The rule
What an accessible PDF actually is
“Accessible” has a precise technical meaning. Here are the twelve things a PDF needs, in plain English, and the standards that define them.
The twelve requirements
- Tags for every piece of content. The file carries a hidden outline that says what each part is: a title, a heading, a paragraph, a list, a table, a picture, a link, a form field. Page numbers, running headers and decorative lines are marked as decoration so a screen reader skips them.
- Tags connected to what is on the page. Each tag points to the exact text or image it describes. This is the hard part, and the reason most quick fixes fall short.
- A logical reading order. The tags are in the order a person would read the page, including two-column layouts and sidebars, and heading levels are not skipped.
- Real text. Text is stored as characters, with embedded fonts and correct character maps, so it can be read aloud, searched and copied. A scanned image of text is not real text.
- Title, language and identifying information. The document has a real title (not a file name), declares its language, and identifies itself as tagged and as PDF/UA.
- Descriptions for images. Every informative image, chart or logo has alternative text that conveys the same information. Purely decorative images are marked as decoration.
- Real tables. Data tables have header cells tied to their rows and columns. Tables used only for layout are not tagged as tables.
- Usable forms. Fillable fields are real form fields with labels a screen reader announces, a sensible tab order, and required fields indicated in text.
- Links that make sense. Links say where they go and are tagged as links.
- Enough contrast. Text has a contrast ratio of at least 4.5 to 1 against its background, or 3 to 1 for large text.
- Bookmarks and no blocking security. Long documents have bookmarks for navigation, and security settings do not stop assistive technology from reading the file.
- Language changes marked. Passages in another language, such as a Spanish or Portuguese translation, are marked so they are read in the right voice.
The standards
The legal standard is WCAG 2.1 Level AA (28 CFR § 35.200). WCAG was written for web pages, so DOJ applies it to documents “to the extent those criteria can be applied” and points to W3C's guidance on non-web content, WCAG2ICT:
“public entities generally must ensure conformance to the WCAG 2.1 Level AA success criteria to the extent those criteria can be applied. … public entities may find it helpful to consult W3C's guidance on non-web information and communications technology.”
The document-specific standard is PDF/UA-1 (ISO 14289-1), tested with the Matterhorn Protocol. DOJ did not adopt PDF/UA as the legal standard, so a PDF/UA-1 pass is strong evidence of conformance, not proof. That is why our conformance statement reports both a PDF/UA-1 validation and a WCAG 2.1 AA document checklist that covers what PDF/UA does not, such as contrast, the quality of alt text, and link purpose.
What automated checks can and cannot verify
| Kind of check | Examples | Who can do it |
|---|---|---|
| Machine-checkable | Tags present and valid, fonts embedded, title and language set, every item tagged or marked as decoration, PDF/UA identifier, measured contrast of text | Software, with a pass or fail answer. We use veraPDF for PDF/UA-1 and our own contrast and content checks. |
| Needs judgment | Is this line a heading, and what level? Is this a data table? Which cells are headers? What is the reading order? What should the alt text say? | A vision model can propose an answer; a person confirms it. |
| Human only | Is the alt text for this chart true? Can this form be filled out without sight? Is this document still in use, or should it be archived? | A trained reviewer, sometimes with a screen reader. |
No checker can measure contrast inside a picture, confirm that alt text is accurate, or tell whether a reading order makes sense. A green checkmark from an automated checker is a good start, not the finish line.