A web page has two readers: the human who reads the text, and the machine that has to understand what that text is. For the human, the difference between an article, a product page and a frequently asked question is clear at a glance. For the machine, it's all just text, unless you tell it otherwise. Telling it that is what schema markup does.

01 / the principleWhat structured data is

Schema markup, also known as structured data, is a piece of invisible code in your page that tells machines what's there: this is an article, this is the author, this is the publication date, this block is a question with an answer. The conventions for this are set out on schema.org and are read by all major platforms.

The common types can be briefly summarised. Article for articles and blog posts, with author and date. FAQPage for blocks with questions and answers. HowTo for step-by-step guides. Product for products with price and reviews. Organization for who you are as a business. For the visitor, nothing about the page changes; the code is only for machines.

02 / the effectWhat AI systems do with it

Not demonstrable for AI citations: Ahrefs measured in May 2026 that pages adding JSON-LD saw no measurable gain in AI citations. The often-cited figure of 30 to 40 percent comes from Princeton's GEO research and concerns content (sources, citations and statistics), not schema. The explanation isn't mysterious. A system that has to work out from a page who the author is, when the piece was written, and which block contains the answer to which question, doesn't have to guess on marked-up pages. What doesn't need to be guessed gets picked up correctly more often.

Schema only reinforces what's already there. A page without content doesn't become a source through markup; a good page becomes more readable through it. So the order is fixed: content that genuinely answers a question first, then the markup on top. How you do the first part is covered in how to write content AI can cite.

Schema markup leaves a machine nothing to guess. What doesn't need to be guessed gets picked up correctly more often.

03 / the nuanceWhat Google does and doesn't ask for

Google is clear about it: no special markup is required for AI Overviews. Google's AI features rest on the regular search and quality systems, and content written for people wins there. Anyone adding schema purely to make it into AI Overviews is doing it for the wrong reason.

At the same time, Google does recommend structured data for your visibility in the broader sense, including for rich results with stars, questions and breadcrumbs. And platforms outside Google do read the markup. So the conclusion is fairly sober: schema isn't required anywhere, demonstrably helps outside Google, and does no harm anywhere. It belongs in the same category as llms.txt: machine readability as a layer on top of good content, not a replacement for it.

04 / in practiceWhere to start

Start small and where it pays off. For most service providers, that's this order:

The direct recommendation for a service provider with a knowledge blog: do steps 1 and 2 now, add FAQPage as soon as you have pages with question-and-answer blocks, and skip step 4 until your site genuinely calls for it.

Many systems handle part of this themselves: WordPress with an SEO plugin, and modern site builders too, write Article markup automatically. So check what's already there before having anything built. You can check this for free with Google's Rich Results Test or the validator on schema.org: enter the link, and you'll see which types the page carries and whether there are any errors.

One warning: only mark up what's genuinely on the page. FAQPage markup on a page without visible questions, or review stars without reviews, fails inspection and can harm your visibility rather than help it.

05 / the limitWhen schema isn't a priority

If your site has five pages and barely any knowledge articles, the gain from markup is small: there's little to mark up, and machines don't get lost across five pages. Work first on the question of whether your pages answer your customers' questions.

Anyone without any measurement yet is also better off leaving schema for now. Without a baseline measurement, you won't know afterwards whether the markup did anything. Measure first where you stand, then make the change, and measure again; that way the effect is an observation rather than a feeling.

This text was produced with AI assistance and checked and approved by a human before publication.