Your PDFs Are Probably the Least Accessible Thing You Publish
A council fixes its website, publishes the accessibility statement, and links to a hundred PDFs that a screen reader cannot read a single word of. This is the most common shape of accessibility failure in the public sector, and almost no scanner looks at it.
The four things that decide it
Is there any text at all. A scan of a printed page is an image. A screen reader gets nothing, search engines get nothing, and no amount of remediation elsewhere helps. This is the single most common and most severe PDF failure.
Is it tagged. Tags tell assistive technology which text is a heading, which is a list, and what order to read in. Without them the reader guesses from visual position, which breaks the moment there are columns, tables or a sidebar.
Is the language set. Screen readers pick a pronunciation engine from it. English read by a German voice is not merely awkward, it is unintelligible.
Is there a title, and is it used. Without one, readers announce the filename. "minutes-final-v3-REVISED.pdf" tells a listener nothing.
The one setting nobody knows about
You can set a document title and still have readers announce the filename, because displaying the title is a separate flag. Adobe calls it "Display document title" in the initial view settings, and it is off by default.
We tested the IRS Form W-9 while building this. Six pages, properly tagged, language set, title set — and that flag off, so assistive technology still reads out the filename. If the IRS misses it, everybody misses it.
Export, do not print
The single highest-value habit: in Word, use Save as PDF with document structure tags enabled. Do not use Print to PDF, which throws the structure away and hands you a visually identical file that is inaccessible.
Most inaccessible PDFs were accessible documents until somebody printed them.
Checking yours
Our PDF check reads the document’s own markers: whether it contains text, whether it is tagged, whether it has a structure tree, its language, its title, and whether that title is displayed. It reports what it finds and what it cannot decide.
It is not a full PDF/UA audit and does not claim to be. Whether a table’s headers are associated correctly, or whether the reading order actually makes sense, still needs a person.