PDF Accessibility Testing: A Team Test Plan
On this page
PDF accessibility testing is the process of confirming, with tools and with a person, that a PDF can be used by someone who cannot see it. It has two halves that are often confused: automated checking, which a machine can do in seconds, and manual testing, which needs a screen reader and a human decision. A workable test plan uses both, in a fixed order, and records the result so you can prove what you did.
Automated testing versus manual testing
Automated testing is a checker reading the file’s structure. It is fast, repeatable and free, and it catches the failures that break a document outright. The tools are covered in the PDF accessibility checker guide: PAC 2024 and veraPDF are the two that matter, and both map their findings back to WCAG 2.1 AA (the WCAG 2.1 specification), the standard most public-sector work is judged against. On a team, one person can run the automated checks on a batch in minutes; the manual pass is where the time goes, so schedule it per document rather than per folder.
Manual testing is a person listening to the document with a screen reader, or reading it with the structure visible, and judging whether it makes sense. A checker can tell you an image has alt text; only a person can tell you the alt text says something useful. The screen reader side is covered in how screen readers read a PDF.
Neither half is optional. Automated-only testing misses meaning; manual-only testing is slow and inconsistent across a team.
Which failures are automatable, and which need a person
| Failure | Test method |
|---|---|
| No tag tree (“Tagged PDF: Failed”) | Automated |
| Document language missing | Automated |
| Document title missing or is the filename | Automated |
| Image has no alt text at all | Automated |
| Alt text is present but meaningless | Manual |
| Table has no header cells | Automated |
| Reading order in a two-column layout is wrong | Manual |
| Heading levels skip (H1 straight to H3) | Automated flags, human confirms |
| Contrast of text and background | Automated flags, human confirms |
| Scanned page with no text layer | Automated |
The pattern is consistent: anything stored in metadata is automatable, and anything that requires judgement about meaning is not.
A numbered PDF accessibility test procedure
- Run the checker. Open the file in PAC 2024 or veraPDF and save the report. Record pass or fail on structure, language and title first.
- Read the structure pass. If tags failed, stop and remediate before testing further; nothing else is meaningful yet.
- Check images. List every figure and confirm the alt text describes its purpose. Mark decorative images as artifacts so they are skipped.
- Check tables. Confirm header cells are tagged and that a screen reader can associate a cell with its header.
- Open it with a screen reader. With NVDA on Windows or VoiceOver on macOS, read from the top and listen for the heading list, the reading order and the pronunciation of the language.
- Test navigation. Use the screen reader’s heading navigation to move through the document. If you can reach each section by heading, the outline is working.
- Record the result. Log the file, the date, the tool versions, the failures found and what you fixed.
Step 5 is where most problems surface. A file can pass every automated check and still read a footer in the middle of a paragraph, or announce a heading as plain text.
Recording your results
Keep a simple register, one row per document: file name, owner, date tested, checker used, pass or fail per category, and what was changed. This does three things. It gives you a defensible record if a complaint arrives, it stops you re-testing the same file twice, and it shows where the failures cluster. If most failures are scans, the fix is an OCR step; if they are exports from one authoring tool, the fix is upstream in that tool, which is a theme of document accessibility.
For a legacy archive, do not test every file by hand. Sample a representative set by type and age, use the sample to estimate how many files fail and how badly, then work through the highest-risk documents first: anything published, anything transactional, anything someone has complained about.
What a passing result looks like
A realistic result for a 30-page annual report:
Automated: structure pass, language en-US, title “2026 Annual Report”, 0 images without alt text after remediation, 1 table flagged for merged cells. Manual: read start to finish with NVDA; heading navigation reaches all six sections; one sidebar on page 4 reads after the main text, which is the correct logical order. Open items: alt text on the cover image reviewed and accepted by the author; merged-cell table reviewed by hand.
That is a file you can publish with confidence, and the register above is the evidence. If you would rather not open every file by hand first, a triage tool such as TagRight tags the file and returns exactly the kind of list in the open-items row, so a person only touches what needs judgement. The accessible PDF checklist turns this procedure into a per-file list, WCAG and PDFs explains the criteria the checks map to, and the free tier is on the pricing page. You can tag a PDF free to see the report a file produces before you commit to a workflow.
Frequently asked questions
What is the difference between automated and manual PDF accessibility testing?
Automated testing runs a checker such as PAC 2024 or veraPDF and reports on structure a machine can verify, like tags, language and title. Manual testing uses a screen reader and human judgement to confirm reading order, alt text quality and whether the document makes sense when you cannot see it.
Which PDF accessibility failures can be tested automatically?
Missing tags, missing language, missing title, images without alt text, tables without headers and missing PDF/UA metadata are all automatable. Whether alt text is meaningful and whether the reading order is logical are not.
How often should a team test its PDFs?
Test every new document before it is published, and re-test anything you remediate. For a legacy archive, sample a representative set to size the problem, then work through the files that carry the most risk first.
Do I need a screen reader to test PDF accessibility?
For a confident result, yes. NVDA is free on Windows, VoiceOver is built into macOS, and TalkBack ships on Android. Listening to a document is the only way to confirm it reads in a sensible order.
Can I pass a PDF accessibility test without fixing every warning?
Many checker warnings are advisories that need a human decision, not hard failures. The aim is a file that reads correctly and meets WCAG 2.1 AA, not a report with zero lines. Document the judgement calls you make.