Schema Markup for AI Search: What Actually Works in 2026

Schema markup does not guarantee a citation in ChatGPT, Perplexity, or Google AI Overviews, and Google said so directly in its own 2026 guidance. What it does is remove ambiguity: clean JSON-LD tells a crawler or a model exactly what an entity is, who wrote it, and how it connects to everything else on your site, which matters more as answer engines pull from fewer, more confident sources.

That is a less exciting answer than most of the content on this topic gives you, and it is the accurate one. There has been a wave of blog posts in the last year claiming specific multipliers, like content with schema is some fixed number of times more likely to get cited by an AI model. Those numbers do not trace back to any named study, dataset, or methodology anywhere. This piece only uses figures and claims that come from a source you can actually check, and where the honest answer is “we do not fully know yet,” it says that instead of inventing a number to fill the gap.

What Schema Markup Actually Is

Schema markup is structured data written in a vocabulary defined at Schema.org, embedded in a page so machines can read what a human would otherwise have to infer from context. Instead of a crawler guessing whether a page is a recipe, a product listing, or a how-to guide from unstructured text and layout, schema states it explicitly: this is an Article, published on this date, by this author, about this topic, referencing these other entities.

The dominant format today is JSON-LD, a script block that sits separately from your visible HTML rather than being woven into individual tags the way older microdata and RDFa formats required. Google recommends JSON-LD specifically because it is easier to implement and maintain without touching your page’s visual markup, and all three formats remain technically valid and equally acceptable to search engines when implemented correctly.

The part that matters for this article: schema was built for search engines and has been part of SEO for well over a decade. What changed is who else is reading it. Large language models and their retrieval systems, GPTBot, PerplexityBot, Google’s AI Overviews pipeline, and Bing Copilot’s crawlers, all encounter the same JSON-LD blocks. The question worth answering is not whether schema exists for AI, it is whether these systems actually use it, and how much that use translates into visibility.

What the Platforms Themselves Say

Start with the only source that can speak with any authority about its own system: the platforms. In May 2026, Google’s Search Central team published an official guide titled “Optimizing your website for generative AI features on Google Search.” It includes a section explicitly titled “Mythbusting generative AI search: what you don’t need to do,” and structured data is one of the myths it addresses head-on:

“Structured data isn’t required for generative AI search, and there’s no special schema.org markup you need to add.”

Google goes further than that single line. The guide states plainly that you don’t need to create new machine-readable files, AI-specific text files, or special markup formats to appear in AI-generated results, and that Google Search ignores files like llms.txt entirely: creating one “will neither harm nor help your site’s visibility or rankings in Google Search.” At the same time, the guide does not say schema is worthless. It recommends continuing to use structured data as part of a normal SEO program because it still helps with eligibility for conventional rich results on Google Search, which is a real and separate benefit from AI citation.

Microsoft has taken a more direct position on the AI side specifically. Bing confirmed in March 2025 that its models, including Bing Chat and Copilot, use schema.org markup to help understand page content. That is a narrower and more concrete claim than anything Google has said publicly about how AI Overviews weighs structured data, and it is worth treating the two platforms differently rather than assuming one confirmed practice applies everywhere. OpenAI and Anthropic have not published documentation on how their crawlers handle structured data, so any claim about how ChatGPT or Claude specifically weighs schema is not something you can verify against a primary source right now, which means it should not be stated as fact by anyone, including us.

If your content strategy needs a foundation before schema even enters the conversation, our AI SEO services start with the same fundamentals Google’s own guide points back to: crawlability, genuine originality, and clear entity structure, before layering in the technical work. If you would rather start by seeing where your own site currently stands with AI crawlers, a GEO audit is a faster read on that than guessing from a blog post.

What the Actual Research Shows

Beyond platform statements, a small number of independent tests exist, and they are more cautious than the marketing claims circulating about them.

Search Engine Land ran a controlled experiment published in September 2025, authored by Molly Nogami and Ben Tannenbaum. They built three nearly identical pages: one with well-implemented schema, one with poor schema, and one with none. The page with well-implemented schema was the only one of the three to appear in a Google AI Overview, and it also achieved the highest organic ranking, hitting position 3, while the poorly-schema’d page peaked at position 8 and the unmarked page was not indexed at all. That is a genuinely interesting result. But the authors themselves were careful about what it proves, writing that they do not consider the results “absolute proof that well-implemented schema plays a role in AI Overview presence,” and noting the sample size was small enough that they planned follow-up testing, including swapping schema between the pages to isolate the variable further. That is the correct level of confidence to have in a three-page test, and it is more honest than most of what gets written about this topic.

A separate study cited in Search Engine Land’s coverage, run by Search/Atlas in December 2024, found no correlation between how much schema markup a site had implemented and how often it got cited by AI systems. That is a meaningfully different finding from the September 2025 test, and the two results together suggest something closer to the truth than either one alone: presence of schema is not the variable that matters, quality and correctness might be, and neither has been proven at scale yet.

There is also older, more general research worth citing outside the SEO industry entirely. A February 2024 study published in Nature Communications found that large language models extract information more accurately when given structured prompts with clearly defined fields, compared to unstructured instructions. That is not a study about schema markup or AI search citations specifically, but it supports the underlying mechanism people assume schema relies on: models handle explicit, labeled structure better than they handle inference from prose. It is a reason to expect schema could matter, not proof that it currently moves citation rates in ChatGPT or Perplexity.

Put together, the honest state of the evidence in mid-2026 is this: one small controlled test suggests a link between schema quality and AI Overview inclusion, one larger correlational study found no link between schema volume and AI citation rate, and no platform has published data quantifying the effect either way. Anyone telling you they know the exact lift schema markup provides for AI search visibility is not working from a source that exists.

A GEO program built on guesses instead of verified findings burns budget on the wrong things. If you want a second opinion on where your site actually stands before investing in a markup overhaul, a strategy call is a faster way to find out than another round of speculative blog posts.

FAQ Schema Is a Different Story, and It Just Changed

This is the part of the schema conversation most 2025-era advice gets wrong now, and it is a concrete, sourced correction rather than a matter of interpretation.

FAQ rich results, the expandable question-and-answer dropdowns that used to appear directly in Google Search results, were already restricted once. In August 2023, Google narrowed eligibility to a small set of well-known government and health websites after widespread abuse by sites stuffing FAQ schema onto pages purely to gain extra search real estate. Then, on May 7, 2026, Google removed FAQ rich results from Search entirely, ending eligibility even for the government and health sites that had kept it. Google added a notice directly to its FAQ structured data documentation: “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.” Google did not publish a separate blog post explaining the reasoning, and as of this writing has not clarified whether it relates to how AI Overviews now handle question-and-answer content instead.

The practical takeaway is not that FAQ content stopped mattering. It is that the rich result reward for FAQPage markup specifically is gone. FAQPage remains a technically valid Schema.org type, and Google has said it will continue parsing it to understand page content, but if your team is still implementing FAQ schema for the SERP dropdown, that outcome no longer exists. Keep writing genuinely useful FAQ sections for the reader and for topical coverage. Stop treating FAQPage markup as a rich-result tactic, because as of May 2026, it isn’t one.

The Five Things Google Says You Don’t Need to Do

Google’s May 2026 guide is unusually direct for an official platform document, because it names specific tactics agencies and consultants have been selling and says not to bother. The mythbusting section lists five practices as unnecessary for generative AI search visibility:

• llms.txt and similar AI-specific text files. Google Search ignores them entirely, and creating one has no effect, positive or negative, on rankings or AI feature eligibility.
• Content chunking for AI consumption. Restructuring pages into isolated, AI-optimized fragments is not something Google’s systems require or reward.
• AI-specific content rewrites. There is no separate version of your content that performs better for AI Overviews than the version written well for a human reader.
• Inauthentic mentions and citations. Manufacturing brand mentions or references across the web to game entity recognition does not work the way some GEO vendors imply.
• Overusing structured data. Piling on schema types beyond what accurately describes your content does not improve AI visibility, and Google is explicit that no special schema.org markup is required at all.

Read those five together and a pattern shows up: every one of them is a shortcut that promises AI-specific results without doing the underlying work of being a genuinely useful, well-structured, credible page. Google’s own framing for the whole guide is that optimizing for generative AI search “is still SEO,” built on the same foundation of crawlable pages, original content, and clear structure that has always mattered, with a sharper bar now for content that is not just competent but the kind of primary, specific material a model cannot simply reassemble from ten other pages that already exist.

Which Schema Types Are Actually Worth Your Time

None of the caution above means skip schema markup. It means implement it for reasons that are actually true instead of ones that are not. Here is what holds up, organized by what each type is verified to do rather than what it is rumored to do for AI citations specifically.

Notice what is missing from that “verified benefit” column: any entry that says “increases AI citation rate.” That is intentional. The table above lists what these schema types are documented to do. Where AI-specific effects are unproven, the honest move is to leave the space blank rather than fill it with a number nobody can source.

The case for implementing this list anyway is straightforward without needing an AI multiplier to justify it. Clean Organization and Article schema is close to free once your CMS supports it, it removes ambiguity for every crawler that touches your site including the AI ones, and it earns real, measurable rich-result placement in classic Google Search, which is not a hypothetical benefit. If schema quality does turn out to correlate with AI Overview inclusion the way Search Engine Land’s early test suggests, you get that upside for free by doing the work correctly the first time. If it doesn’t, you still got a legitimate SEO improvement.

How to Implement It Without Making It Worse

Bad schema is worse than no schema, because it introduces contradictions between what your markup claims and what your page actually shows, and that gap is exactly the kind of signal an authoritative system is built to distrust.

A few practical rules keep implementation clean:

Match markup to visible content exactly. If your JSON-LD claims an AggregateRating of 4.8 stars from 200 reviews, that number needs to be visible and accurate on the page itself. Markup that overstates or invents data violates Google’s structured data policies and risks manual action, separate entirely from any AI consideration.

Use JSON-LD, placed in the page head or body, not woven into inline attributes. It is easier to generate programmatically, easier to audit, and easier to keep in sync with page content when it lives as a single script block rather than scattered across dozens of HTML tags.

Validate before you publish, not after. Google’s Rich Results Test and the Schema.org validator both catch structural errors, missing required fields, and type mismatches before they reach production. A broken schema block that a validator would have caught in thirty seconds is a common, avoidable failure.

Keep entity data consistent across your whole site. If your Organization schema, your author bios, and your About page describe your company or contributors differently, you are handing every reading system, human or machine, a small contradiction to resolve. Consistency is a cheap way to reinforce the entity clarity that both Google’s guide and Bing’s confirmed usage point to as what actually matters. If you want a shared reference for terms like entity, structured data, and JSON-LD across your team, our marketing glossary keeps definitions in one place so they do not drift between whoever writes copy and whoever ships markup.

Don’t chase novelty schema types with no real audience. Speakable schema, for instance, was built for voice assistant use cases that never reached meaningful scale. Time spent implementing niche, thinly-supported schema is time not spent getting your core Article, Organization, and Product markup fully correct.

Working through this list against a real site, most teams find the gap is not the exotic schema types, it is that Organization and Article markup were set up once years ago and never updated as the company, authors, or product line changed. If you want a second set of eyes on what your current markup actually says versus what your site actually shows, book a call and we will walk through it with you.

Where Schema Fits Into a Real GEO Program

Schema markup is infrastructure, not strategy. It supports a page’s ability to be understood correctly once a crawler or model reaches it, but it does not create the content worth citing, the topical depth that makes a page the best answer to a question, or the credibility signals that make a model trust your domain over a competitor’s. Those are the things Google’s own 2026 guide identifies as what actually drives inclusion in generative results, and they are also where most of the visibility gap between sites actually gets decided.

If you have not audited where your site currently stands in AI search results, that is a more useful starting point than a schema overhaul on its own. Our GEO audit checklist walks through exactly how to check your current ChatGPT and Perplexity citation visibility, and our broader guide to generative engine optimization covers the fundamentals schema sits underneath: content structure, entity authority, and the technical groundwork that makes any of this work reliably.

None of this means treat schema as optional busywork either. Implemented correctly, it costs relatively little, it produces confirmed benefits in conventional search and on Bing’s AI surfaces, and it removes a category of ambiguity that costs you nothing to close. The mistake is expecting it to be a lever you pull for a guaranteed AI citation bump, because right now, no platform, no study, and no honest practitioner can back that promise with a real number. If you want help sorting the schema work worth doing from the tactics Google has already told the industry to stop selling, our team can walk through your specific site and tell you which is which. Book a call and we will show you exactly where the gap is.

Alex Carter LinkedIn

SEO & Content Strategy, MV3 Marketing

Alex Carter leads SEO and content strategy at MV3 Marketing, specializing in generative engine optimization, technical SEO, and AI-driven content systems for B2B companies.