in search altogether, the FAQ search appearance filter and rich result report are being retired, and Search Console API support for the feature is being phased out entirely. Google gave no public explanation beyond updating its own documentation, which is itself worth noticing, since a feature this widely implemented disappeared with no blog post and no warning.

This does not mean FAQ schema is now pointless, and it does not need to be ripped out of existing pages. The visual reward is gone, but the underlying value, an unambiguous machine-readable version of a Q&A section, still exists for any system parsing the page for extractable answers. What the change actually exposes is a planning mistake: building a schema strategy around a specific visible SERP feature is fragile, because that feature can be pulled by a platform you do not control at any time. Schema that reflects the genuine structure of your content holds its value regardless of whether a search engine decides to display it that particular way this year.

Which AI Platforms Have Actually Confirmed Using Structured Data?

This is the part most schema advice skips, and it is the part that actually determines where your engineering time should go. Google confirmed in 2025 that structured data gives its AI Overviews feature an advantage in surfacing and summarizing content, which lines up with AI Overviews drawing heavily on Google’s existing, structured-data-aware search index. Microsoft has separately confirmed that schema helps Bing Copilot understand page content more reliably. Neither ChatGPT nor Perplexity has made an equivalent public statement, despite both platforms having the technical capability to parse structured data during retrieval.

A December 2024 industry study looking specifically at the relationship between schema markup coverage and AI citation rates found no meaningful correlation between the two across the sites it examined. That finding does not mean schema is worthless. It means schema alone, absent clear, well-structured, corroborated content, is not sufficient to earn a citation, which is a very different claim than the one most schema checklists imply. Our approach to AI search visibility treats schema as one input among several rather than the whole strategy, precisely because the data does not support treating it as the whole strategy.

Which Schema Types Are Actually Worth Your Engineering Time?

Organization and Person schema tend to matter more than teams expect, because they do the specific job generative engines struggle with most: confirming who is actually behind a piece of content. Attaching a sameAs property linking your Organization entity to verified profiles elsewhere on the web gives a language model a corroboration signal it cannot get from prose alone, which is exactly the kind of cross-referencing that influences whether a system trusts a source enough to cite it.

Article schema, specifically the headline, author, and datePublished or dateModified fields, matters for a different reason. Retrieval systems built around freshness, Perplexity in particular, weigh recency heavily, and Article schema is the cleanest way to state exactly how current a page is without forcing a system to guess from context clues in the body text. Product, Offer, and AggregateRating schema still carry real weight for ecommerce, since Google’s shopping-oriented AI features lean on that structured pricing and review data directly rather than trying to extract it from a product description.

FAQPage schema still earns a place on the list, just not for the reason it used to. With the rich result gone, its value is purely structural: it hands a retrieval system a clean question-and-answer pair instead of forcing it to identify one inside a paragraph. HowTo schema, which Google restricted around the same time as FAQ schema, sits in a similar spot: still useful as machine-readable structure, no longer useful as a guaranteed visual reward.

The following example is illustrative and not a real client engagement. Assume a services business publishes 40 articles a year with no Article or Organization schema at all, and assume that gap contributes to a modest, hard-to-isolate drag on how consistently AI Overviews and Bing Copilot surface the brand as a source, since both platforms have confirmed weighing structured data. Even a conservative assumption, that complete entity and article schema recovers a few additional citation appearances a month across a business’s core queries, compounds meaningfully over a year once multiplied across dozens of published pages. The number is unverifiable in isolation. The direction, that missing entity schema is a real but modest tax rather than a catastrophic one, is the more defensible takeaway.

Why Do Most Teams Get Schema Markup Wrong?

The most common mistake is markup that does not match what is actually on the page. Schema describing a five-star average rating for a product with three reviews, or an FAQ block listing questions that were never actually answered in the visible content, is not a shortcut. It is a mismatch that search engines and AI retrieval systems are both built to catch, and a pattern of mismatches erodes the trust a source needs to ever get cited at all.

A second mistake is chasing every schema type available rather than prioritizing the ones that reflect the actual business. A local service business implementing Recipe schema because a plugin made it easy gains nothing, since the markup has no relationship to what the business does. In our experience, the businesses that see the most benefit from structured data are the ones that implement a small, accurate set of entity and content schema deeply and correctly, rather than a large, shallow set of every available type.

Schema markup that does not match the page is not neutral. It is a trust signal working against you, whether the reader is a search engine, a language model, or eventually a person who notices the mismatch themselves.

The third failure mode is never validating the markup after it ships. JSON-LD can break silently during a site migration, a template update, or a CMS plugin change, and because it is invisible to a visitor, nobody notices until an audit turns it up months or years later. A page can carry broken or empty schema for a long stretch of time while everyone assumes it is working exactly as designed. Getting this right across an entire site, rather than one page at a time, is usually where a proper technical SEO program earns its cost over a one-time schema push.

How Do You Actually Measure Whether Schema Markup Is Working?

Validation comes first, and it is a binary check rather than a judgment call. Running a page through a structured data testing tool confirms the markup is present, correctly formatted, and free of errors, but it says nothing about whether that markup is influencing anything downstream. Treat it as a floor, not a success metric.

Search Console’s rich result and enhancement reports show whether Google is actually reading and using the markup you have shipped, which is a meaningfully different question than whether the markup is merely valid. For AI visibility specifically, the only direct measurement available right now is manual prompt testing, running your core queries through ChatGPT, Gemini, and Perplexity on a regular schedule and logging whether your brand is cited, mentioned without citation, or absent, since none of these platforms currently offers anything like a Search Console equivalent.

Tracking citation frequency before and after a schema implementation, alongside referral traffic segmented from chatgpt.com, perplexity.ai, and Gemini-related referrers, is the closest thing to a real feedback loop that exists today. Reviewing what that measurement actually looked like across different accounts is easier with real examples in front of you, and a look through our documented case studies shows a few different versions of that tracking in practice.

Frequently Asked Questions

Sources