PDF Accessibility Checker

About PDF Accessibility Checker

This tool reads a PDF and changes nothing. It walks the catalog, the structure tree, every page's content stream, annotations and fonts, and reports about twenty checks that commonly fail under PDF/UA (ISO 14289) as pass, fail or warning, with the page numbers where each problem occurs.

It is a basic self-check in the spirit of PAC and the Matterhorn Protocol, not a certification. A machine can tell whether a figure has alternate text; it cannot tell whether that text actually describes the figure, or whether the tagged reading order matches the order a person would read. Even with every check green, listening to the file in a screen reader is still part of the job.

What tagging actually adds

Under the hood a PDF page is a list of drawing commands: put these glyphs at these coordinates. In a two-column layout, the first line of each column sits at the same height, and the order of the commands is whatever the authoring application chose. From those commands alone, a screen reader cannot know which line is a heading or where a paragraph ends.

A tagged PDF numbers each piece of page content (a marked-content ID, or MCID) and keeps a separate structure tree that records the logical order and role: H1, then P, then Figure, then Table. Running headers, page numbers and decorative rules are marked as Artifacts so assistive technology skips them. PDF/UA builds on that and requires a document language, a title shown in the window, alternate text for figures, header cells in tables, descriptions for form fields and more.

What gets checked, and the usual culprits

Checks fall into five groups: document, structure, content, annotations and fonts. The failures seen most often in real files:

• No MarkInfo /Marked and no structure tree. Nearly every file made by printing to a PDF printer driver lands here. Export with the save-as-PDF path and turn on the accessibility/tags option instead.
• No /Lang. A screen reader may pronounce the whole document with the wrong language rules. The value is a language tag such as en-US.
• Missing title or DisplayDocTitle. The window should show the document title, not report_final_v3.pdf.
• Untagged content. A properly tagged file later stamped, annotated-and-flattened or merged in another tool typically leaves only the new material outside the tags.
• Image-only pages. Scans contain pictures of text and nothing to read until OCR adds a text layer.
• Figures without /Alt, tables without TH cells, skipped heading levels (H1 followed by H3), links without descriptions or structure parents, form widgets without a tooltip (/TU), pages with annotations whose tab order isn't structure order (/Tabs /S), non-embedded fonts and fonts without a ToUnicode map.

Why the contrast check is only an estimate

Text contrast is computed with the WCAG 2.x formula against the background right under the text. The tool follows the content stream in order and takes the color of the last box or shape painted before the text that covers its position, or white paper if there is none. So white text on a navy panel passes, and dark gray text on a black panel gets a warning. When text sits on a photo or a gradient, the content stream alone doesn't say what color is there, so the page is rendered in your browser and the most common color around the text is read as its background.

It is still an estimate: shapes are judged by their bounding boxes, transparency, blend modes and special color spaces are ignored, and only one point at the start of each text run is looked at. That is why this check only ever raises a warning. The thresholds are 3:1 for text 18 pt and larger and 4.5:1 for everything else.

PDF/UA-1, PDF/UA-2 and where the rules bite

PDF/UA-1 (ISO 14289-1:2014) is built on PDF 1.7; PDF/UA-2 (ISO 14289-2:2024) is built on PDF 2.0. The tool reads pdfuaid:part in the XMP metadata to show which one a file claims, and otherwise concentrates on requirements the two share. A missing claim is a warning, and the claim only means something once the requirements are really met. Dynamic XFA forms are not allowed under PDF/UA and fail immediately.

In the United States, documents posted by federal agencies fall under Section 508, whose standards incorporate WCAG 2.0 and apply it to electronic documents; PDF/UA is widely used as the technical way to get a PDF there, but the law points to WCAG rather than naming PDF/UA itself. In the EU, the web accessibility directive for public-sector sites and apps likewise works through the harmonised standard EN 301 549, which covers downloadable documents.

Worked examples

How to use it

1. Choose a PDF. It is only read, never modified.
2. Review the pass/fail/warning summary, the grouped checks and the affected pages.
3. Download the report if you need to hand it to whoever produced the document.

Key terms

MCID

A number attached to a stretch of page content so a structure element can point to what was actually drawn.

Artifact

Content marked as not part of the reading: page numbers, running heads, decorative lines.

ToUnicode

A map from a font's glyph codes back to Unicode characters. Without it, screen reading, copying and search break.

Frequently asked questions

Related guides

• PDF/A and PDF/X Explained: Archive-Grade PDFs, Print-Ready PDFs, and Who Actually Requires Them