How to Run a PDF Accessibility Check Before Submitting or Publishing It
What an automated PDF accessibility check can and cannot tell you, how to run one, which failures to fix first, and the ten-minute manual check.
Running an accessibility check before you publish is worth doing, and it is worth being honest about what it proves: an automated checker finds mechanical failures. It cannot tell you the document is accessible, because most of the things that make a document usable are judgements no tool can make.
Used properly it is still the fastest way to catch the problems that would otherwise reach a reader.
What automated checking can and cannot see
Machine-checkable — is the document tagged, do figures have alternative text, do tables have header cells, is a document language set, is there a title, does page content sit outside the tag structure, do form fields have accessible names. These are yes-or-no and a tool answers them reliably.
Not machine-checkable — whether the alt text describes the image usefully, whether the reading order makes sense, whether headings reflect the real hierarchy, whether colour contrast is adequate in context, whether link text means anything out of context. A tool can see that alt text exists. It cannot see that it says "image1.png".
This is why every serious checker, including the one on this site, reports findings in three buckets: failures, things needing human review, and things that could not be tested. The middle bucket is not padding — it is the part that actually decides whether a document works.
Run the check
The free PDF accessibility checker here runs a WCAG 2.2 AA-relevant, PDF/UA-1-oriented ruleset across eight families of check — tags and structure, images and figures, tables, forms, headings and lists, and the rest — against the tag tree, the page content, the metadata and the rendered pages, and shows the evidence behind each finding.
Other routes: Adobe Acrobat Pro has a built-in full check, and PAC, a free Windows tool, is the usual reference for PDF/UA conformance specifically. Running two checkers is not wasted effort — their rulesets differ and they disagree at the edges.
Fix in this order
Not everything found is equally serious, and fixing in the wrong order wastes time.
- Is it tagged at all? An untagged PDF fails everything downstream. Nothing else is worth doing first.
- Content outside the structure. Text a screen reader cannot reach is invisible, whatever else is right.
- Document language and title. Two fields, seconds to set, and both affect how the document is announced.
- Alt text on informative figures, and artifact marking on decorative ones.
- Table headers. A data table without header cells is unusable non-visually.
- Reading order, checked by listening rather than by looking.
- Contrast and link text — human judgement, and usually the last pass.
The check that matters most takes ten minutes
Install a screen reader and listen to the document. NVDA is free on Windows; VoiceOver ships with macOS.
You do not need to be proficient. Listening to the first two pages tells you more than any report: whether the headings make sense, whether the images are described, whether the order is followable, whether page furniture interrupts the text. Every serious accessibility practitioner does this, and it is the step most people skip.
Fix the source where you can
Almost everything above is easier to fix in Word or InDesign and re-export with tagging enabled than to repair in the PDF. Repairs to a PDF's tag tree also have to be redone after every revision, which is how documents silently regress between versions.
If the source is gone, PDF-side repair needs a tool that edits tag structure — Acrobat Pro being the common one. This site's editor changes page content on files up to 10 MB and 100 pages; it does not edit the tag tree, so it is not the tool for structural remediation.
Frequently asked questions
Does passing an automated check mean the document is compliant? No. It means no machine-detectable failures remain. Compliance depends on judgements a tool cannot make.
Which checker should I use? Run more than one. Their rulesets differ. PAC is the usual reference for PDF/UA.
What is the single most common failure? The document is not tagged, which fails almost everything else by implication.
Do I need to check a scanned PDF? A scan with no text layer is inaccessible by definition — there is nothing to read. It needs OCR and then tagging.
How long does remediation take? A tagged document exported properly from a good source may need minutes. An untagged, complex, scanned document can take a very long time, which is the argument for fixing the source.
