A PDF is a picture of a document with some text attached. Whether that text makes sense to a screen reader depends on a layer most people never see: the tag tree. Two PDFs can look identical on screen, and one reads as a clear document with headings and tables while the other reads as a jumble of words in whatever order the layout engine drew them.
The standards
PDF/UA (Universal Accessibility) is the ISO standard for accessible PDF. PDF/UA-1 is ISO 14289-1 and applies to PDF 1.x files, which is still most of them. PDF/UA-2, ISO 14289-2:2024, applies to PDF 2.0 files. The PDF Association's Matterhorn Protocol turns PDF/UA-1 into a list of testable failure conditions, which is how most checkers, ours included, organise their tests.
Laws that mention WCAG generally cover PDFs published on a website too. In practice, a PDF that meets PDF/UA goes a long way toward meeting WCAG for that document.
What makes a PDF accessible
- It is tagged. Headings, paragraphs, lists, tables and figures are marked as what they are. An untagged PDF has no structure for assistive technology to use.
- The reading order is right. The tag order is the order a screen reader reads. Multi-column layouts, sidebars and footnotes are the usual trouble.
- Figures have alternative text, and purely decorative graphics are marked as artifacts so they are skipped.
- Tables have header cells that say what they are headers for.
- The document has a title and a language, and the viewer is set to show the title rather than the file name.
- Fonts are embedded with Unicode mappings, so the text can be extracted as real characters.
- Links and form fields have names, and the tab order follows the structure.
- Encryption doesn't block assistive technology. Some security settings stop screen readers from reading the text at all.
Getting it right at the source
Fixing a PDF after the fact is slow. Fixing the source document is fast:
- In Word, use the built-in heading styles, real lists, alt text on images and header rows on tables. When exporting on Windows, check that "Document structure tags for accessibility" is ticked in the PDF options.
- In InDesign, map paragraph styles to export tags and set the articles panel order before exporting.
- Scanned documents need OCR first; a scan without it is an image with no text at all.
What a checker can and can't tell you
Automated PDF checks are good at structural facts: whether the file is tagged, whether figures have alt text, whether tables have header cells, whether a language and title are set, whether fonts are embedded, whether headings skip levels. AccessiSight's PDF scanner checks all of these and more, and it tells you plainly when a PDF 2.0 file needs PDF/UA-2 checks it doesn't run.
What no checker can fully confirm is whether the reading order makes sense, whether alt text is accurate, and whether table headers are the right ones. Open the tag tree, or read the document with a screen reader, for anything important: forms, statements, anything people need to act on.
References: ISO 14289-1:2014 (PDF/UA-1) and ISO 14289-2:2024 (PDF/UA-2). The Matterhorn Protocol is published by the PDF Association.