JSON-LD that matches the page — otherwise it is a script you hope with

Structured data is a promise to two parties: the crawler, which would like to understand the page without guessing, and the quality rater (human or system) that checks whether you are cheating. The promise holds when FAQ markup matches the questions you actually show, when Article has the author and date the page shows, when Organization has the name and company id the footer has. It rots when you hope for stars, FAQ extraction or a knowledge panel by describing something the user cannot see.

Teams do it because a plugin or an agency slide said 'turn on schema and you get rich results'. The first test tool goes green. It becomes the habit. Three months later the homepage has LocalBusiness, Product, FAQPage and Review in one script, and none of it stands as content. Google does not show the stars. Now and then Search Console shows 'either/or' or 'non-critical issues'. You fix one warning. You do not touch the truth.

The damage is not only missing rich results. Markup that overclaims is a signal you have asked to be treated as that. A Review you wrote yourself. An AggregateRating without sources. An FAQ that only exists in JSON. A price you put on Product that you do not show — and that we do not put on a marketing site anyway. You have told a machine something you will not stand behind on the page. That is the opposite of SEO. It is asking for your honest pages to be met with scepticism too.

We write this because 'we have JSON-LD' has become a tick you add at go-live. An honest setup is boring: one @type that fits the page, fields you can point at, and no nodes you turned on 'to be safe'. Without that habit you have a script from the plugin. You do not have structured data.

JSON-LD is a script of type `application/ld+json` that describes the page with schema.org types. It is not a secret ranking factor you can switch on. It is a label on what you have already written. WebSite and Organization on a site that has a name and a homepage. Article on a post with an author and a date. Service on a service you describe. FAQPage on a page that shows questions and answers. BreadcrumbList on the trail the user can see. Each type is a job. Not a shelf you empty into `<head>`.

It is not the same as Open Graph and Twitter tags. Those are sharing: title, description, image, when someone posts the URL. JSON-LD is meaning: what kind of page this is, who publishes it, which questions sit on it. You can have honest OG without JSON-LD. You can have JSON-LD that contradicts OG and the H1. The crawler and the messenger unfurl then choose. You have not chosen. Write the H1 and the meta first. Write JSON-LD as a mirror, not as a second draft.

It is also not the same as a sitemap or a canonical. The sitemap is the list of URLs you want crawled. Canonical is which variant is preferred. JSON-LD says what the preferred page means. Pasting Organization on every URL and forgetting Article on the post is saying every page is the company. Pasting FAQPage on every URL is saying every page is a Q&A. Both are lies if the page is a marketing section without that block.

And it is not 'Yoast / RankMath / a React component that dumps the whole graph'. A plugin can be a tool if you turn off what you have no content for. A component can be a tool if it reads the same dictionary the page renders. What typically goes wrong is a dump: every type, empty fields, an address you do not want to be LocalBusiness on, and a review you never received. The inventory is the script you deploy and the HTML it mirrors. Everything else is a hope you parked with a vendor.

The first failure is turning on types the page is not. LocalBusiness on a site that sells nationally and internationally — you have a Danish address, you are not a shop with opening hours as the product. Product on a service without a visible product, without an offer, without what you may show as a price. Review and AggregateRating you wrote yourself, 'so we can get stars'. Each wrong type is a place a test tool can go green and a rater can say no. Pick the type you can defend in a meeting where someone opens the page next to the script.

The second failure is fields that do not sit on the page. A `description` that is a different draft from the lead. An `author` who is not named. A `datePublished` you set to today on an old post every time you deploy. A `telephone` on Organization that you do not show. Google has repeated: markup must match visible content. That is not a guideline you can 'take later'. It is the whole job. If the field is true, show it. If you will not show it, do not mark it up.

The third failure is pasting the same graph on every page. Homepage, service, post, 404 — the same Organization + FAQ + WebSite in one lumped script. Organization and WebSite can belong on the homepage and in a site-wide block you keep honest. FAQ belongs where the FAQ is. Article belongs on the article. A 404 with Article is a lie. The error-pages post is chrome. This is letting the script follow the template, not letting the layout be a dump.

The fourth failure is chasing rich results as the goal. Stars, FAQ extraction, sitelinks searchbox. Some of those are restricted, retired, or given to pages that already have trust. You control the markup. You do not control the extract. An honest Article with author, date and an image the page shows beats a Review you invented. If you want FAQ extraction, write the FAQ as content first. The script comes after. The speed post is LCP. This is refusing to let a hope of SERP chrome drive what you put in `<head>`.

The homepage is the job that most often gets the whole catalogue. It is the company and the site: Organization with legalName, an address you will stand behind, sameAs to the profiles you actually own, WebSite with the name you use. It is not LocalBusiness unless the local place is the product. It is not FAQPage unless the homepage shows the questions. It is not Product. An honest site-wide block beats a dump you copied from a theme.

The article is the job that most often forgets author and date, or puts them in the script and hides them in the UI. BlogPosting / Article with headline = H1, datePublished = the date you show, author = the byline you show, image = the hero you show. Update dateModified when you actually edit, not when you deploy. A post without a visible date should not have a date in JSON-LD you hope for in the SERP.

The service is the job that most often becomes Product with Offer, because someone wanted price markup. A service is Service: name, description, provider pointing at Organization. Area served, if you mean it. Not a price field you do not show — and that you should not show on this site. An Offer without a visible price is a lie. An Offer with a price you will not defend on a call is worse. Write what you do. Mark that up. Leave the price in the conversation.

The FAQ is the job that most often becomes a hidden array. FAQPage is right when the page shows questions and answers as content — not as an accordion you have `display:none` on, and not as H2s you never asked as questions. Four to six real questions beat twenty thin ones. The JSON-LD is the same array the block renders. If you remove the block, you remove the node. Search Console extraction is a bonus you do not control. The visible FAQ is the job.