Schema markup for AI search

Let’s set expectations before anyone sells you a markup audit: schema will not, by itself, get you cited by an AI engine. We’ve seen unmarked pages win citations on the strength of their content, and immaculately marked-up pages lose on the weakness of theirs. And yet we add or fix schema on nearly every engagement. Both things are true, and the resolution is simple: schema doesn’t win citations, it stops you from silently losing them.

What schema does in an AI pipeline

Engines assemble an entity picture of your organization from everything they can read. Structured data is the one place you state the facts in a format built for machines: what you are, what you sell, what things cost, who wrote this, when it changed. When markup and visible content agree, confidence in both goes up. When your About page, your Organization schema, and your LinkedIn say three different things, you’ve manufactured doubt, and engines resolve doubt by citing someone less confusing. Entity clarity is the first filter in the GEO playbook; schema is its load-bearing wall.

The types that earn their keep

• Organization, everywhere, complete. Name, legal name, URL, logo, sameAs links to every official profile, founding date, address. This is your entity’s passport. Most B2B sites we audit have a five-field stub.
• Product / Service with real offers. Engines love answering “what does X cost”; machine-readable pricing makes you the safest source. If your pricing is “contact us,” that’s a strategy decision with a GEO cost; at least mark up what you can.
• FAQPage, used honestly. Mark up real questions answered in 40–80 word self-contained blocks, the same extraction-friendly shape that wins AI Overview chips. Don’t wrap your nav links in it; everyone sees that.
• Article with honest dates. datePublished and dateModified feed the freshness signals engines visibly use. Which leads to the one rule people break: never bump dateModified without changing the content. Engines compare snapshots. Getting caught lying about freshness is worse than being stale.
• BreadcrumbList, cheap, and helps engines map which page owns which sub-topic in a cluster structure like this one.

What to skip

Exotic types with no consumer (Thing hierarchies five levels deep), markup describing content that isn’t on the page, and any “AI-specific” tags promising special model treatment, we’ve tested, they don’t. JSON-LD in one block per page, validated, matching the visible content. Boring wins.

The audit we actually run

1. Validate what exists; broken markup is worse than none.
2. Complete Organization and reconcile it with every public profile, same name, same description, same facts.
3. Add Product/Service + offers to money pages, FAQPage to genuine Q&A, Article with true dates to guides.
4. Re-check citations weekly and attribute honestly, schema changes land alongside content changes, so claim the win for the combination, not the markup alone. (Method: tracking AI citations.)