A PDF that looks perfectly clear can be completely unusable to a blind reader, and nothing on screen tells you which kind you have made.
What a screen reader needs
Visual layout communicates structure to a sighted reader: this is large and bold so it is a heading, these are indented with bullets so they are a list, this is a grid so it is a table.
None of that is available to a screen reader. It needs the structure stated explicitly, as tags in the file — an H1 tag, a list tag, a table with header cells marked as headers. Without them the reader receives a stream of text with no signposts and no way to skip between sections.
This is why "it looks clear" is not a test. The visual and the semantic are separate layers, and a PDF can have a perfect first layer and an empty second one.
The five things that matter
| Requirement | Why | How to check |
|---|---|---|
| Tagged content | Structure for navigation | Document properties: "Tagged PDF: Yes" |
| Correct reading order | Order text is announced in | Read the tag tree, not the page |
| Alt text on images | Meaningful images described | Every non-decorative image |
| Real text, not a scan | Nothing to read otherwise | Try selecting a line |
| Document language set | Correct pronunciation | Properties: Advanced, Language |
| Table headers marked | Cells announced with context | Header cells tagged as headers |
Reading order is the one that most often fails silently. It follows the tag tree, not the visual layout, so a two-column page can be tagged in a sequence that interleaves the columns. It looks fine and reads as nonsense.
The legal position
This is not a best-practices question for a lot of organisations.
In the United States, ADA Title II and Section 508 require public sector digital content to be accessible, and the 2024 DOJ rule set WCAG 2.1 AA as the standard with compliance deadlines in 2026 and 2027. In the European Union, the Web Accessibility Directive covers public sector bodies and the European Accessibility Act extends obligations to many private services from June 2025. In the United Kingdom, the 2018 public sector accessibility regulations apply, and documents published after September 2018 are in scope.
The practical consequence: a scanned PDF published on a public sector site is a compliance failure. Not suboptimal — non-compliant. That is worth knowing before publishing a scanned annual report.
Get it right in the source
Tagging an untagged PDF afterwards is slow, manual and error-prone. Exporting correctly from a document that already has structure takes no extra time at all.
- Use real heading styles in Word or Docs, not text made large and bold. Heading 1 and Heading 2 become H1 and H2 tags; manually formatted text becomes an untagged paragraph.
- Use real lists and real tables, not tabs and spaces arranged to look like them.
- Add alt text to every meaningful image, and mark decorative ones as decorative so they are skipped rather than announced as "image".
- Export, do not print. Print to PDF discards the tag tree entirely. Save As or Export preserves it — the same distinction that preserves hyperlinks, as in how to convert PowerPoint to PDF properly.
- Set the document language in the properties, so a screen reader pronounces it correctly.
That fourth point catches a lot of otherwise careful work. Everything is structured correctly in the source, then Print to PDF flattens it into an untagged page and all of it is lost.
Writing alt text that helps
Alt text is where good intentions produce useless output most often.
The test is functional: if the image vanished, what would the reader need to have been told? Not what the image looks like — what it conveys.
- Decorative image: mark it decorative. Do not describe a border.
- A photograph illustrating text: describe what matters, briefly. "A queue outside a job centre" beats "photograph".
- A chart: state the finding, not the shape. "Revenue grew 40% between 2023 and 2025" is the content; "bar chart" is not.
- A logo: the organisation name.
- Text in an image: the text itself, verbatim.
- Never start with "image of". The screen reader already announces that it is an image.
For a complex chart, alt text is not enough — put the underlying figures in a table nearby. A sentence cannot carry a dataset, and pretending otherwise fails the reader more subtly than no alt text at all.
What about a scan?
A scanned PDF is a picture. A screen reader announces an empty page, and there is no tagging that fixes that because there is nothing to tag.
OCR adds a text layer, which makes it readable, and that is a genuine improvement. It does not make it accessible: OCR produces text without structure, so there are still no headings, no list semantics and no table headers.
So the honest sequence for a scanned document is OCR with Image to Text, then rebuild it as a real document with real styles, then export tagged. That is more work than people expect, and it is why publishing scans is a decision with a cost rather than a neutral shortcut.
Frequently asked questions
How do I know if my PDF is accessible?
Check document properties for "Tagged PDF: Yes" as a first pass, then run an accessibility checker. Tags alone are not sufficient — reading order follows the tag tree rather than the visual layout, so a tagged document can still be announced in the wrong order.
Is a scanned PDF accessible?
No. It is a picture, so a screen reader finds nothing to announce. OCR adds a text layer and makes it readable, but the result has no headings, list semantics or table headers, so it is still not structurally accessible.
Do I legally have to make PDFs accessible?
For public sector bodies in the US, EU and UK, yes. ADA Title II and Section 508 in the US, the Web Accessibility Directive in the EU, and the 2018 regulations in the UK. The European Accessibility Act extends obligations to many private services from 2025.
Why does Print to PDF break accessibility?
Because it sends the document through a print driver, which produces a flat rendering with no tag tree. All the structure you created with heading styles and real tables is discarded. Use Save As or Export instead.
What makes good alt text?
Describe what the image conveys, not what it looks like. For a chart, state the finding — "revenue grew 40%" — rather than "bar chart". Never begin with "image of", since the screen reader already says so. Mark decorative images as decorative so they are skipped.
Can I fix accessibility after exporting?
You can add tags to an existing PDF, but it is slow and manual, and reading order usually needs fixing by hand. Adding real heading styles in the source and exporting properly takes no extra time and gets it right the first time.