Does schema help Google AI Overviews?

What structured data can and cannot do

Google's position on AI Overviews and AI Mode is direct: there are no additional technical requirements beyond normal Search eligibility, and you do not need new machine-readable files, AI text files, or special markup to appear. Adding schema does not submit a page to those features, make it rank, or make a citation happen.¹

That does not make structured data pointless. Google separately describes it as a standard way to provide explicit clues about a page, and notes that schema.org properties outside Google's supported Search features may still be useful to other search engines, services, tools, and platforms. Its defensible job is to label facts that already exist on the page so a parser has less ambiguity to resolve. It cannot make a weak claim true, authoritative, competitive, or observed in an answer.

In Robot Visible's evidence model, valid matching schema belongs to Eligible: an intended parser can access and interpret a declaration. It does not establish Competitive evidence, and it never counts as Observed. Only a captured answer with its query, provider, surface, source URL, and timestamp can establish a citation.

Three layers, with visible content first

A useful implementation order starts with what a person can read, then adds labels and discovery controls. Missing schema does not automatically leave a capable search engine or assistant guessing: clear text can be understood without it. Schema is most useful when the relationship between otherwise clear facts would benefit from an explicit type or property.

What changed for FAQPage in 2026

Google stopped showing FAQ rich results on 7 May 2026. This completed a long reduction: in 2023 Google had already limited the display largely to well-known government and health sites, so most commercial sites had received no regular FAQ rich result for nearly three years.

The display feature and its reporting are gone; the schema.org FAQPage vocabulary is not. Those are different things. Google Search no longer offers the expandable FAQ treatment, while a standards-aware non-Google parser can still read an FAQPage declaration. That possibility is not evidence that any particular assistant uses it for retrieval or citations.

Is FAQPage still worth implementing?

Sometimes, but the reason must survive the loss of Google's rich result. Keep a genuine visible FAQ when it helps a reader make a decision or resolve a recurring question. Add FAQPage only when the markup is generated from those same visible questions and answers, a known consumer needs it, or the maintenance cost is low enough to justify standards-based reuse.

Do not add it to chase extra Google result space, promise AI Overview inclusion, or turn thin filler into an apparent knowledge base. There is no published evidence that FAQPage increases the chance of an AI citation. If the JSON-LD is a second hand-maintained copy with no known consumer, removing it is a reasonable maintenance decision; the useful on-page questions can stay.

Robot Visible currently emits FAQPage on guides that have visible Quick answers because both outputs come from the same question-and-answer objects. That avoids a second factual copy and leaves the content available to schema.org-aware parsers at low maintenance cost. We do not treat the markup as a Google AI requirement or as evidence that any answer engine used it.

The schema types that earn their maintenance cost

Choose types from what the page visibly and truthfully contains, not from a wish list of search features. A small accurate graph is more useful than a large speculative one.

• Organization — identify the business consistently, including its canonical URL, logo, and official profiles where appropriate.
• SoftwareApplication, Product, or Service — describe what the page is actually about. Add prices or Offer only where the visible page makes the same offer; do not mark up a competitor's price as your own.
• Article — connect a guide to its named author, publisher, publication date, meaningful modification date, and cited sources.
• BreadcrumbList — declare the page's place in a real, navigable site hierarchy.
• FAQPage — label genuine visible questions and publisher-written answers for standards-aware parsers, with no Google FAQ rich-result expectation.
• One rule above all: the markup must match the visible page. Structured data is a declaration about content, not a hidden place to make a stronger claim.

llms.txt: useful, but know what it is

llms.txt is an emerging convention: a plain-text index at the site root that points to useful pages. In June 2026 Google clarified that it does not use llms.txt for Search and that maintaining one has neither a positive nor negative effect on Google visibility or rankings. Google's AI-features guide also says an AI text file is not required for AI Overviews or AI Mode.

Another service may choose to consume the file, so publishing one can still be a cheap compatibility decision. Treat it as an optional index for named consumers, not as a strategy or a substitute for crawlable pages, clear visible facts, internal links, and supported structured data.

Implement and measure without overclaiming

Start with one self-contained sentence a reader can verify: what the thing is, who it is for, and the concrete fact the page supports. Then express that same information with the most specific applicable schema.org types and properties. Generate visible copy and JSON-LD from one source where possible, so a correction updates both.

Validate syntax with the Schema.org validator. Use Google's Rich Results Test only for structured-data features Google currently supports; it is no longer an FAQPage eligibility test. Then inspect the rendered page and the markup together for factual agreement.

Measure each outcome separately. A successful fetch and valid matching markup are eligibility evidence. Stronger coverage, sources, and corroboration are competitive evidence. A mention or citation is observed only after a live answer check captures the query, provider, surface, source URL, and time. Schema can be part of a change worth testing, but it cannot be credited with an outcome the evidence did not isolate.

Run a free scan to inspect your visible content and schema