Published March 19, 2026
There are three formats for implementing structured data on a website: JSON-LD, Microdata, and RDFa. The SEO industry has recommended JSON-LD for years. Google has stated its preference for JSON-LD. But the question for AI visibility is whether the format matters when AI systems — not just search engines — are reading the data.
The AIFDS blueprint library uses JSON-LD exclusively. This article explains why.
JSON-LD is a standalone script block in the <head> or <body>, separate from page content. The structured data lives in one clean block readable without parsing the rest of the page.
Microdata embeds structured data directly into HTML elements — a <span> gets an itemprop, a <div> gets an itemscope. The data is woven into the page content itself.
RDFa works similarly — structured data in HTML attributes using a different syntax. Like Microdata, it requires parsing the HTML to extract the data.
All three can express the same schema.org types and fields. The difference is how the data is delivered.
The same principle behind the @graph approach and page weight as a structural advantage: reduce the work the crawler has to do.
An AI crawler reading JSON-LD can extract structured data without parsing a single line of HTML. The data is clean, self-contained, and already formatted as a structured object.
Microdata and RDFa require parsing the full HTML, extracting schema attributes from individual elements, and assembling data from pieces scattered across the document. The end result may be the same — but the process is more complex and more dependent on well-structured HTML.
For a simple page, the difference is marginal. For a complex page with thousands of lines of HTML, JSON-LD gives the crawler a shortcut: structured data, already assembled, in one place.
Google has publicly recommended JSON-LD for years. Every major CMS — WordPress, Shopify, Squarespace — generates or supports it. AI systems have been trained on and optimized for pages using JSON-LD.
The AIFDS blueprints did not test Microdata or RDFa against JSON-LD. The decision was practical: JSON-LD is proven, cleanest for AI processing, and requires the least integration with existing HTML. There was no compelling reason to test alternatives.
Separation from HTML. JSON-LD lives in a script tag. Changing structured data does not require changing page content. Microdata and RDFa are embedded in HTML — updating them means editing elements throughout the page.
Easier implementation. A developer pastes a JSON-LD block into the <head>. No HTML restructuring required. Microdata and RDFa require integrating schema attributes into existing page structure.
Easier validation. JSON-LD can be validated by pasting the script block into any validator. Microdata and RDFa require parsing the full page HTML — more prone to errors with complex or dynamic pages.
Cleaner for AI. A JSON-LD block in the <head> is the first structured data the crawler encounters. No assembly required. This aligns with making the crawler's job easier — the less parsing required, the more reliably AI processes the data.
If a site already has Microdata or RDFa that is working — generating rich results, passing validators — there is no urgent reason to replace it. But for any new implementation, JSON-LD is the format to use.
Every AIFDS blueprint provides exact JSON-LD for a specific industry and page type. The AI prompts generate complete schema from business details. The validator checks JSON-LD against the fields AI systems use. The entire framework is built on JSON-LD because it serves every stakeholder with the least friction.
David Valencia writes about how AI systems find, parse, and cite websites.