Accessible PDF: tags, reading order and testing

A PDF can look perfect to the person viewing it and be almost unusable for someone listening to it with a screen reader or reading it zoomed in on a phone. The difference lies in something you cannot see: structure. An accessible PDF contains, besides the appearance of the pages, a description of what each element is: this is a heading, this a paragraph, this a table with its headers.

Making your documents accessible is not specialist work. In most cases it is enough to build the source file properly and export it with the right options.

What tags are

Tags are a hidden tree inside the PDF that assigns a role to the content, much like the HTML of a web page:

Document
├── H1  «Installation manual»
├── P   «This document describes…»
├── H2  «Requirements»
├── L   (list)
│   ├── LI «Up-to-date operating system»
│   └── LI «Network connection»
└── Figure (alt text: «Wiring diagram»)

A screen reader uses this tree to announce headings, allow jumping from one section to another, read tables cell by cell and describe images. A PDF without tags is, for these technologies, an undifferentiated stream of text at best.

The reference standard for accessible PDFs is PDF/UA, defined by ISO 14289.

Start from the source document

Almost all accessibility is decided before export:

1. Use real heading styles, not enlarged bold text. Only then do they become H1, H2, H3 tags.
2. Respect the hierarchy: a single main heading, no skipped levels.
3. Create lists with the program's list feature, not with dashes typed by hand.
4. Write alternative text for images and charts that convey information; mark as decorative those that do not.
5. Define the header row of tables and repeat it on following pages.
6. Set the document language, so speech synthesis uses the right pronunciation.
7. Write descriptive links: «request form» rather than «click here».
8. Fill in the title in the document properties: it is what appears in the window bar.

Export with tags

• Word: Save As › PDF › Options, checking that the option for document structure tags for accessibility is enabled.
• LibreOffice: Export as PDF, General tab, tagged PDF option; the same dialog also offers PDF/UA compliance.
• LaTeX: tagging support is under active development in recent releases; for documents that must certainly be accessible, check the result carefully.

Printing to PDF through a «virtual printer», on the other hand, almost always produces an untagged file: that is the method to avoid.

Reading order

Tags also define the order in which content is read. In single-column documents it matches the visual order. Problems arise with:

• text boxes and anchored objects, which can end up read at a random point;
• multi-column layouts built with invisible tables;
• repeated headers and footers, which should be marked as artefacts and not read on every page.

The practical rule is simple: the more linear the layout, the more likely the reading order will be correct without manual intervention.

Tables

Tables are the most delicate element. An accessible data table has:

• a header row declared as such;
• no merged cells, or the bare minimum;
• no empty rows used to create space;
• a title or caption saying what it contains.

Tables used only for layout, for example to place a logo next to an address, should be avoided: they confuse people navigating by keyboard or voice.

Testing

An automatic check does not replace a direct test: it verifies that alternative text exists, not that it is useful. Listening to your own document for two minutes reveals problems no validator reports.

In short

• An accessible PDF contains tags describing the role of each element.
• Accessibility is built in the source: heading styles, real lists, alt text, language.
• It must be exported with tagging enabled; virtual PDF printing loses the tags.
• Linear layout and simple tables ensure a correct reading order.
• PAC and built-in checkers help, but a test with a screen reader remains decisive.