Accessible PDFs: What PDF/UA Requires and How to Check Your Documents

Who has to care

Public bodies in the EU have had to make their websites and the documents on them accessible for some years under the Web Accessibility Directive, implemented in Germany by the BITV 2.0. The European Accessibility Act (Directive (EU) 2019/882) extends accessibility requirements to private businesses for certain products and services from 28 June 2025, among them e-commerce, banking services, e-books and passenger transport services. Germany implemented it in the Barrierefreiheitsstärkungsgesetz (BFSG). Micro-enterprises with fewer than ten employees and an annual turnover or balance sheet total of no more than 2 million euros are exempt from the requirements for services.

Whether a particular PDF falls under these rules depends on the service it belongs to; terms and conditions, product information and statements in an online shop or a banking app are typical candidates. Even where no law requires it, an accessible PDF reaches more readers, works better on phones and is easier for search engines and AI tools to read correctly. This guide is not legal advice; for a binding assessment, ask a specialist.

What makes a PDF accessible

The core is the tag structure, a hidden layer that says what each part of the page is: this is a level-two heading, this a list with three items, this a table whose first row holds the column headers, this a figure with the description Sales by quarter, 2025. Assistive technology reads this structure, not the visual layout. A PDF without tags is called untagged; for a screen reader it is just a stream of text in whatever order it was drawn.

Around the tags come further requirements: the document language must be set so that the screen reader uses the right pronunciation, the title should appear in the window instead of the file name, every figure needs alternative text, decorative elements must be marked as artifacts so they are skipped, form fields need a description, links need a meaningful text and fonts must be embedded with a correct mapping to Unicode, so that the letters read are the letters shown.

The international standard for this is PDF/UA (ISO 14289). Its first part, PDF/UA-1, is the version most validators and procurement requirements refer to. The Matterhorn Protocol, published by the PDF Association, breaks PDF/UA-1 into concrete checkpoints and states which of them a machine can decide and which require human judgment.

Accessible from the start

Accessibility is far easier to create in the source document than to add to a finished PDF. In Word, use the built-in heading styles instead of bold, enlarged text, real lists instead of hyphens, tables with a marked header row and alternative text on every image. When saving as PDF, keep Document structure tags for accessibility switched on in the options. LibreOffice, InDesign and other layout programs have similar export options for tagged PDF.

Scanned documents are a special case: a scan is a picture, and a picture of text is silent for a screen reader. Text recognition with OCR PDF adds a text layer first. Tagging a scan properly afterwards still takes a specialist tool, so for important documents the better route is to recreate them digitally.

Checking a PDF step by step

Drop the PDF on the PDF Accessibility Checker. The report lists every check with passed, failed or warning and the pages concerned: tags, document language, title, PDF/UA identifier, alternative text of figures, table and heading structure, content outside the tags, embedded fonts and Unicode mapping, tab order, descriptions of form fields and links, encryption settings and scanned pages.

The tool then offers the fixes that have one right answer or that you confirm yourself: the document language, detected from the text, the title, the tab order following the structure, tooltips for form fields suggested from their names, descriptions for links, and alternative text for existing figures, which you write next to a crop of each figure. Improve accessibility applies them and you download the improved file.

• Language and title set, and the title shown in the window
• Every figure described, decorative images marked as artifacts
• Headings in a logical order, without skipped levels
• Tables with header cells, lists tagged as lists
• Form fields and links with a description
• No scanned pages without a text layer

What a checker cannot decide

An automatic check can see that a figure has alternative text, but not whether that text describes it well. It can see that headings exist, but not whether the heading levels match the meaning. It can see that there is a reading order, but not whether it follows the logic of the content. The Matterhorn Protocol lists these checkpoints as requiring a human, and a passing report is not proof of accessibility.

For a real test, open the PDF with a screen reader such as NVDA on Windows or VoiceOver on macOS and listen to a page, jump through the headings and read a table. You quickly notice whether the document makes sense when heard instead of seen.

Toolula deliberately never adds tags to an untagged PDF. Guessed structure, where a large bold line is assumed to be a heading and two columns are assumed to be read left before right, misleads screen readers more than no structure at all. For untagged files the report explains how to export them again with tags.

The PDF/UA identifier

A PDF can declare in its metadata that it conforms to PDF/UA. Validators and assistive technology trust this declaration. Toolula therefore adds it only when every other check passes after the fixes, never as a cosmetic label on a document that still has open faults. Signed PDFs are not changed, because any change would break the signature; encrypted PDFs are only checked for the permission that lets assistive technology read them.

Fixing the common faults at the source

Most faults the checker reports are fixed fastest in the program the document came from, followed by a new export. A few examples from Word, the most common source of everyday PDFs:

• Missing alternative text: right-click the picture, choose Edit Alt Text and describe what the picture shows and why it is there. Mark purely decorative pictures as decorative.
• Headings not recognised: apply the styles Heading 1, Heading 2 and so on instead of enlarging and bolding normal text.
• Tables without header cells: select the first row and switch on Repeat Header Rows in the table properties.
• Wrong language: set the language of the text under Review, Language, so that the PDF gets the right language tag.
• File name shown instead of a title: enter a title under File, Info, Properties. Toolula's checker can also set the title and make readers show it.

Frequently asked questions