Accessible PDFs from LaTeX:
PDF/UA and the EU Accessibility Act

What changed in June 2025

The European Accessibility Act (Directive 2019/882) has applied since 28 June 2025. Among the products it covers are ebooks and, more broadly, digital publications sold in EU markets. Micro-enterprises have partial exemptions and existing titles have transition periods. The direction is unambiguous, though: a publisher selling digital publications into the EU needs an accessibility answer, and "our PDFs are whatever Word exported" is not one.

The scale of the existing problem: a 2024 study of scholarly PDFs found that about 75% fail every accessibility criterion tested.

What PDF/UA actually requires

PDF/UA (ISO 14289; PDF/UA-2 is the current revision, built on PDF 2.0) is the standard a screen reader can rely on. The requirements that matter in practice:

• Tagged structure. Headings, paragraphs, lists, tables, and footnotes marked as such in a structure tree. Without tags, a screen reader gets a bag of positioned characters and has to guess the reading order.
• Reading order. Two-column layouts, floats, and sidebars must declare the order in which they are read. Visual order and reading order routinely differ.
• Alternative text for every figure, and actual text (not an image of text) wherever possible.
• Math as MathML. An equation rendered as positioned glyphs is noise to a screen reader; PDF/UA-2 can embed MathML alongside the visual rendering, which screen readers can speak correctly.
• Metadata and language. Document title, primary language, and language switches marked inline.

Retrofitting these onto a finished PDF is remediation work: slow, manual, and priced per page by specialist vendors. The economical route is generating the structure at typesetting time, from a source that already knows what is a heading and what is a footnote.

Why LaTeX is well positioned for this

LaTeX sources are structured by construction: \section, \caption, \footnote, and math environments say exactly what each element is. The LaTeX Project's tagged-PDF effort makes that structure flow through to the PDF: with a current TeX distribution, documents can opt in via \DocumentMetadata:

\DocumentMetadata{
  pdfversion = 2.0,
  pdfstandard = ua-2,
  lang = en-US
}
\documentclass{article}
...

With this, standard constructs (sectioning, lists, footnotes, table of contents) produce tagged structure automatically, and math can carry MathML ("math-first" association of formula source with the rendered output). Coverage of the package ecosystem is still expanding, so validate real documents: run them through a PDF/UA checker such as veraPDF (free) or PAC, and fix what they flag. The most common finding is missing alt text on figures, which no tool can invent for you.

The practical checklist

• Compile with LuaLaTeX and a current TeX Live; declare \DocumentMetadata with pdfstandard=ua-2.
• Write alt text for every figure at authoring time. It is source material, like the caption.
• Declare document language, and mark inline language switches in multilingual texts.
• Validate with veraPDF; treat its report as the gate before publication.
• For book-length or math-heavy work, budget review time for tables and equations, where automated tagging still needs human checking.

Accessible by default

texlode generates tagged PDF/UA-2 output as its standard export: document structure comes from the LaTeX source, equations carry MathML, and the editor prompts for alt text when you add a figure, so accessibility is part of the normal workflow and needs no remediation project afterwards. The validation step from the checklist above is built in too: texlode runs veraPDF automatically after every export and shows the compliance result in the editor, and a pre-export check flags missing alt text, skipped heading levels, and undeclared document language before you publish. The same source also produces the print PDF and the EPUB, described on the main page.