PDFs are visually focused, not AI-friendly, but transforming tagged PDFs into structured formats such as HTML or XML makes their content accessible to machine learning systems.
This free application note summarizes precisely where PDF creators should add XMP metadata to PDF 2.0 objects, and thus where PDF 2.0 processors should search for it.
To help implementers locate the latest information, the PDF Association maintains a page of links to current versions of documents normatively referenced from ISO 32000-2:2020.