PrivatePDF

Check a PDF for accessibility

Under the European Accessibility Act, a published PDF has to work for someone using a screen reader — which means it has to be tagged, titled, language-marked, and its images described. Checking that normally means paid desktop software or an upload service, and an upload service is exactly what you cannot use on a document that has not been published yet. This checks in your browser and gives you a report you can attach to a ticket. It is read-only: it tells you what is missing and it cannot invent the answers, because the missing thing is usually a judgement about meaning.

or drop a file anywhere on this panel

opened locally · 0 bytes sent

How it works

  1. 01Open the PDF.
  2. 02Download the report.
  3. 03Work through the failures in whatever produced the document — Word, InDesign, your publishing system.
  4. 04Export the PDF again and re-check.

What it checks

Eight things, each with an explanation you can act on rather than a code: whether the document is tagged with a structure tree at all; whether it declares its language; whether it has a title and is set to show that title rather than its filename; whether tagged images carry alternative text; whether headings are marked up and nest without skipping levels; whether tables mark their header cells; whether the text can be extracted and read aloud at all; whether form fields have labels; and whether pages set a keyboard tab order.

Why it does not fix anything

Every failure it finds needs a human decision. Alternative text is a sentence describing what an image conveys in its context — a tool that generated one would be producing confident, plausible, unverifiable text, and a wrong description is worse than a missing one because it silently passes the check. The same goes for which cells of a table are its headers, and for whether a reading order makes sense. So this reports, precisely, and leaves the writing to you.

Tagging has to come from the source

The most common result is a document with no structure tree at all, and that cannot be repaired after the fact by any tool — the information about what is a heading and what is a caption was discarded when the PDF was made. The fix is in whatever produced it: use real heading styles in Word, set alt text on images before exporting, and export with the accessibility option enabled rather than by printing to PDF, which throws the structure away.

What the report does not establish

The report says this itself, at the end, because it is the kind of document that gets forwarded as evidence. Colour contrast is not checked — that needs the rendered pixels compared with the text drawn on them. Reading order is checked for presence, not for sense: a tagged document whose tags are in the wrong order passes. Alternative text is checked for presence, not accuracy. Passing every check is necessary and not sufficient, it is not a statement of PDF/UA or WCAG conformance, and none of it substitutes for testing with a screen reader or an audit by somebody who uses one.

Questions

Can it make my PDF accessible?
No, and no tool honestly can. It tells you exactly what is missing; adding tags and descriptions has to happen in the program that produced the document, because that is where the meaning still exists.
Does passing every check mean the document is compliant?
No. These checks are necessary but not sufficient, and several important things — contrast, whether the reading order makes sense, whether a description is accurate — cannot be checked by a program at all. Treat a clean report as a floor, not a certificate.
Why is a document with no tags such a problem?
Without a structure tree a screen reader has to guess the reading order from where things sit on the page. That works on a simple page and fails on columns, tables, sidebars and captions — producing text that is fluent and in the wrong order, which is hard for a listener to even detect.
Is my document uploaded to be checked?
No, and for this tool that matters more than most: the documents people need to check are usually unpublished drafts. Everything is read by your own browser.