Which schema types actually matter.

Schema advice usually arrives as a list of everything that exists. This is the shorter version: the types our grader validates, the properties it insists on, and the two that carry most of the value for AI answers.

What we read

Schema markup is a shared vocabulary for labelling what a page is about, so a machine does not have to infer it from your layout. JSON-LD is the format we read, and the only format we score.

A required property is a field we treat as load-bearing, so a type that declares itself and then omits one loses points. We follow a graph wrapper and an array of types, because a page that uses either is not doing anything wrong and should not be punished for it.

The ten types we validate

Anything outside this table is ignored rather than penalised. Adding types we do not check is not a mistake, it simply earns nothing here.

Type	Properties we require	Worth adding when
Organization	name, url	Always, on every page
Person	name	You are the brand
WebSite	name, url	Always, once per site
FAQPage	mainEntity	The page answers real questions
Article	headline, datePublished	Guides and explainers
BlogPosting	headline, datePublished	Dated posts in a feed
NewsArticle	headline, datePublished	Reporting rather than evergreen
LocalBusiness	name, address	Customers come to a premises
Product	name	A specific thing is for sale
BreadcrumbList	itemListElement	The page sits under a parent

The mainEntity property is the array of questions and answers inside FAQ markup. A breadcrumb list refers to the trail of parent pages above the current one, which is how a model places a page inside a site.

How the scoring works

The Schema Markup module starts every site at 100% and deducts from there. The deductions are small and specific, which makes the module unusually easy to reason about.

No structured data at all. The module drops to 60% immediately, and that is the largest single deduction.
A required property missing. Ten points off, charged once per missing property rather than once per block.
Markup that will not parse. Fifteen points off, and the block is then worth nothing to anyone reading it.

Read the second and third rules together and the warning is obvious. Three half-finished blocks can cost more than having written none of them.

Why the schema module understates the value

Schema Markup itself carries only 5% of the overall grade, which makes it one of the lightest modules we run. Judged on that number alone, markup looks like a rounding error.

That reading is wrong, because three further modules read the same JSON-LD for their own purposes. This is the real reason to get it right.

Module	Weight	What it reads out of your markup
Entity Clarity	12%	Organization or Person, and a postal address
Citability	12%	FAQ markup, alongside your question headings
Trust Signals	5%	An author, and a published date
Schema Markup	5%	Every type in the table above

We will not claim those weights simply add up, because each module scores other things too. The honest statement is narrower and still useful. Correct markup is an input to four of the ten modules, and the AI-search half of the grade is 45% of the total.

What to add, in order

Organization on every page, carrying a name, a web address, a logo and social profiles.
FAQ markup wherever you already answer questions in the visible copy.
Article markup on guides, carrying a headline, an author and a published date.
LocalBusiness with a full postal address, if customers visit you.
Breadcrumb markup on interior pages, which is usually generated rather than typed.

Mark up only what a visitor can actually see. Questions in your FAQ block must be on the page, and a date in your markup should match the date printed on the page.

Common questions.

Does schema markup make an AI assistant cite me?

Not on its own, and anyone promising that is selling something. Markup makes your claims unambiguous, which removes a reason not to cite you. The passage still has to be worth quoting, which is a writing problem rather than a markup problem.

Which single type should I add first?

Organization on every page, then FAQPage on the pages that answer questions. Organization is what lets a model say who you are, and it feeds the heaviest of the AI-search modules. FAQ markup is the only type that directly earns points in a second module as well.

Is Microdata or RDFa good enough instead of JSON-LD?

Our grader reads JSON-LD only, so other formats score nothing here. That reflects where the ecosystem has settled rather than a preference of ours. JSON-LD also lives in one block in the head, which makes it far easier to keep accurate than attributes scattered through the markup.

Can adding schema make my score worse?

Yes, and this surprises people. A type we validate that omits a property we require costs points that the missing block would not have cost, and invalid JSON costs more again. Incomplete markup is worse than none, so finish the block or leave it out.

The free scan parses your JSON-LD and names every type it found and every required property it did not. No account and no card. Paid teardowns start at $79 and include the corrected markup.

Run a free scan →