Schema is an entity tool now, not a snippet tool

Structured data stopped being a way to get a bigger search listing. Most of the markup types that once drew something on the results page have been retired, and Google's own May 2026 guidance says structured data is not required for generative AI features either.

That leaves schema doing one job well: telling every machine that reads the web exactly who you are, what you sell, and which profiles across the internet belong to you. That is entity work, and it is worth doing. It is just not what most schema scopes are still selling.

What still renders

The list is short and worth knowing precisely, because it is where display value still exists.

Product markup with price, availability and review data still produces enhanced listings and feeds merchant surfaces. Review snippets still show stars in eligible categories. Breadcrumbs still replace the raw URL path. Video markup still drives thumbnails and key moments. Events, recipes and job postings still render in their specialised treatments. Article markup contributes to some news surfaces.

That is close to the whole list. FAQ rich results were deprecated globally in May 2026, including the government and health exemption that had survived since 2023. HowTo went earlier. Several others narrowed to the point of irrelevance for commercial sites.

If a proposal promises rich results from a markup type not on that list, ask which surface it renders on.

What schema is genuinely for now

Search systems, AI assistants, shopping engines, mapping products and voice systems all need to resolve one question before anything else: is this the same company across all these mentions.

That problem is harder than it sounds for real businesses. Names collide with unrelated brands, with product models, with towns, with other companies in other countries. A company can have a trading name, a legal entity name, three regional office listings and six social profiles that a machine has no automatic reason to connect.

Four things do that work.

Organization on the homepage, with legal name, ABN or company number where relevant, logo, founding date, address and contact points. This is the anchor record.

sameAs pointing at every profile you control: LinkedIn company page, YouTube channel, Instagram, Facebook, industry association listings, Crunchbase, Wikipedia if you have an entry. This is the single highest-value and most-ignored property in the whole vocabulary. It is the explicit statement that all these accounts are one entity.

LocalBusiness per physical location, with consistent name, address and phone matching your Google Business Profile exactly. Inconsistency here is the most common cause of a business being treated as two entities.

Product for what you actually sell, with real identifiers. GTIN, MPN, SKU, brand. These connect your listing to product records that exist outside your site.

The mistake almost every large site makes

Marketing owns the brand and IT owns the templates, so sameAs gets populated once at launch and never touched again. Three years later the company has a new LinkedIn page after a rebrand, a YouTube channel nobody linked, an Instagram handle that changed, and a Facebook page from a regional office that is not referenced anywhere.

The consequence is not abstract. Systems assembling a picture of your company from the web will pick up whichever profiles they can find and associate them with whatever entity they think owns them. If your own markup does not claim your channels, something else may. Companies have found search systems attaching the wrong employer, the wrong founder or a former trading name to their brand, and the fix in nearly every case starts with correcting the entity signals they control.

Audit sameAs once a quarter. It takes twenty minutes and it is the cheapest correction in technical SEO.

What not to spend money on

Do not add schema to win AI citations. Google says it is not required for generative features. If a scope justifies a large schema project on AI visibility grounds, that justification does not survive contact with the primary source.

Do not mark up content that is not visible on the page. That has always been against the guidelines and enforcement is real. The rich result loses eligibility and, in bad cases, a manual action follows.

Do not build custom vocabularies nobody consumes. Stick to schema.org types that documented consumers actually read. Inventing properties produces valid JSON-LD that nothing uses.

Do not treat validation as success. A page passing the Rich Results Test means it is syntactically correct, not that it earns anything. Most schema reports celebrate the wrong number.

A sensible implementation order

1. 1.Organization plus sameAs on the homepage. One block. Correct, complete, current. Half a day of work, largest single benefit.
2. 2.LocalBusiness per location, with name, address and phone matched character for character against your Google Business Profile.
3. 3.Product markup with real identifiers if you sell products. Only where price and availability are accurate, because wrong data here creates worse problems than no data.
4. 4.Breadcrumbs sitewide. Cheap, still renders, improves how your URL structure reads on the results page.
5. 5.Video markup on anything hosted or embedded, because video is one of the few areas where structured data still buys visible placement.

Then stop. Everything past that point is diminishing returns dressed up as thoroughness.

How to report it honestly

Do not report schema coverage as a percentage of pages. That number rises without anything improving.

Report three things instead. Whether the entity record is complete and current. Which rich result types you are eligible for and how many impressions each earns. Whether your knowledge panel and business profile show the correct information.

Schema became infrastructure. Infrastructure gets checked, not celebrated.