Structured data is widely misdescribed as a ranking factor. Google has been consistent on this: it does not directly boost rankings. What it does is make a page eligible for particular search result treatments, and eligibility is worth pursuing because those treatments change click-through behavior.
The useful question is therefore not "should we add schema" but "which types currently produce a visible result, and does our content genuinely qualify."
Types that reliably still produce something
Product — price, availability, and review data in shopping-adjacent results. High commercial value, and one of the most strictly enforced types. Prices and availability must match the visible page.
Review and AggregateRating — star ratings, where they are permitted. The eligibility rules narrowed significantly: self-serving reviews of your own business embedded on your own site have been excluded from rich result treatment. Reviews of products you sell remain eligible.
Recipe, Event, JobPosting, Course — vertical types with dedicated result formats. Genuinely valuable if you have the underlying content, entirely irrelevant if you do not.
Breadcrumb — replaces the URL string in the result with a readable hierarchy. Modest, near-universally applicable, and very cheap to implement.
Video — thumbnails and key moments in results. Increasingly valuable as video surfaces expand.
Organization and LocalBusiness — feeds entity understanding and knowledge panel data. No direct rich result for most sites, but it is the type most relevant to being correctly identified as an entity, which matters more as AI search grows.
Types whose value has declined
FAQPage is the significant one. Google narrowed FAQ rich result display to well-known authoritative government and health sites, which removed the treatment from the overwhelming majority of sites that had implemented it. Many sites still carry FAQ markup that no longer produces the visual result it was built for.
Worth being precise here, because the conclusion people draw is often too broad: FAQ markup is still valid structured data, still parsed, and still contributes to machine understanding of the page — including for AI systems reading the page. What it no longer reliably does is produce expanded FAQ entries in Google's results. If it was implemented purely for the SERP real estate, its business case has changed. If it was implemented to make a genuinely structured Q&A page machine-readable, it still does that.
HowTo was similarly deprecated as a rich result type.
Implementation details that decide whether it works
Use JSON-LD. Google's stated preference, easiest to maintain, and separable from presentation markup. Microdata and RDFa are supported but harder to keep correct.
Mark up what is visible. The most common cause of a manual action for structured data is markup describing content that does not appear on the page. Rating a product 4.8 stars in the JSON-LD when no reviews are shown on the page is a policy violation, not a shortcut.
Fill required properties, then recommended ones. Google documents required and recommended fields per type. Missing required fields disqualify the page outright; missing recommended fields reduce eligibility.
Use @id to connect entities. A graph in which the Organization node is referenced by @id from the WebSite, the Article's publisher, and the LocalBusiness is far more machine-legible than three disconnected blobs repeating the same name string.
Validate both ways. The Rich Results Test tells you what Google can produce from the markup. The Schema.org validator tells you whether the markup is technically correct. Passing one and failing the other is common and worth catching.
What structured data does for AI search
This is where the calculus has shifted. Language-model-driven search systems parse pages to extract facts, and explicit structured data removes ambiguity that prose leaves in. A price, a service area, an author, a publication date, and an organization identity stated in JSON-LD are unambiguous in a way that the same facts embedded in sentences are not.
There is no confirmed ranking mechanism here and it would be dishonest to imply one. But the direction is defensible on first principles: systems that need to extract structured facts do better when structured facts are provided, and the cost of providing them is low.
The pragmatic priority
For most business sites the order is: Organization or LocalBusiness first (entity identity), Breadcrumb second (cheap, universal), Article or BlogPosting on editorial content, and then whichever vertical type actually matches your content — Product, Event, JobPosting, Course.
Skip the types that no longer render. Do not add markup describing content the page does not contain. And treat the whole exercise as machine-readability infrastructure rather than a ranking lever, because that is what it is.