Check a PDF against WCAG 2.1 AA and PDF/UA — tags, reading order, alt text, contrast and document language. Runs entirely in your browser; the file is never uploaded.
or drop a file here
It reports whether the document is tagged, whether every mark on the page is either tagged or marked decorative, whether it declares a language and a real title, whether each image carries alternative text, whether tables have header markup, whether the structure tree announces the page in the order it actually reads, whether any page is an un-OCR'd scan, whether text meets WCAG contrast ratios, and whether form fields and annotations have descriptions. Each finding cites the WCAG success criterion and PDF/UA clause it relates to. On a document that is longer than the preview limit, the pages it did not examine are reported as unchecked rather than counted as passing.
No. Parsing, rendering and every check run inside your browser tab using our own PDF engine, so the file never leaves your device. That matters for exactly the organisations facing accessibility deadlines — government, healthcare and legal teams — who often cannot send documents to a cloud service at all. The one exception is a password-protected PDF, which needs a server step to unlock before it can be read.
No, and any tool claiming otherwise is misleading you. Software can prove a document fails — a missing tag tree is a fact. It cannot prove a document passes, because whether alt text is meaningful, or a heading level is right, needs human judgement. This checker lists those items separately as manual checks rather than quietly counting them as passes.
Almost every PDF produced by a normal Save as PDF is untagged: it has no structure tree, so assistive technology sees one undifferentiated block of text with no headings, lists, tables or reading order. It looks perfect on screen and is close to unusable with a screen reader. That gap is the entire reason accessibility remediation exists as an industry.
Both are legal requirements that documents published by covered organisations be accessible. The European Accessibility Act has applied since June 2025. The US Department of Justice's ADA Title II rule was amended in April 2026 and now sets compliance dates of 26 April 2027 for public entities serving 50,000 people or more, and 26 April 2028 for smaller entities and special district governments. PDFs are in scope, which is why document accessibility moved from good practice to a deadline.
Yes, for the ones a machine can fix. It writes a real structure tree — headings, paragraphs, lists, tables and figures, in column-corrected reading order rather than draw order — plus the document language, title and PDF/UA metadata, and it fills in descriptions for links that have none. It then re-renders the result and shows you what fraction of pixels changed, which on a normal document is zero. Our remediation is developed against veraPDF, the ISO 14289-1 reference validator, and every change is checked against a conformance corpus before it ships — but your own file is remediated in your browser, so run it through veraPDF or Acrobat yourself if you need that on record.
From you. After the check, every image on the page is listed with a thumbnail and a box to describe it, or a checkbox to mark it decorative so it is written as an artifact and skipped by screen readers. Nothing is generated: a description invented by software would satisfy a validator and tell a blind reader nothing. Images you leave blank stay reported as failures rather than being quietly filled in.
Yes — a PDF report and a CSV of the findings, both generated in your browser and available as soon as the check finishes, whether or not you run the fix. The report lists every finding with the WCAG success criterion and PDF/UA clause it comes from, the checks that passed, what was changed, and a SHA-256 of the file so anyone can repeat the check and compare. It states plainly that it is evidence of what was checked, not a certificate of conformance. The report is itself a tagged PDF, generated through the same remediation, and the tool tells you whether that succeeded for your report rather than assuming it.
Yes, and it says so rather than guessing. A document whose fonts were never embedded, or whose glyphs carry no Unicode mapping, cannot be made conformant by any remediation tool — the information simply is not in the file. Text drawn inside form XObjects, un-OCR'd scans and a document that already has a structure tree are all left alone and reported: rewriting someone else's tag tree risks producing two structures that disagree, which is worse than one honest gap.
