---
title: "Structured Data for AI Search: What Schema Markup Actually Does (and Doesn't)"
description: "Google says schema isn't a ranking or AI-citation lever. Here's the real evidence on what structured data does — and doesn't — do for AI search visibility."
answer_summary: "Google says schema isn't a ranking or AI-citation lever. Here's the real evidence on what structured data does — and doesn't — do for AI search visibility."
canonical: "https://nqz.ai/blog/geo-structured-data-for-ai-search-useful-signals-without-overpromising"
published_at: "2026-07-18T09:56:14.222Z"
updated_at: "2026-09-12T13:29:08.359Z"
author: "nqzai Editorial Team"
category: "GEO"
tags: ["structured data","schema markup","AI search","GEO","JSON-LD","AI Overviews","technical SEO"]
image: "https://nqz.ai/blog/covers/geo-structured-data-for-ai-search-useful-signals-without-overpromising.webp"
---

# Structured Data for AI Search: What Schema Markup Actually Does (and Doesn't)

TL;DR

Adding schema markup to 1,885 pages produced no statistically significant boost in AI citations: Google AI Mode rose 2.4%, ChatGPT rose 2.2%, and AI Overviews actually declined 4.6% — all within the noise range. Google's own Search Advocate John Mueller has stated flatly that structured data "won't make your site rank better" and is only used for displaying rich results like star ratings, not for ranking or AI visibility.

A separate experiment by searchVIU found that when five AI systems (including ChatGPT and Gemini) fetched pages live, they all ignored JSON-LD entirely and extracted only visible HTML text

## What structured data actually is

Structured data is a standardized vocabulary — usually schema.org, encoded as JSON-LD — that describes what's on a page in a machine-readable format: this is a Product, its price is $40, its rating is 4.6 out of 178 reviews; this is a Recipe, it takes 30 minutes, it serves four. Google's own documentation defines its purpose narrowly: structured data helps Google understand page content and, when the data qualifies, display it as a "rich result" — a star rating under a search listing, a recipe card, a job posting panel ( Google Search Central, Intro to How Structured Data Markup Works ).

That's the whole job description. It is not a ranking signal, and — despite two years of GEO marketing claiming otherwise — there is no published evidence that adding it reliably increases how often an AI answer engine cites your page. This piece separates what Google, Bing, and independent researchers have actually said and measured from what schema.org markup has been sold as.

## What Google has actually said

Google's clearest statement on this comes directly from Search Advocate John Mueller, who has said it more than once. In April 2025, responding to a developer on Bluesky asking whether structured data helps SEO, Mueller wrote that structured data "won't make your site rank better" and that it's "used for displaying the search features listed in" Google's search gallery — nothing more. He added that using it for other schema.org purposes "won't cause problems, but you're unlikely to see any visible change from it in Google Search" ( Search Engine Roundtable, "Google: Structured Data Does Not Make Your Site Rank Better," April 2025 ; also reported by Search Engine Journal ). This wasn't new — Mueller made the same point in 2018 and again in 2023.

Google's written policy backs this up. The General Structured Data Guidelines state plainly that a structured-data violation "means that a page loses eligibility for appearance as a rich result; it doesn't affect how the page ranks in Google web search" — and separately, that "Google does not guarantee that your structured data will show up in search results, even if your page is marked up correctly" ( Google Search Central, General Structured Data Guidelines ).

On the AI side specifically, Google's dedicated page for AI Overviews and AI Mode is even more direct: "there's no special schema.org structured data that you need to add" to appear in those features. The guidance instead points site owners toward verifying in Search Console and following the same content-quality and technical-access basics that apply to classic search ( Google Search Central, AI Features and Your Website ).

Microsoft has said something more encouraging, but for a narrower system. At SMX Munich in March 2025, Bing's Fabrice Canel confirmed on stage that schema markup helps Microsoft's LLMs (which power Copilot) understand page content, since Bing's index feeds into its generative layer through a grounding process ( Search Engine Land, "Microsoft Bing/Copilot use schema for its LLMs" ). That's a real, on-record claim — but it applies to Bing's own indexing pipeline, not to how any AI system behaves when it fetches a page live.

## What the actual research shows

The most rigorous test to date is Ahrefs' 2026 study, which tracked 1,885 pages that added JSON-LD schema between August 2025 and March 2026, matched against 4,000 control pages, using a difference-in-differences design across four statistical tests. The result: no meaningful citation lift on any platform. Google AI Mode moved +2.4% (statistically indistinguishable from zero), ChatGPT moved +2.2% (also indistinguishable from zero), and Google AI Overviews actually saw a small, real but hard-to-attribute decline of -4.6%. The authors' own conclusion: "If the only reason you're adding it is to get more AI citations on pages that are already visible, our data doesn't support that bet." ( Ahrefs, "We Tracked 1,885 Pages Adding Schema. AI Citations Barely Moved." )

The important caveat, which Ahrefs states openly: every page in the study already had 100+ AI Overview citations before schema was added. It's a study of pages already inside the consideration set, not of pages trying to get discovered for the first time — so it doesn't settle the question for a brand-new or low-authority page.

A separate, mechanical explanation for the null result comes from searchVIU's October 2025 experiment. Researchers tested five AI systems — ChatGPT, Claude, Perplexity, Gemini, and Google AI Mode — fetching pages live, and found that none of them used JSON-LD during that direct-retrieval step; every system extracted only the visible HTML text, ignoring JSON-LD and hidden Microdata entirely ( SearchVIU, "Schema Markup and AI in 2025" ). That finding covers one specific moment in the pipeline — live fetch, not whatever happens further upstream when Google or Bing build their indexes — but it's a concrete reason schema-only pages don't reliably get quoted back correctly.

Otterly.ai ran a parallel controlled experiment across seven AI platforms and reported no isolated citation lift on ChatGPT after adding structured data, with only Google AI Mode and AI Overviews showing any gradual increase during the test window ( Otterly.ai, "GEO Experiment: Does Schema Markup Really Impact AI Search?" ). Search Engine Land's own synthesis of this research phrased it well: schema is a supporting technical signal, not a lever you pull for visibility ( Search Engine Land, "How schema markup fits into AI search — without the hype" ).

None of this is peer-reviewed academic research — it's SEO-industry experimentation, and every study has a caveat worth reading before you cite it. But three independent efforts converging on "no clear positive effect on already-visible pages" is a meaningfully different picture than the "add schema, get cited" claim that circulates in a lot of GEO content.

## Schema types: what they actually enable

Schema typeWhat it's officially forConfirmed ranking effectAI citation evidenceArticle / NewsArticle / BlogPostingEligibility for article-specific rich result treatmentsNoneNo isolated effect found in the Ahrefs 2026 studyProductPrice, availability, and shopping-result eligibilityNoneSome correlational studies suggest Product+AggregateRating pages get cited more for commercial queries, but this hasn't been isolated from confounds like site authorityReview / AggregateRatingStar ratings in search snippetsNoneSame caveat as above — correlation, not demonstrated causationFAQPageFormerly expanded FAQ snippetsNoneRetired by Google — see belowHowToFormerly step-by-step rich cardsNone (feature discontinued)Retired by Google — see belowLocalBusinessLocal pack / knowledge panel eligibilityNoneNot isolated in any published studyOrganizationKnowledge panel and entity disambiguationNonePlausible support role for entity grounding; not measuredBreadcrumbListBreadcrumb trail in the SERPNoneNot measured for AI citationVideoObjectVideo rich results, key momentsNoneNot measured for AI citationThe FAQPage and HowTo rows aren't hypothetical caution — they're the clearest illustration of the risk of treating a rich-result type as a durable asset. Google scaled back FAQ and HowTo rich results in August 2023, then formally deprecated FAQPage rich results in 2026, stopping the search appearance, the rich results report, and Rich Results Test support entirely ( Google Search Central Blog, "Changes to HowTo and FAQ rich results," August 2023 ). Sites that built content strategy around FAQ schema lost that surface with no ranking penalty attached — because, as Google states in that same post, the markup itself was never doing more than unlocking a display feature.

## How to implement structured data correctly (7 steps)

1. Start from the content, not the markup. Decide what's factually true and already visible on the page — price, author, rating, ingredients — before writing any schema. Markup that states facts not present in the visible HTML violates Google's policy and won't be trusted by AI systems either.
2. Pick the most specific type schema.org and Google both support. Use Article only for an article, Product only for an actual product page. Generic or mismatched types are more likely to be ignored or flagged.
3. Encode it as JSON-LD in a single script block (application/ld+json), placed in the page and kept in sync with the visible content whenever that content changes. Google explicitly recommends JSON-LD over Microdata or RDFa because it's easier to maintain correctly at scale.
4. Fill in the recommended properties with real values , not filler. Google's own guidance says results with genuine, complete optional properties — actual prices, actual review counts — are treated as higher quality than the bare minimum.
5. Validate before publishing using Google's Rich Results Test and the Schema Markup Validator, and fix every warning, not just the errors that block eligibility. For a fuller toolkit — including WordPress plugins, a JSON-LD sandbox, and enterprise-scale scanners — see our roundup of free structured data tools .
6. Confirm Google can actually see it with the URL Inspection tool in Search Console — a page blocked by robots.txt, noindex, or a login wall won't get credit for its markup no matter how correct it is.
7. Say the same thing in plain text, not just in the script block. Given that live-fetch AI systems have been shown to read visible HTML and skip JSON-LD, the facts that matter — price, rating, key specs — should also appear as ordinary readable text on the page, not only inside the structured data.
8. Monitor the Enhancements reports in Search Console , not rankings, to judge whether the markup is working — track eligibility and error counts, since that's the only thing structured data is designed to move.
9. Revisit the site's schema periodically. Google has quietly discontinued whole rich-result types (FAQPage, HowTo, and seven more types retired in June 2025). Markup for a feature that no longer exists isn't harmful, but it isn't doing anything either.

## What this doesn't guarantee

Direct answer: Structured data does not improve rankings — Google has said this on the record repeatedly, most recently in 2025. It does not guarantee rich-result display even when implemented correctly. It does not have a demonstrated, isolated effect on AI citation rates for pages that are already indexed and visible, per the largest controlled study available. And for at least five major AI systems tested in live retrieval, it is not even read at the moment the system fetches your page — the same information has to exist in plain visible text to be reliably picked up.

The honest position: structured data is a hygiene and eligibility mechanism, not a growth lever. It's worth doing correctly because it's low-cost, because it prevents you from losing rich-result eligibility you'd otherwise qualify for, and because it may support entity understanding upstream in ways current experiments aren't equipped to isolate. It is not worth doing as a substitute for the actual content quality, factual accuracy, and topical authority that both Google and independent research agree are the real drivers of AI citation.

## Where nqzai fits

Direct answer: Rather than treating structured data as a box to check, nqzai's site analysis looks at whether a page's schema actually matches what's stated in its visible text, flags markup for rich-result types that no longer exist, and tracks whether a page is actually appearing in AI answer citations over time — so the question "is this working" gets answered with observed citation data instead of an assumption baked into a checklist.

## FAQ

Direct answer: Does adding schema markup guarantee my page gets cited by ChatGPT or Google AI Overviews?

No. Google has stated structured data doesn't affect rankings, and the largest controlled study on AI citations (Ahrefs, 2026) found no meaningful citation increase on Google AI Mode, ChatGPT, or Google AI Overviews after adding schema to already-visible pages.

If schema doesn't help rankings or AI citations, why bother?

Because it's the only supported way to become eligible for rich results (star ratings, product info, recipe cards) in classic Google Search, it's low-cost to implement correctly, and it may support entity grounding in systems like Bing/Copilot, per Microsoft's own on-record statement — even though that effect hasn't been isolated in controlled testing the way Google's ranking claim has.

Do ChatGPT, Perplexity, and Gemini read structured data the same way Google does?

Not demonstrably. Testing in October 2025 found that all five systems tested (ChatGPT, Claude, Perplexity, Gemini, Google AI Mode) ignored JSON-LD during live page fetches and read only visible HTML text. Whether any of them use structured data further upstream, during indexing, hasn't been independently confirmed the way Bing's Fabrice Canel confirmed it for Copilot.

Should I remove old FAQPage or HowTo schema now that Google discontinued those rich results?

You don't have to — Google has said unused structured data doesn't cause problems. But it's also not doing anything for you anymore, so there's no reason to keep investing effort maintaining it.

Is JSON-LD better than Microdata for AI search specifically?

There's no evidence either format is read differently by AI systems, since neither is reliably read during live retrieval by the systems tested. Google's preference for JSON-LD is about maintainability for the Search product, not a documented advantage for AI visibility.

Can bad structured data get my page penalized?

It can make a page ineligible for a rich result via manual action, but Google's own guidance is explicit that this doesn't affect how the page ranks in web search.

## Evidence and scope

**Review date:** 2026-09-12.

**Reproducible use.** Use the framework with a defined audience, source data, and review date; test material recommendations against your own evidence before making a production or buying decision.

**Limit.** This article is educational guidance, not legal, financial, security, or performance assurance.