Cost
Free
/ Free tool · PDF accessibility

Free PDF accessibility checker. Results in seconds.

Accessibility law covers documents, not just pages, and most organisations sit on thousands of PDFs nobody ever checked. Upload one and see whether assistive technology can read it at all.

The file is analyzed in memory and discarded. We store nothing but your email, if you leave one.

/ PDF accessibility guide

What this PDF check tests, and what it leaves to a person

The checker opens the file you upload and runs six structural tests on it. They catch the faults that stop a screen reader cold. They do not prove a document is accessible, and the sections below say exactly where the line is.

/ The laws

What each statute says

Web Accessibility Directive

Directive (EU) 2016/2102, Art. 1(4)(a) and Art. 4

Who it binds
Websites and mobile apps of public sector bodies in the EU, including the documents published on them.
What it requires
Content made perceivable, operable, understandable and robust. Office file formats published before 23 September 2018 are outside its scope, unless they are needed for active administrative processes.
Enforcement
Set by each member state through its national law.

EN 301 549, clause 10

ETSI EN 301 549 V3.2.1, clause 10 Non-web documents

Who it binds
Documents that are not web pages, including documents offered for download from a web page.
What it requires
WCAG success criteria applied to the document itself. A PDF rendered inside the page falls under clause 9 instead.
Enforcement
Through the laws that point to the standard, such as the national acts carrying Directive (EU) 2016/2102.
Source
etsi.org

BFSGV (Germany, under the EAA)

§ 12 BFSGV

Who it binds
Services covered by the Barrierefreiheitsstärkungsgesetz, the German law carrying the European Accessibility Act.
What it requires
Information about how the service works offered through more than one sensory channel, in text formats that can be turned into other accessible formats, readable, with alternatives for non-text content, and perceivable, operable, understandable and robust when digital.
Enforcement
The market surveillance authority of the Länder (MLBF).

Section 508 (United States)

36 CFR part 1194, Appendix A, E205.4 and E205.4.1

Who it binds
Electronic content of US federal agencies, public facing content in particular.
What it requires
WCAG 2.0 Level A and AA. For a non-web document the word document replaces web page, and four success criteria do not apply to non-web documents.
Enforcement
Agency complaint procedures and federal procurement.

What the check looks at

The file is read in memory with an open-source PDF parser and then discarded. Six tests run on its structure, each tied to the WCAG success criterion or PDF/UA clause it is closest to. A file over 15 MB, one that is password-protected or one that does not parse as a PDF gets an error message instead of a score.

  • Tags. Whether the file is marked as tagged and has a structure tree. Without one, a screen reader cannot tell a heading from a table from body text, or know the reading order. Critical.
  • Text layer. Whether any of the first five pages has text that can be extracted. If none does, the document is pictures of text, a scan, and reads as silence. Critical.
  • Language. Whether the document declares its language, so a screen reader picks the right voice. Serious, WCAG 3.1.1.
  • Title. Whether the metadata carries a title, and whether the file tells viewers to show that title instead of the file name. Serious when missing, moderate when it is there but not shown. WCAG 2.4.2.
  • Bookmarks. Whether a document longer than ten pages has an outline, so a keyboard user can jump between sections. Moderate, WCAG 2.4.5.
  • Form fields. Whether each field of a fillable form has a description. Without it a screen reader reads out the internal field name or nothing. Serious, WCAG 4.1.2.

What it cannot tell you

A tagged file is not the same as a well tagged file. The check confirms that a structure tree exists. It does not read the tree, so it cannot tell whether a heading is tagged as a heading, whether a table has header cells, or whether the reading order matches the visual order. Those are the faults that remain after a quick fix.

It does not judge meaning. Whether alt text describes the image, whether a link text makes sense on its own, whether the colours have enough contrast: all of that needs a person looking at the document, or a full PDF/UA validation plus a person.

It is not a PDF/UA validation and not a certificate. A clean result means none of the six faults was found. It is a good reason to keep going, not a statement you can publish that the file conforms.

How to read the result

The score starts at 100 and loses points for each finding by its weight: 18 for a critical one, 11 for a serious one, 6 for a moderate one. A finding that repeats, such as ten form fields without a description, costs a little more than a single one. The grade follows the score: A from 90, B from 80, C from 70, D from 55, F below that.

Fix the critical findings first. A file with no tags or no text is unreadable for many users, and nothing else on the list matters until that changes. Serious findings come next, then moderate.

Under the score, a box says how many of the faults found are ones our repair puts right automatically, and how many need a person. That split is computed from the findings, not from a guess about your file. A missing title or language is mechanical. Tagging a scan is not, because there is no text to tag until somebody recognises it.

A worked example

Take a one-page document exported from a design tool: real text, an 18 point heading and a few paragraphs, but no tags, no title and no language. The check returns 60 out of 100, grade D, with three findings: The PDF is not tagged (critical, PDF/UA 7.1), No document language (serious, WCAG 3.1.1) and No document title (serious, WCAG 2.4.2). The repair box says all three are faults the repair puts right automatically.

Now the same page as a scan, with no text layer. It scores 42, grade F, with four findings: not tagged, no extractable text (both critical), no language and no title. Here the box says two of the four are automatic. The other two need a person, because the page has to go through text recognition before there is anything to tag.

A properly tagged file with a title, a language and text scores 100 and shows no findings. That is the result to aim for before the manual checks start.

What the law asks of documents

In the EU, documents on a public body's website are part of the website. Directive (EU) 2016/2102 leaves out only office files published before 23 September 2018 that are not needed for active administrative processes. EN 301 549 treats a downloadable PDF as a non-web document under clause 10 and applies the WCAG criteria to the file. Poland's 2019 act carries the same exclusion date for office files.

Under the European Accessibility Act, the question is the information a covered service gives its customers. In Germany, § 12 BFSGV asks for that information in text formats that can be turned into other accessible formats, with alternatives for non-text content, and perceivable, operable, understandable and robust when digital. When the terms, instructions or account documents come as PDFs, those PDFs are where the requirement lands.

In the United States, Section 508 holds federal agencies' electronic content to WCAG 2.0 Level A and AA. A non-web document is tested with the word document in place of web page, and four success criteria that only make sense for sets of pages do not apply to it. Vendors selling documents or document systems to agencies are asked about this in procurement, often through a VPAT.

Fixing a PDF that fails

Most faults are cheaper to fix in the source than in the PDF. Turn on tagged PDF export in the word processor or layout tool, use real heading styles and real tables, and export again. Set the title and language in the document properties before export. W3C's PDF techniques describe each of these steps: PDF9 for headings, PDF16 for the language, PDF18 for the title and how it is shown.

When the source is lost, or the file comes from a system that produces untagged output by the thousand, the PDF itself has to be repaired. A scan has to be recognised into text first. Alt text for images and descriptions for form fields have to be written by somebody who knows what the image shows and what the field is for.

/ Questions

Questions people ask

Is a PDF accessible if I can select the text?
No. Selectable text means the characters are in the file. Accessibility depends on the tags, the reading order and the alternatives, and selection tells you nothing about those.
Does this check validate against PDF/UA?
No. It runs six structural tests that catch the most common blocking faults: tags, text layer, language, title, bookmarks and form field descriptions. A PDF/UA validation tests far more, and some of it still needs a person.
Can a scanned document be made accessible?
Yes, but it has to be recognised into real text first and then tagged. A scan is an image of a page, and an image cannot be read out.
Is my file stored?
No. The file is read in memory and discarded after the check. If you type an email address, we keep that address with the file name and the score so we can follow up, and nothing else.
Why is a password-protected PDF not checked?
If the file opens without a password, it is checked as normal. If it needs one, the checker cannot read its structure and says so instead of guessing.
Do you repair PDFs?
Yes. The repair is priced by the length of the document, and the result is checked again with the same tests before it comes back. Faults that need a person are named in the result rather than left out.
/ Sources

Where each statement comes from

Checked against the sources on .

This page explains the rules in general terms. It is not legal advice, and it does not tell you whether a rule applies to your organisation. For that, ask a lawyer who practises in the jurisdiction.