PDF/UA Checker - Test PDF/UA (ISO 14289) Compliance

What PDF/UA is really about

PDF/UA has very little to say about how a page looks. Its concern is the layer underneath: the tag tree. Tags tell a screen reader that a line is a heading, that three lines form a list, that a picture shows a sales chart. Anything that is only decoration, such as a page border or a running footer, has to be marked as an artifact so it gets skipped.

Most of the standard follows from that. Text that is neither tagged nor marked as an artifact gets read out of place, or not at all. A figure without alternate text is silent. A font without a Unicode map can look perfect on screen and still come out as nonsense when a screen reader tries to speak it.

What we check, and how

The checker reads the file directly rather than guessing from how it looks. It walks the whole tag tree, follows the RoleMap for custom tags, reads every font dictionary and annotation, and pulls the text from every page to see which characters sit inside tagged content. The rules it covers:

• The file is marked as tagged and has a structure tree.
• Page text is tagged or marked as an artifact. We measure this per character, so a missed footer on page 40 shows up.
• There is a document title, and the viewer is told to show it instead of the file name.
• The default language is set to a valid code such as en-IN or hi-IN.
• Figures and formulas have alternate text.
• Headings exist and do not jump levels, for example from H1 straight to H4.
• Tables have header cells and lists have list items.
• Custom tags map to standard ones.
• Fonts are embedded and have a Unicode mapping.
• Links have descriptions, annotations are tagged, and pages with links or fields use structure tab order.
• Form fields have a tooltip name, and there are no XFA forms.
• Password settings do not block screen readers.
• The XMP metadata declares PDF/UA (pdfuaid:part).

A typical result, and what it means

When we tested a 65-page employee handbook exported from Word 2024, it was tagged properly: over 3,000 tags, a language set, headings, lists and two tables with header rows. It still failed on 1,548 figures with no alternate text. Word tags every decorative shape and icon as a figure unless you mark it as decorative, so a handbook full of small icons fails this rule badly even when the real images are described.

That is a common pattern. Exporting from Word or Google Docs with tagging switched on gets most of the structure right. What is left is usually alt text, the document title, and headings that skip levels because someone picked "Heading 4" for its font size.

What still needs a person

Software can prove a tag tree is present. It cannot tell whether the tags follow the order a reader would expect, or whether "image1" is a useful description. Those items appear under "Needs a human check" in the report.

The quickest manual test is to listen. Open the file in NVDA, which is free on Windows, or use Read Out Loud in Acrobat Reader, and follow along for a couple of pages. If the order would confuse you, it will confuse a screen reader user.