PDF/UA and WCAG 2.1 AA
The rules you are held to name WCAG 2.1 Level AA. The standard that says what that means for a PDF specifically is PDF/UA. They are not competitors and you do not choose between them — one is the obligation and the other is how you satisfy it in this file format.
The short version
| WCAG 2.1 Level AA | PDF/UA (ISO 14289) | |
|---|---|---|
| What it is | A standard for web content of any kind | A standard for PDF files specifically |
| Who requires it | ADA Title II and HHS Section 504, by name | Nobody, by name — it is the means, not the mandate |
| What it says about tags | “Information and relationships can be programmatically determined” | Exactly which structure elements to use, how to nest them, and what each must carry |
| Testability | Partly automated; much of it is judgment | Largely machine-checkable, by design |
In practice: a PDF that conforms to PDF/UA will satisfy most of the WCAG criteria that apply to documents, and conformance checking tools are built around PDF/UA because it is the one written precisely enough to check. That is why a scan report speaks in PDF/UA checks and then maps them back to WCAG criteria — the check is what a machine can verify, the criterion is what you are accountable for.
Which WCAG criteria documents actually fail
Measured, not theorised. Across 15 published federal documents scanned on September 27, 2026, the findings mapped to these criteria. The count is how many separate findings referenced each one.
| Criterion | What it requires | Findings |
|---|---|---|
| 1.3.1 | Info and Relationships — structure must survive being read aloud | 109 |
| 4.1.2 | Name, Role, Value — form fields need labels | 19 |
| 2.4.4 | Link Purpose — a link must say where it goes | 5 |
| 1.1.1 | Non-text Content — images need a text alternative | 5 |
| 2.4.2 | Page Titled — the document needs a real, displayed title | 5 |
| 1.3.2 | Meaningful Sequence — the reading order must make sense | 5 |
| 2.4.3 | Focus Order — keyboard order must be predictable | 4 |
| 3.1.1 | Language of Page — the document must declare its language | 4 |
| 3.1.2 | Language of Parts — passages in another language must say so | 4 |
| 2.4.6 | Headings and Labels — descriptive, properly nested | 4 |
| 2.4.5 | Multiple Ways — more than one route to a page | 3 |
The distribution is extremely lopsided. 1.3.1 — Info and Relationships — structure must survive being read aloud — accounts for 109 of the findings, several times any other criterion. That is worth sitting with, because it says something useful about where to spend effort: the overwhelming majority of document accessibility failure is structural. Not colour contrast, not alt text, not language declarations — structure. Headings that are not tagged as headings, tables that are not tagged as tables, lists that are visually lists and structurally paragraphs.
It follows that the single highest-leverage change most organisations can make is upstream of remediation entirely: get the people producing documents to use real heading styles and real tables in the authoring tool, and export with tagging enabled. That costs nothing and prevents the expensive category from being created.
Where the two standards come apart
Conformance to PDF/UA is necessary but not sufficient. Three WCAG criteria in particular are not fully decided by it:
- 1.3.2 Meaningful Sequence. PDF/UA requires that a reading order exists. It cannot require that the order is sensible, because that is a judgment about content. A document whose tags run right-column-first passes conformance and fails a reader.
- 1.1.1 Non-text Content. PDF/UA requires that a figure has an
/Altentry. Whether the text in it describes the image is not machine-checkable, soAlt="image1.png"conforms. - 1.3.1 Info and Relationships. A table tagged with header cells conforms even if the content is not a table at all — the oldest layout hack there is, and still common in documents exported from Word.
The PDF Association’s Matterhorn Protocol makes this explicit: it sorts its checkpoints into machine-verifiable and human-judgment categories, and the second category is not small. Any tool that claims a document is accessible on the strength of an automated pass is overstating what it checked.
PDFCraft’s second layer goes at exactly those three. We reconstruct the visual reading order from page geometry and compare it with the tag order; we test whether a tagged table’s cells align on a grid; and we flag alt text that is a filename, a placeholder, or pasted onto every figure. Those findings sit alongside the 39 machine checks in the same report, labelled by which layer found them. The overview has the detail.
Which should you ask a vendor for?
Ask for WCAG 2.1 AA, because that is what the rules name and what you will be measured against. Expect the vendor to work to PDF/UA, because that is how the job is actually specified for this format. If a vendor offers only a PDF/UA conformance report as the deliverable, ask specifically how reading order and alt text quality were verified — those are the parts a conformance pass does not cover, and the answer should involve a person.