A PDF that looks fine on screen can be unusable for someone who relies on a screen reader. If the language is not declared, the title is missing, the file is not tagged or the images have no description, assistive software has little to work with. AuraDoc checks for exactly these gaps and can repair them for you, producing a corrected copy while your original stays untouched.
What the check looks at
The accessibility check opens the PDF and looks for five things: whether the document language is declared, whether it has a title, whether it is marked as a tagged PDF, whether it has a structure tree, and whether every image carries alternative text. Each problem found is listed with its rule and a severity, and images are reported with their page and name so you know where to look. The report also gives a score from 0 to 100, which starts at 100 and drops by 10 points for each issue found. It is a quick way to see whether a file is in good shape or needs work.
A separate answer on PDF/UA
The score is a practical guide, not a certificate. When the PDF/UA validator is available in the environment, the check also runs a formal PDF/UA-1 validation and reports whether the file is compliant, with the list of failed clauses when it is not. If the validator is not available, the report says the compliance result is unknown rather than guessing. Keeping the two results apart means a high score will never be mistaken for a formal pass.
What the repair fixes
Repair runs as a background job, so you can keep working while it finishes. It declares the document language when it is missing, sets a title if there is none, marks the file as tagged, and builds a basic structure tree from the text it can read. It also asks viewers to display the document title instead of the file name, adds basic metadata, and writes the PDF/UA identification only when the formal validation passes. When the document has no readable text yet, the structure tree is reported as remaining work, which is why running OCR on scanned files first gives better results.
Alt text for every image
Images without a description are the most common barrier, so the repair goes through them one by one. An image that already has alternative text is left alone. For the others, AuraDoc writes a short description. When an AI vision provider is configured and your organization has agreed to third-party AI processing, the image is described by the model, with a short, plain instruction to keep the text concise. Without a provider or without that consent, the repair does not send anything out: it writes a simple placeholder that mentions the page, so no image is left without any text. You can then refine these descriptions where it matters. Read more about how consent is handled in privacy and compliance.
Report, fixed items and what remains
When the repair finishes you get a report with the new score, the list of items that were fixed, and the list of things that remain, such as tables that could only be handled on a best-effort basis or PDF/UA clauses that still fail. That honesty matters: you see what was changed and what still needs a human eye. The corrected PDF can be downloaded from the report, and it is a new file, so the original is never overwritten.
Where to find it
On the document page, the accessibility panel has two buttons: one to check and one to repair. It becomes available once the document has finished processing, and it shows the latest check and the latest repair, so you can come back later and find them. The same operations are available through the API for documents you manage in bulk, and they need permission to update the document, while reading the reports only needs read permission. If your work involves public-sector, education or customer-facing documents, make the check part of your routine alongside the other PDF tools and pipelines. To try it, create an account, upload a PDF and run the check.