AI Optimization
Structured data: when it helps and when it hurts
Machine-readable statements about a page are useful when accurate and actively damaging when they contradict what the page says.
What it is for
Structured data lets a page state explicitly what it is about: that this is a business, with this name, at this address, offering these things. It removes the need for a system to infer, and inference is where misunderstanding enters.
In AI optimization terms it is a comprehension aid rather than a ranking mechanism, and treating it as the latter is the source of most of the disappointment around it.
When it genuinely helps
When it removes real ambiguity. A business name that is also a common word. A service area that is not obvious from the address. A relationship between entities that a reader would infer and a system cannot.
When it makes an implicit fact explicit. Opening hours in an image. A price stated in prose. An identifier that only appears in a footer.
When it connects an entity to records elsewhere. Explicit references to authoritative identifiers do more than almost anything else on a page, because they let a system resolve rather than guess.
The rule that governs all of it: structured data should describe what the page already says. The moment it asserts something the visible page does not support, it has stopped being a comprehension aid and become a claim.
When it hurts
Contradiction with the visible page. Markup saying one category while the page describes another does not add a signal, it undermines both.
Ratings and reviews the business generated for itself. Widely abused, widely discounted, and a reliable way to be treated as less trustworthy rather than more.
Stale markup. Hours, prices and offers that were accurate two years ago. Structured data does not age visibly, so nobody notices, and a confidently stated wrong fact is worse than an absent one.
Everything, everywhere. Marking up every conceivable property produces noise and increases the surface area for contradiction. Fewer, accurate, load-bearing statements beat comprehensive ones.
Why it is not a shortcut
Structured data is a description of an entity, not evidence about it. It can make a business legible. It cannot make a business credible, because anyone can write it, and signals that anyone can write carry the weight that implies. A business with excellent markup and no independent corroboration is clearly described and unsupported.
How AI optimization work should sequence it
After the page says the right thing in prose, and after the descriptions across sources agree. Markup applied on top of an unclear page encodes the lack of clarity precisely, which is not an improvement.