Structured data needed for AI search? Google answers

TL;DR: Google’s own documentation says you don’t need special schema.org markup to appear in AI Overviews or AI Mode, and an Ahrefs study of 1,885 pages that added JSON-LD schema found no meaningful citation lift on Google AI Overviews, AI Mode, or ChatGPT. Structured data still earns you rich results and helps machines parse your page correctly, so it stays worth doing, just not as an AI-citation lever.

The short answer
What Google’s documentation actually says
Inside the 1,885-page study
Why the FAQ schema deprecation confused everyone
What structured data still does for you
Structured data vs what actually moves AI citations
A decision framework for your time
Schema types still worth implementing in 2026
How to audit your structured data
Common structured data mistakes
Structured data and E-E-A-T signals
What to do instead if citations are the goal
Frequently asked questions
Key takeaways

The short answer

No, structured data is not required for AI search. Google says so directly in its own AI features documentation: “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” (Google Search Central, AI features and your website).

That single sentence ends the debate about whether you owe your AI Overview visibility to a missing <script type="application/ld+json"> tag. You don’t. Google’s systems read your rendered HTML, your prose, your headings, and your entity signals whether or not you wrap any of it in schema markup.

The harder question, and the one worth actually answering, is whether structured data helps anyway, even without being required. An Ahrefs study published May 11, 2026 tested that directly on 1,885 real pages and found the answer is mostly no (Ahrefs, “We Tracked 1,885 Pages Adding Schema”). Keep reading for the numbers, because “not required” and “doesn’t matter” turn out to be different claims with different evidence behind them.

What Google’s documentation actually says

Go to the source before you trust a summary of the source, including this one. Google’s AI features and your website page is the closest thing to an official policy statement on this question, and it says structured data is not a prerequisite for AI Overviews or AI Mode inclusion. No special AI-facing schema exists. No new file format. No markup checklist you’re secretly failing.

John Mueller, part of Google’s Search Relations team, reinforced the same point from the ranking side in an April 2025 Search Central office-hours recap: “Structured data won’t make your site rank better,” he told a site owner asking about schema’s ranking impact, adding that it’s used for the search features listed in Google’s structured data gallery, not for boosting position (Search Engine Roundtable, Daily Search Forum Recap, April 15, 2025).

Neither statement says structured data is useless. Google’s own structured data introduction page frames it as an enhancement layer: “Adding structured data can enable search results that are more engaging to users” and cites Nestlé’s implementation as an example, reporting an 82% higher click-through rate on pages with rich results versus pages without (Google Search Central, Introduction to structured data). That’s a real, measured business outcome. It’s a click-through-rate outcome in classic blue-link search, not a citation-rate outcome in an AI answer.

That distinction is the entire article. Structured data was built for rich results in traditional SERPs. Nobody built an AI-citation schema, because Google’s language models read the page like a person does, not like a parser hunting for JSON-LD.

Inside the 1,885-page study

Ahrefs ran the closest thing to a controlled experiment on this question that exists in public. Between August 2025 and March 2026, researchers Louise Linehan and Xibeijia Guan tracked 1,885 pages that added JSON-LD schema markup, all of which had at least 100 AI Overview citations as of February 2025 before the treatment (Ahrefs, “We Tracked 1,885 Pages Adding Schema”, reviewed by Ryan Law). They matched each treated page against three control pages from the same site or topic cluster that didn’t add schema, for roughly 4,000 controls, then ran four separate statistical tests: matched difference-in-differences, two-sample t-tests, and event studies across a 30-day window before and after the schema addition.

The results, measured across Google AI Overviews, Google AI Mode, and ChatGPT:

Google AI Overviews: citations dropped 4.6%, a small but statistically significant decline, working out to roughly 12 fewer daily citations per page on average.
Google AI Mode: citations moved 2.4%, which the study’s own confidence intervals classify as indistinguishable from random variation.
ChatGPT: citations moved 2.2%, again inside the noise band.

The study’s blunt conclusion: “Adding schema produced no major uplift in citations on any platform.” Not a wash that happens to look neutral. A small negative signal on Google’s own AI surface, and statistical noise everywhere else.

That’s worth sitting with, because it inverts a year of AEO advice telling you schema is a citation lever. It isn’t, at least not the generic JSON-LD most sites are adding. The study pooled Article, FAQ, Product, HowTo, and Organization schema together rather than isolating individual types, so it can’t tell you whether one specific schema type quietly outperforms the rest. What it can tell you is that bolting on schema as a standalone tactic, without changing the underlying content, does not move the needle you’re hoping it moves.

Why the FAQ schema deprecation confused everyone

Google added to the confusion in the same window the Ahrefs study was running. As of May 7, 2026, FAQ rich results stopped appearing in Google Search entirely, with Search Console reporting and the Rich Results Test dropping support in June and API support ending in August (Google Search Central, FAQPage structured data). The changelog entry is terse: the feature is gone from the search results page, full stop.

That reads like Google killing FAQ schema. It didn’t. The FAQPage type itself is still valid, still crawlable, still something you can add to a page. What disappeared is the visual rich result, the expandable accordion of questions in the blue-link SERP. The markup, and whatever signal it sends about page structure, is unaffected by the deprecation.

If you were adding FAQ schema purely to win that accordion in classic search, that reason to bother is gone as of August 2026. If you were adding it because a clearly labeled question-and-answer block helps a language model, human reader, or crawler parse a distinct claim on your page, nothing in the deprecation changes that. The mistake is treating the rich-result sunset as proof the underlying schema type stopped mattering for anything. It only stopped mattering for one specific SERP feature.

What structured data still does for you

Three things, none of them “gets you cited by ChatGPT.”

Rich results in classic search. Product, Review, Recipe, Event, and Job Posting schema still power star ratings, price ranges, and other SERP enhancements that measurably lift click-through rate, per Google’s own Nestlé case study above. If you sell products or run local listings, this alone justifies the implementation cost.

Machine-readable disambiguation. Structured data tells a parser explicitly what a number, name, or date means, rather than leaving it to be inferred from surrounding prose. A price is unambiguously a price. A rating is unambiguously a rating out of a defined scale. That reduces the odds of any system, AI or otherwise, misreading your page.

A forcing function for content clarity. Filling out Product schema requires you to actually specify a price, an availability status, a review count. Filling out FAQPage schema requires you to write a real question and a real, self-contained answer. The discipline of writing markup often improves the underlying content quality, and content quality is what actually correlates with AI citation, per every study on the topic including Ahrefs’ own.

None of these three benefits require you to believe schema is an AI-visibility lever. They stand on their own.

Structured data vs what actually moves AI citations

Signal	Moves AI Overview / AI Mode citations	Moves classic SERP rich results	Evidence
Adding JSON-LD schema to existing content	✗	✓	Ahrefs, 1,885-page study: -4.6% on AI Overviews, noise elsewhere
Rewriting the answer to be extractable in one paragraph	✓	✗ (no dedicated field)	Consistent across AEO practitioner analysis
Product schema (price, availability, rating)	✗ (not measured as a citation driver)	✓	Google Search Central: 82% CTR lift, Nestlé case
FAQ schema for the visual FAQ rich result	N/A (rich result retired May 2026)	✗ (deprecated)	Google, FAQPage docs
Third-party mentions on high-trust domains (Reddit, review sites, press)	✓	Indirect	See brand mentions as the new backlinks
Clear entity naming (consistent brand/person/product names site-wide)	✓ (helps disambiguation)	✓	Google AI features documentation
Generic Organization schema with no unique data	✗	✗	No measured lift in either surface

Read that table as a budget allocation problem, not a checklist. Every hour spent tagging generic Organization schema with your logo URL and social profiles is an hour not spent rewriting your top pages into an extractable, self-contained answer format, which is the thing every current study ties to actual citation gains.

A decision framework for your time

flowchart TD
    A[Do you sell products, list events,\nor run recipes/jobs pages?] -->|Yes| B[Implement Product/Event/Recipe/\nJobPosting schema for rich results]
    A -->|No| C{Is your goal AI Overview\nor ChatGPT citations?}
    C -->|Yes| D[Skip generic schema.\nRewrite content into\nself-contained, quotable answers]
    C -->|No, it's classic SERP CTR| E[Implement Article, Review,\nor Breadcrumb schema for rich results]
    B --> F[Validate with Rich Results Test\nand monitor Search Console]
    D --> G[Track citations directly:\nsee AI overview traffic recovery plan]
    E --> F

If you landed in the “rewrite for extractability” box, that’s not a schema project. It’s an editorial one: put the direct answer in the first sentence of the section, keep the paragraph self-contained enough to be lifted out of context, and name the specific fact, number, or mechanism instead of describing it vaguely. That’s the pattern every AI-citation study converges on, and it has nothing to do with markup.

Schema types still worth implementing in 2026

Not all schema is equally dead weight. Rank these by whether they power something you can still measure:

Product: price, availability, aggregate rating. Directly tied to shopping-result eligibility and the CTR lift Google documents. Keep this current; stale price data can trigger a Search Console warning.
Review / AggregateRating: star ratings in classic search remain one of the highest-CTR rich results available. Only use it on pages with genuine, verifiable reviews; Google’s guidelines explicitly prohibit self-serving or fabricated ratings.
Article: helps disambiguate author, publish date, and headline. Low effort, no measured citation harm, and it supports Google’s News and Discover surfaces independent of AI features.
Organization: worth doing once, site-wide, for brand entity consistency (name, logo, sameAs links to your verified social profiles). This is closer to entity hygiene than a citation lever, but entity consistency is one of the few factors Google’s own AI features page treats as relevant to how its systems understand who you are.
BreadcrumbList: cheap, mechanical, and still renders breadcrumbs in classic search results. No downside, minimal upside.
FAQPage: implement only where you have a genuine question-and-answer format your reader would actually search for, since the rich-result payoff is gone. Treat it as content structure, not an SEO trick.
HowTo: Google deprecated the visual HowTo rich result for most queries in 2023, well before this year’s FAQ sunset, so treat any lingering advice to add it purely for a step-by-step SERP card as outdated.

Skip generic, low-information schema entirely: Organization markup with no sameAs links, WebSite schema with a search-box action nobody uses, or Article schema duplicated across every template without page-specific values. That’s the schema equivalent of a keyword-stuffed meta description, work with no attached outcome.

How to audit your structured data

Four checks, in order:

Validate syntax. Run every template through Google’s Rich Results Test or the Schema Markup Validator. A single missing comma in JSON-LD can invalidate the entire block silently.
Check Search Console coverage. The Enhancements section under Legacy Report flags structured data errors and warnings by type. Fix errors first; warnings are often optional fields.
Match markup to visible content. Google’s guidelines require that structured data reflect what’s actually on the page. A Review schema block claiming a 4.8-star rating that isn’t visible anywhere in your rendered HTML is a policy violation, not a growth hack.
Confirm it’s still eligible. After the May 2026 FAQ deprecation, run a spot check on every template that still emits FAQPage markup purely for the rich result. If that was the only reason it existed, decide whether the content itself still earns its place on the page.

None of this audit work is about AI citations. It’s hygiene that protects the CTR gains structured data can still deliver in classic search, and it catches the kind of manipulative markup that draws a manual action.

Common structured data mistakes

Markup that doesn’t match visible content. The single most common cause of a Search Console structured data warning, and the fastest way to lose eligibility for the rich result entirely.
Copy-pasting a competitor’s schema wholesale. Different CMS, different field names required, near-guaranteed validation errors.
Adding every schema type “just in case.” More markup isn’t more signal. It’s more surface area for a validation error and more maintenance burden when a field goes stale.
Treating FAQPage as a citation hack post-deprecation. As covered above, the rich result is gone. Adding walls of thin Q&A pairs purely for markup, with no reader value, wastes effort that both Google’s guidelines and basic content strategy discourage.
Ignoring aggregateRating accuracy. If your review count or star average drifts from what’s actually shown to users, that’s a policy violation that risks the whole rich-result feature for the page, not just a missed opportunity.

Structured data and E-E-A-T signals

Structured data touches E-E-A-T indirectly, through the Organization and Person schema types that name an author, cite their credentials, and link to verifiable profiles. That’s entity clarity, which supports how confidently any system, human or AI, can attribute a claim to a real, checkable source. It is not the same as expertise or trustworthiness itself. Markup can describe a credential; it can’t manufacture one.

If your site struggles with authority signals and has no press mentions or backlink profile to lean on, structured data is a minor lever compared to the actual content and citation work covered in E-E-A-T for small sites with no brand. Don’t mistake filling out a Person schema block for building the credibility it’s supposed to represent.

What to do instead if citations are the goal

If the real objective is AI Overview, AI Mode, or ChatGPT citations rather than classic SERP rich results, redirect the effort schema won’t repay:

Write the direct answer first. Put the specific number, name, or mechanism in the first sentence of a section, then explain it. This is the single most consistently cited pattern across AEO research and matches what you’d find in why you rank on Google but stay invisible in ChatGPT.
Earn mentions on high-trust third-party domains. Reddit threads, review sites, and press coverage carry more weight in AI grounding than your own markup does; see brand mentions as the new backlinks for AI search and how to use Reddit for AI citations without getting banned.
Track your share of answer. Citation share is the metric that replaces click-through rate once the goal shifts from ranking to being quoted; see share of answer: the metric replacing share of voice.
Build genuine topical depth, not a single article trying to cover everything. See how to build topical authority when everyone publishes daily.
Measure citations directly, not proxy metrics like schema coverage. Set up referrer tracking per how to track ChatGPT and Perplexity traffic in GA4 and pull query-level detail from Google Search Console data analysis.
Run a full visibility audit before assuming schema is your gap at all; a lot of what looks like an AI-citation problem is actually a crawlability, freshness, or thin-content problem covered in the AI visibility audit checklist.
Skip the file-format detours. llms.txt doesn’t do what most people think it does either, and chasing it instead of content quality wastes the same kind of effort schema chasing does.

If you’re weighing this against a broader list of AEO tactics that sound plausible but don’t hold up, cross-reference against 11 AEO mistakes that keep you out of AI answers before you commit engineering time to any one of them. And if traffic already dropped and you’re trying to figure out where to spend the next sprint, the AI Overviews traffic loss recovery plan sequences this against the fixes that pay off fastest.

None of this means structured data is a waste of time everywhere. It means the specific claim “add schema to get cited by AI” doesn’t survive contact with the one controlled study that’s tested it. Keep the schema you have for the rich results it still earns you. Stop expecting it to do a second job it was never built for.

Frequently asked questions

Is structured data required for AI Overviews?

No. Google’s own AI features documentation states you don’t need any special schema.org markup or new machine-readable file to appear in AI Overviews or AI Mode (Google Search Central). Standard, crawlable, well-written HTML content is what its systems read and summarize.

Does adding schema markup improve ChatGPT citations?

Not measurably. Ahrefs tracked 1,885 pages that added JSON-LD schema and found ChatGPT citation rates moved 2.2%, a change the study classifies as statistically indistinguishable from noise (Ahrefs). Content extractability and third-party mentions carried more weight in that dataset.

Did structured data hurt AI Overview citations in the Ahrefs study?

The study measured a 4.6% decline in AI Overview citations after schema was added, described as small but statistically significant, roughly 12 fewer daily citations per page on average. It’s not proof schema causes harm broadly, but it firmly rules out schema as a citation booster on that platform in this dataset.

What happened to FAQ rich results in 2026?

Google stopped showing the FAQ rich result in Search results as of May 7, 2026, with Search Console reporting and the Rich Results Test dropping support in June and API support ending in August (Google, FAQPage documentation). The FAQPage schema type itself is unaffected; only the visual accordion in classic search disappeared.

Should I remove FAQ schema from my site now?

Only if you added it purely for the retired rich result. If the underlying question-and-answer content is genuinely useful to readers, keep the markup as a content-structure signal. If it exists solely as thin, repetitive Q&A pairs bolted on for the old rich result, that’s a good moment to rewrite or remove it.

Does John Mueller say structured data affects rankings?

No. In an April 2025 Search Central recap, Mueller told a site owner that structured data won’t make a site rank better and clarified it powers the specific search features listed in Google’s structured data gallery rather than boosting position (Search Engine Roundtable).

What’s the difference between structured data and llms.txt?

Structured data (schema.org JSON-LD) is a markup standard Google, Bing, and some AI systems can parse for specific fields like price or rating. llms.txt is an unofficial proposed file format some vendors promote as an AI crawler instruction file; Google has stated it doesn’t use it for ranking. Neither is required for AI search visibility; see the full llms.txt breakdown for the engine-by-engine truth.

Which schema types still have a measurable payoff in 2026?

Product, Review/AggregateRating, and Event schema remain tied to rich results with documented click-through-rate lifts in classic Google Search. Article and BreadcrumbList are low-effort and low-risk. Generic Organization or WebSite schema with no unique field data has no demonstrated payoff in either classic search or AI citations.

Does structured data help with E-E-A-T?

Indirectly. Person and Organization schema can name an author or entity and link to verifiable profiles, which supports clarity about who is making a claim. It doesn’t substitute for actual expertise, real credentials, or third-party validation, which carry far more weight in how any system judges trustworthiness.

Is Product schema still worth implementing for AI shopping features?

Yes, for the classic and shopping-adjacent search surfaces where price and availability data are directly consumed. The Ahrefs study didn’t isolate Product schema specifically from the pooled JSON-LD sample, so treat its AI-citation impact as unmeasured rather than proven negative, and keep it for the rich-result value alone.

How do I check if my structured data is valid?

Run each page template through Google’s Rich Results Test, then check the Enhancements section of Google Search Console for ongoing error and warning tracking by schema type. Fix errors immediately; they can disqualify a page from the associated rich result entirely.

Can bad structured data get my site penalized?

Markup that doesn’t match visible page content, or fabricated review ratings, violates Google’s structured data guidelines and can result in a manual action or loss of eligibility for that rich result across the site. It’s a policy risk, not just a wasted opportunity.

Do AI Overviews use structured data to select which pages to cite?

Google hasn’t published a mechanism confirming this, and its own AI features documentation explicitly says no special markup is needed to appear in AI Overviews. The Ahrefs study’s near-zero and slightly negative results are the closest empirical evidence available, and they don’t support structured data as a selection signal.

Is schema markup a ranking factor?

No. Google has repeatedly stated structured data is not a ranking factor; it powers specific SERP features rather than adjusting where a page sits in results. Any ranking improvement associated with adding schema is more likely explained by the content changes made alongside it.

What should I prioritize instead of adding more schema?

Rewriting your highest-traffic pages so the direct answer sits in the first sentence of each section, earning mentions on high-trust third-party sites, and building genuine topical depth on your core subject. All three have stronger evidence tying them to AI citation gains than markup does.

Does HowTo schema still show a rich result?

No. Google restricted the visual HowTo rich result to a small set of eligible sites starting in 2023, well before the 2026 FAQ deprecation. Adding it broadly for the old step-by-step SERP card is outdated advice; keep the content structure if it helps readers, not for the markup.

How long does it take to see results from adding structured data?

For classic rich results, Google typically needs to recrawl and reprocess the page, which can take anywhere from days to several weeks depending on crawl frequency. For AI citation purposes, the Ahrefs study measured a 30-day before/after window and found no meaningful lift to wait for.

Should ecommerce sites deprioritize structured data based on this research?

No. The research is specific to AI-citation lift, not to the CTR and shopping-eligibility benefits of Product and Review schema, which remain well documented. Ecommerce sites should keep that markup current and accurate; they should just stop expecting it to move AI Overview or ChatGPT citation rates.

Does adding structured data ever directly hurt SEO?

Only when it’s inaccurate, mismatched with visible content, or manipulative, which risks a manual action. Accurate, well-implemented schema that matches your page has no documented ranking or citation downside outside the specific 4.6% AI Overview citation decline the Ahrefs study measured, which the authors themselves describe as small.

Is it worth hiring someone specifically to implement structured data for AI search?

Not on the evidence available as of August 2026. Hire for structured data if the goal is classic SERP rich results, shopping eligibility, or content-structure discipline. If the stated goal is AI Overview or ChatGPT citations specifically, that budget is better spent on content rewrites and third-party mention building, per the sections above.

Key takeaways

Structured data is not required for AI Overviews, AI Mode, or ChatGPT citations, by Google’s own documented statement and by the only controlled study that has tested the claim directly. Adding JSON-LD schema to 1,885 real pages produced a small, statistically significant citation decline on Google AI Overviews and statistically insignificant noise on AI Mode and ChatGPT. Structured data still earns its keep for classic SERP rich results, shopping and review eligibility, and the content-clarity discipline that filling it out forces on you. The May 2026 FAQ rich-result deprecation removed one specific reason to add FAQPage markup, not the schema type’s broader validity. If AI citations are the actual goal, the effort belongs in rewriting content for direct, self-contained answers and earning mentions on trusted third-party domains, not in a markup checklist.

Run your own audit before deciding: validate what you already have with Google’s Rich Results Test, check Search Console for errors, and confirm every schema block still matches something a reader can actually see on the page. Then spend the time you’d have spent on more schema somewhere the evidence says it pays off.