Schema.org: which structured data should you ask for on your site?
Check that your business, your services and your articles are described correctly, with no outdated prices and no promise of AI citations.

- Published
- Updated
- Reading time
- 7 min
- Author
- SBSébastien BalieuFounder of Numinam
Structured data describes certain elements of your site in a standard format: a business, a product, an article. Your developer can generate it from the page content, so the visible text and the code never drift apart.
Google uses this markup to understand pages and to build some of the ways they appear in its results. Valid markup guarantees neither a rich result nor a citation in an AI assistant.
Do AI assistants read this markup?
Google answers for its own features. Its page on AI features in Search states that you need no special Schema.org structured data to appear in AI Overviews or AI Mode. It does recommend that your structured data match the visible text of the page.
OpenAI, Anthropic and Perplexity do not publish how their assistants handle JSON-LD. So I can’t tell you how much weight it carries in an answer from ChatGPT or Claude, and nobody can claim to know with evidence to back it up.
The markup still has a use you can verify. It gives your name, logo, official profiles and article dates in an unambiguous form, and Google uses it, for example, to choose the logo shown in its results. Wrong markup spreads an error in a format that machines take at face value.
Which types match your pages?
| Page | Type to consider | What to check |
|---|---|---|
| Business identity | Organization, or LocalBusiness where appropriate | Exact name, URL and contact details |
| Service offer | Service | Provider, type of service and area served |
| Article | Article or BlogPosting | Title, author, publication date and actual last update |
| Product page | Product | Product, offer, visible price and availability |
The Schema.org vocabulary covers far more types than the rich results Google supports. Adding a Service type is no promise of any particular display.
For your identity, Google advises the most specific subtype. An online shop takes OnlineStore, a restaurant or a physical shop takes a subtype of LocalBusiness. The LocalBusiness type requires a physical address: a consultant who sees clients at an office can use it, while a business with no address open to visitors stays on Organization.
Organization markup has no required properties at Google. It goes on the home page or on a page that presents the business, once, without repeating it everywhere. Google recommends including, among other things, the name, URL, logo, address, contact details, sameAs and VAT number.
HowTo still exists in the Schema.org vocabulary but has gone from Google’s list of rich results. Only ask for it if another specific tool makes use of it.
What does a correct JSON-LD example look like?
This minimal example describes Numinam as an organisation. It is there to show the structure, so it leaves out any address or logo that hasn’t been verified.
{
"@context": "https://schema.org",
"@type": "Organization",
"@id": "https://www.numinam.com/#organization",
"name": "Numinam",
"url": "https://www.numinam.com"
}
Your developer has to adapt it to your business details and to the requirements of the type you choose. The identifier lets you link a service or an article back to the same organisation. The available properties are documented on the Organization and Service pages.
How do you link the business, its services and its articles?
Each block can point to another through its @id. A service names its provider, an article names its author and its publisher. A tool reading the markup can then attach every block to the same business and the same person, across the whole site.
A service offer, linked to the organisation above:
{
"@context": "https://schema.org",
"@type": "Service",
"name": "Conversion rate optimisation",
"serviceType": "CRO",
"provider": { "@id": "https://www.numinam.com/#organization" },
"areaServed": { "@type": "Country", "name": "Belgium" },
"url": "https://www.numinam.com/en/website-conversion-rate-optimisation"
}
On my site, every article gets a BlogPosting block built from the article’s own fields: title, summary, image, publication date and last-modified date. The author and the publisher point to identifiers declared only once:
{
"@context": "https://schema.org",
"@type": "BlogPosting",
"headline": "Schema.org: which structured data should you ask for on your site?",
"datePublished": "2026-03-10",
"dateModified": "2026-10-28",
"author": { "@id": "https://www.numinam.com/#sebastien-balieu" },
"publisher": { "@id": "https://www.numinam.com/#organization" }
}
When I edit an article, the date changes in the text and in the markup at the same moment, because both come from the same field. Nobody has to copy a date across by hand.
The sameAs property lists the organisation’s official profiles: LinkedIn page, Google Business Profile listing, professional directory. Only include profiles you control or that describe you accurately. The GEO and SEO guide explains how to check these profiles before you list them.
Where does the code go?
Google accepts JSON-LD, microdata and RDFa, and recommends JSON-LD in most cases because it is easier to maintain. The block sits in a <script type="application/ld+json"> tag, in the head or the body of the page.
Google also reads JSON-LD injected by JavaScript. That flexibility has a downside: a plugin, a tag manager and the theme can each add a block, and you end up with three different versions on the same page. Ask for the list of everything that generates markup on your site, then keep one source per type.
What should you check on prices and dates?
A price must state its unit and match the published offer. A day rate is not the total price of a project, and a price range must never be turned into a guaranteed amount.
Ask for a single source that feeds both the prices shown on the page and the structured data.
For articles, keep the original publication date and only change the last-modified date after real work on the text. The author’s name must match a person the reader can identify on the site.
Should every question be marked up as FAQPage?
An FAQ earns its place when it answers useful questions the page doesn’t already cover. Its markup must repeat the text the reader sees.
Google stopped showing FAQ rich results from 7 May 2026. So don’t pay for a long, repetitive FAQ to get that display. The answers your clients need are still worth writing.
Which mistakes does Google penalise?
Google’s structured data policies forbid several practices, and a breach can lead to a manual action that removes eligibility for rich results.
- Markup describing content that is missing from the page or hidden from the reader.
- A type unrelated to the page, such as an “About” page marked up as Product.
- Fake reviews, or ratings unrelated to the subject of the page.
- Missing required properties for the rich result you are aiming for.
- Images in the markup that Google can neither crawl nor index.
Google also asks you to put the markup on the page it describes. A general Product block on the home page does not replace the one on each product page.
The most expensive mistakes pass the validators, though: an old price, an office address you moved out of, a service you dropped. The code validates and the information is wrong.
How do you check your developer’s work?
Ask for a check with the Schema.org validator and, for the types concerned, with Google’s Rich Results Test. The two tools check different things.
Then check the rendered page: prices, URLs, author, logo and dates. A syntax test can pass while the information is wrong. Inspect several page templates, and run the check again after any price change.
Keep an update procedure too. Markup typed in by hand and then forgotten can end up contradicting the site. For content discovery, add the GEO and SEO guide and crawler access to these checks.
If you hand this review to a contractor, the deliverable to expect is a list of corrections you can verify.
Frequently asked questions
Do AI agents need special markup?
No. Google says so in writing for its AI Overviews and AI Mode, and the other vendors document no type reserved for their assistants. The usual types, accurate and consistent with the page, are enough.
JSON-LD or microdata?
JSON-LD in most cases. Google accepts all three formats and recommends the one that is easiest to maintain. JSON-LD keeps the markup separate from the visible HTML.
Does every page need markup?
No. The organisation is declared on the home page or the About page. Each article, product page or service page gets the block that describes it, linked to the organisation by its identifier.
Does my CMS already generate this data?
Possibly, in part. An SEO plugin such as Yoast SEO already outputs Organization and Article blocks, and some themes add their own. Have these existing blocks listed before adding more, so you don’t end up with contradictory duplicates.
Can markup do harm?
Yes, if it is wrong or misleading. Google can remove eligibility for rich results through a manual action, and a wrong piece of data stays readable by every tool that picks it up.


