The PDF checker: what it checks and what it does not
Our PDF checker has had its own site since 7 September 2026: pdf-toegankelijk.nl. You can check a PDF there for free, without an account. The result page is in Dutch for now. Repairing sits in the same tool and is still in closed testing. Your document goes to our own server in Germany and leaves it as soon as the check is done.
Want to know whether your PDF has to be accessible in the Netherlands, or have an auditor look at it? Get in touch.
What the check shows you
How many errors are in your document, how many points need a human to look at them, and how many suggestions come with it. Below that a count per topic: tags, title, language, headings, images, tables. And the three that weigh heaviest, with the success criterion and a link to the explanation in our knowledge base.
If your document is a scan without a text layer, the checker says so. A document like that is empty for anyone using a screen reader, however well the rest is arranged, and the file itself does not show you that.
Up to 10 MB and 20 pages, five documents a day. The 20 pages is a limit of the check itself: above that it no longer reads the whole text layer, and the result would say less than it appears to say.
What we keep
From the check itself we keep nothing. No row about you or your document goes into our database, we do not keep the name of your file, and we delete the file itself as soon as the check is done. The result stays for half an hour at an address only you have, and then it goes.
Two things do exist, and both are in the privacy statement. We count how many checks come from your internet address, because the limit of five a day hangs on that. We do not store the address to do it: we compute an irreversible fingerprint of it, using a secret number that changes every day, and that counter expires after 24 hours. Our web server also keeps a log holding your internet address and the time, as every web server does. We keep that log for 14 days. What happens exactly is in the privacy statement of pdf-toegankelijk.nl, which is in Dutch.
Nothing about you goes to a US server when you use the tool. The pages of the tool load no script, no stylesheet and no font from another company. So your browser opens no connection to Google or to anyone else to display them, and your internet address does not reach them either. The server of the tool does not allow it either: it tells your browser to load files from our own server only. Two tests in the code of the tool check four pages and fail as soon as a script or a stylesheet from another company appears in one of them.
The step after the check
Repairing sits in the same tool and is in closed testing at the moment, with access arranged per organisation.
You upload a PDF and you get your own document back, with a tag structure where there was none and with the title, the language and the matching viewer settings written in. Same pages, same layout. We do not build a second document that looks different.
For every document we compare the pages before and after the repair as images. If anything changed, we tell you which page and how large the difference is.
The full list comes with it: per element what is wrong, which page it is on, and the wording you can lift into an audit report. That is the same arrangement we use in our audit reports, so everything wrong with one element sits together in one place. That list downloads as CSV and as JSON. The free check gives you the totals and not that list.
What the tool does not do
A repaired code layer is not an accessible document. Whether the reading order is right, whether a table header sits in the right place, whether a description covers the image it belongs to: no tool can establish that. So we do not issue a statement that your document meets the standard. If you work for a Dutch public body, that standard is EN 301 549. The decree that names it is the Besluit digitale toegankelijkheid overheid (BDTO), the Dutch implementation of the European Web Accessibility Directive. Your PDF counts as a downloadable document under it. If you fall under the European Accessibility Act, it depends on the service your document belongs to. What we do deliver is a list of the problems that are still there, and a person can work from that.
In a document that has tags we write the PDF/UA identifier into the metadata. That is the line in which your file states it was made according to PDF/UA-1. We write it because checkers such as PAC and veraPDF report a missing one as an error. It says nothing about the content of your document: whether that really meets PDF/UA follows from the check and not from this line.
Where your document stays
Every step runs on our own server in the EU, at Hetzner in Falkenstein, Germany. Your document does not go to Adobe, to Google, to a language model or to any other supplier. We tested that by running the repair with the network connection closed. Hetzner makes a copy of the whole server every night, and that copy stays in Germany.
After a check your file is gone as soon as the check is done. If you have a document repaired, we delete the repaired file 7 days after your last action in it. If you are done sooner, the “ik ben klaar” button at the bottom of your work list throws it away straight away. After that it can still be in a nightly copy, and that copy is itself wiped after 7 days.
One thing does stay, and only when you repair. Of every repair we keep a record: your email address, the time, the number of pages, and which findings were on your document before and after. It also holds the number of headings, tables and images. Your document itself is not in there. We keep that record for 12 months.
Do you need an answer now?
If you need to know whether a document is ready to publish, an audit is the answer and a tool is not. An auditor judges the reading order, the tables and the descriptions of images. No tool can do that, and better tools will not move that line.