Schema Markup for AI Answers: What Actually Helps and What's Cargo Cult
Structured data shows up on every GEO checklist, usually as an undifferentiated "add schema markup" bullet. It is one of the few recommendations on those lists with a defensible mechanism, but the mechanism is narrower than the advice implies, and it explains which markup is worth your time and which is decoration.
The mechanism
Schema does not make an engine prefer you. What it does is remove ambiguity about what an entity is and how it relates to other entities. A page that says "Acme" in a heading requires the engine to infer, from context, whether that is a company, a product, a person or a place, and which of the eleven companies called Acme it is. A page with an Organization block carrying a sameAs array pointing at the company's Wikidata entry, LinkedIn page and Crunchbase profile has answered that question outright.
That matters for AI answers specifically because entity disambiguation is where a lot of brand-level failure happens. Being described as a different company with a similar name, or having a competitor's pricing attributed to you, is an entity resolution failure, not a ranking failure, and it is the failure mode schema is genuinely good at preventing.
What is worth marking up
Organization, done properly. Legal name, logo, founding date, location, and a thorough sameAs array. This is the single highest-value block for most B2B sites and the one most often implemented as a two-line stub.
Product or SoftwareApplication with real offers. If your pricing is public, mark it up with actual numbers and currency. Pricing is one of the most frequently hallucinated attributes about SaaS products, and a machine-readable price on a canonical page is the cheapest available defense.
FAQPage where the questions are real. The value is not the rich result, since Google curtailed those for most sites. It is that the markup pairs an explicit question with an explicit, self-contained answer. That is precisely the shape a generative engine wants to lift.
Article with author and datePublished. Attribution and recency are inputs to whether a source is treated as credible and current.
What is cargo cult
Marking up things that are not on the page. This is the most common mistake and it is actively harmful. Schema that describes content a human visitor cannot see is a policy violation in Google's guidelines and a credibility problem everywhere else.
Review and AggregateRating on your own site, about yourself. Self-serving review markup has been devalued for years. It will not persuade an engine and it can flag you.
Exhaustive nesting for its own sake. A twelve-level-deep graph that no consumer reads is engineering effort spent on the wrong problem. Depth is not the signal; correctness and coverage of the entities that matter are.
Assuming it substitutes for content. Schema describes what is there. If the page has no specific claims on it, perfectly valid markup describes an empty room.
The check worth running
Fetch your own key pages with JavaScript disabled and see what survives. A substantial number of sites render their product details, pricing and even their main copy client-side, which means the markup validates in a browser-based testing tool and is invisible to anything that does not execute scripts. Validate the served HTML, not the rendered DOM. That one check catches more real GEO problems than most audits.