Why most schema does nothing
Most structured data on the web restates what the page already says in a form nobody disputed. Marking up a heading as a heading changes no model’s belief about your business. Schema earns its keep only where it resolves an ambiguity that the prose leaves open.
That reframing matters because it turns a checklist exercise into an editorial one. The question stops being “which types can we add?” and becomes “what could a model get wrong about us, and which markup settles it?”
Identity is the ambiguity worth resolving
The highest-value markup establishes who you are and that all your surfaces refer to the same entity. A single canonical Organization node, referenced by @id from every other schema block on the site, is worth more than a dozen decorative types. It tells a model that the service pages, the articles, and the contact page are one organization rather than three loosely related documents.
Consistency between markup and visible copy is not optional here. Where they disagree, you have not added information — you have added a contradiction, and contradictions reduce the confidence that produces a recommendation.
Which types actually pull weight
In practice four do most of the work: Organization for identity, Service for what you sell, Article for anything you publish, and FAQPage where you genuinely answer discrete questions. Each maps to a category of prompt — who are you, what do you do, what do you know, what about this specific thing.
Schema is not a ranking trick. It is a disambiguation tool, and it only helps where something was ambiguous.
The knowsAbout field is underrated
Organization schema lets you state the subjects you are authoritative about. This is one of the few places you can directly assert topical scope rather than hoping it is inferred from your page mix. Keep it honest and narrow: five subjects you can defend beat twenty you have mentioned once.
What to leave out
Don’t mark up what you can’t substantiate, and don’t mark up data you don’t want repeated. Anything you put in structured data is an explicit invitation to quote it — including addresses and phone numbers, which is worth a deliberate decision rather than a default.
Aggregate ratings you can’t evidence, employee counts that flatter, service areas you don’t really cover: all of these are liabilities the moment a model surfaces them to a buyer who then checks.
How to tell whether it worked
Ask the engines directly. Prompt several assistants to describe your company, list your services, and name what you specialise in, then compare the answers to what your markup asserts. Gaps between the two are your backlog, and the exercise takes an afternoon.