Schema markup

A search engine reading an ordinary page is doing inference. It sees a phone number and infers a business; it sees a date and infers a publication; it sees a suburb and infers a service area. It is usually right, and when it is wrong there is nothing you can do about it, because you never told it anything. Markup is the part where you tell it.

The vocabulary itself is schema.org, published in 2011 by the major search engines so there would be one agreed way to say "this is an organisation" rather than four. The vocabulary is enormous and deliberately so, covering everything from a bus route to a medical trial. What matters commercially is much narrower: the handful of types Google will actually do something with.

There are three ways to write it. JSON-LD is a self-contained block that sits in the page source and touches nothing else; microdata and RDFa weave attributes through your visible HTML. Google recommends JSON-LD, and so do we, for the boring reason that it can be changed without anyone touching the layout.

Here is the part that makes most schema advice out of date. In August 2023 Google announced that frequently-asked-question rich results would be shown only for well-known, authoritative government and health websites. For every other site they would no longer be shown regularly. At the same time, step-by-step how-to rich results were limited to desktop. In September 2023 they were removed from desktop as well, and Google deleted the how-to documentation entirely.

Neither has returned. Yet both are still recommended on page after page of SEO advice, sold as free real estate on the results page, and still shipped by plugins by default. A small business paying for question markup on twelve pages in 2026 is paying for a feature that was withdrawn three years ago.

That does not make the markup harmful, and it is not why we avoid it. We simply do not ship a type that cannot pay for the code it adds, and we hold a harder line on a second group: ratings, reviews and prices that a business writes about itself. Markup that asserts a five-star average nobody can verify is the fastest way to lose the trust that the rest of the markup is there to build. On our own site a page will not ship if it carries a rating, a review or a price we cannot stand behind, and that rule has caught more mistakes than it has cost us features.

For a normal Australian business, four types carry almost all of the value, and a fifth if you sell things. None of them is exciting. All of them answer a question a machine would otherwise guess at.

If you sell products, product markup remains fully supported and genuinely worth the effort, because stock status and shipping detail change what a shopper sees before they click. Everything beyond these is a judgement call, and the honest test is the one Google publishes: does this type have a documented search feature attached to it, and does your page truly contain what the type claims?

The second half of that test is where most sites fail. Markup has to describe what is visibly on the page. Question markup with no visible questions, an author who appears nowhere, a rating pulled from a different site: each of those is a policy breach rather than a clever shortcut, and Google states plainly that markup which does not match the visible content can cost you the eligibility you were chasing.

The rich-result argument for markup has been shrinking for years. The reason to keep writing it has changed, and it is now the stronger of the two.

An answer engine composing a reply has to decide what a page is, who stands behind it and whether the two agree with what it already believes. Clean markup makes that resolution cheap and unambiguous: this organisation, this address, this author, this date, stated the same way across every page of the site. Messy or absent markup does not make the page invisible, it makes it ambiguous, and an ambiguous source is a poor candidate for a sentence a model has to be confident about.

That is why we treat markup as identity work rather than decoration. It is also the least glamorous half of generative engine optimisation, and the half that quietly decides whether the rest of the effort resolves to your brand or to nobody in particular. If you want the reasoning behind the rest of it, start with what generative engine optimisation is and how to get cited by AI in Australia.

Entity consistency is the practical version of all this: the same name, the same address, the same phone number, written identically wherever they appear. It sounds trivial. It is the single most common thing we correct on an Australian site, and it is covered properly in entity SEO.