We Published the Observatory as a Schema.org Dataset. Here Is the Markup

Structured data

A Schema.org Dataset is the one markup type almost nobody in financial marketing uses, and it is the type that describes what a research study actually is. On 10 September 2026 we published the AI Asset-Manager Observatory as a Dataset, alongside Person and Organization nodes and an Article node, all in one JSON-LD graph on the live dashboard. This is the markup, the reasoning, and what to copy.

Why the usual markup does not describe a study

Article and BlogPosting describe a document. They tell an engine that a page exists, what it is called, who wrote it and when it was published. That is the minimum, and it worked for search engines built on links.

It says nothing about the study inside the document. It does not say there is a measurement, what was measured, how many records it covers, when it was collected, how it was produced, or who to credit. Those are exactly the fields an answer engine needs if it wants to quote a number rather than a sentence.

Dataset fills that gap. It carries measurement technique, variable measured, temporal and spatial coverage, licence, creator and publisher. A study that declares those things is easier to cite correctly, and harder to cite wrongly.

The graph we published

Four nodes, one graph, three @id references so the relationships are explicit rather than re-stated on every page.

Node	What it declares	Why it matters
Dataset	Name, description, measurement technique, variables, coverage, licence, creator	Tells an engine this is original measurement, not commentary
Person	Author, role, languages, LinkedIn	Connects the study to a named expert, which is an E-E-A-T signal
Organization	Publisher, founder, area served, topics	Establishes who stands behind the data
Article	Headline, section, date published and modified	Keeps the page itself properly described

The Dataset node carries a citation field with a ready made reference line: Roodhorst, V. (2026). AI Asset-Manager Observatory, Q3 2026. roodhorst.eu. That single field removes the most common failure in AI citation, which is a correct number attributed to the wrong source.

What the Dataset says about our own study

Two fields do most of the work in our case. measurementTechnique states that a fixed 20 prompt buyer intent set was run against multiple AI engines in five languages, with reach defined as the share of completions naming the firm at least once. variableMeasured names the three quantities we actually report: AI share of voice, first place rate, and the split between European and global mentions.

That is the difference between an engine guessing at our method from prose and an engine being told the method directly. The live dashboard carries both, next to the 2,243 logged answers they describe.

The four marks of a citable study

1. A stated sample

Not “we analysed European asset managers” but “1,741 logged completions across five languages, collected between July and September 2026”. A sample size is a claim an engine can carry, and a claim a reader can check.

2. A published method

Name the instrument, the roster and the counting rule. Our counting rule is deliberately blunt: a firm is named once per answer however often it appears, and first place goes to whoever appears earliest. Blunt rules survive contact with sceptical readers.

3. A traceable number

Every figure in our write ups resolves to a saved raw answer. When someone challenges a number we can show the file. That is what makes the study safe to cite, and safety is what gets you cited.

4. Machine readable provenance

Creator, publisher, licence, coverage and citation, in one graph. It is unglamorous structural work, and it is the part most competitors skip.

What Dataset markup will not do

It will not create a citation. Markup describes content, it does not manufacture authority. A thin page with a beautiful graph is still a thin page.

It will not fix a study nobody can read. If the numbers are on slide nineteen of a PDF behind a form, the markup has nothing to describe. Publish the data on the page first, then describe it.

It will not survive a wrong image or a wrong date. Our own styleguide exists because a guessed uploads path in schema output is worse than no schema at all. Every URL in the graph above was read back from the live site and verified before publish.

How to copy it in an afternoon

List the studies you have published. For each one, write down the sample, the method, the coverage window and the variables you report. Then add one Dataset node per study, one Person node for the author, one Organization node for the publisher, and reference the @id values from your existing Article markup. Keep the graph in one script tag and keep the citation line short enough to be copied by hand.

The AI visibility pillar sets out what the measurement is for, and the crawler audit covers the precondition: none of this matters if the crawler is blocked at the door.

What is a Schema.org Dataset?

Dataset is the Schema.org type that describes a body of data rather than a document. It carries fields such as measurement technique, variables measured, temporal and spatial coverage, licence, creator and publisher, which Article markup does not.

Does Dataset schema improve AI citations?

It makes the study machine readable and gives engines a correct citation line, which reduces misattribution. It does not create authority on its own. We have published it since September 2026 and treat any citation effect as something to measure, not something we can yet claim.

Should small marketing teams use Dataset markup?

Only if they publish original research. If you run a survey, a benchmark or a recurring study, the markup is worth the afternoon. If you publish opinion posts, Article and BlogPosting already cover you.

Where does the JSON-LD go?

In the page’s own content, in a single script tag of type application/ld+json. On WordPress that can live inside one Custom HTML block, which is how we ship it on the dashboard page.