What PDF/A is, and when somebody will ask you for one
Last updated: September 22, 2026
You are filling in a form for a public body, or sending a document to a court, or depositing a thesis, and the instructions say the file has to be in PDF/A. The PDF on your desk looks fine, opens fine, and prints fine. It is still, almost certainly, not a PDF/A.
What the format actually is
PDF/A is a PDF with promises attached. The standard is ISO 19005, and every rule in it comes from one idea: the file has to be readable in twenty years by somebody who has nothing but the file.
- Every font travels inside it. An ordinary PDF may just name a font — Helvetica, Verdana — and leave it to your computer to have one. On a machine that does not, the reader substitutes something else and the layout moves.
- The colours say what they mean. An embedded colour profile turns "this much red" into a colour that can be reproduced, instead of whatever the screen of the day decides.
- Nothing reaches outside the file. No JavaScript, no actions that run when it opens, no links to files that have to exist somewhere else, no encryption. An archive cannot keep a document whose content depends on a password somebody will have forgotten.
- It says what it is. A metadata packet inside the file declares which level of the standard it claims.
Nothing there is about how the document looks. A PDF/A is not smaller, or prettier, or more official: it is the same document with its dependencies packed inside.
Who asks for it
- Public administration. In Spain, the technical standard for electronic documents lists PDF/A among the formats accepted for filing and archiving, and you meet it in tender portals, grant applications and professional bodies' submissions.
- Courts. In Germany, electronic filing (ERVV) requires PDF/A outright; in the United States many federal courts require or recommend it for documents that go into the record.
- Universities and repositories, for theses and anything meant to survive a change of software.
- Archives and libraries, which is where the format came from in the first place.
- Electronic invoicing. Factur-X and ZUGFeRD invoices are a PDF/A-3 with the invoice XML carried inside — that is exactly why part 3 exists.
Which one to send
Two numbers describe a PDF/A file, and people confuse them constantly. The part says which PDF it is built on:
- PDF/A-1 — the oldest and strictest. No transparency, no layers.
- PDF/A-2 — what most conversions produce today: transparency, layers and JPEG 2000 allowed.
- PDF/A-3 — like 2, but it may carry other files inside it. This is the one invoices use.
The level says how much it promises:
- B (basic) — it will look the same. This is what is asked for in practice.
- U — as B, plus the text can be extracted as proper Unicode.
- A (accessible) — as U, plus the document is tagged for screen readers.
So PDF/A-2B unless you have been told otherwise, and you cannot honestly turn an untagged document into level A by converting it: the tags have to come from whoever wrote the document.
Converting one
PDF to PDF/A writes PDF/A-2B in your browser. Before it converts anything it tells you which of two things is going to happen, because that is the file's decision and not the tool's:
- Its pages are kept. Every font it uses already travels inside it, so nothing is re-encoded and the text stays selectable. Scans are the easiest case of all: there are no fonts to be missing.
- Its pages are redrawn as images. This happens when the file only names a font, or when it paints in print inks (CMYK) that a screen colour profile cannot describe. The document looks identical and the text stops being selectable — which is the honest outcome, because there is no way to embed a font nobody has.
Either way, what PDF/A forbids comes out: JavaScript, actions on opening, attached files and any encryption. Annotations stay and are marked printable.
One order matters: convert first, sign afterwards. A digital signature covers a specific sequence of bytes, and converting rewrites them, so a file signed before conversion arrives with a broken signature. If you need both, convert, then sign with your certificate.
Checking a file
A file that says PDF/A is not necessarily one. Plenty of converters write the metadata that declares the format and change nothing else, which produces a document that claims PDF/A-2B and fails the first validator it meets.
Check PDF/A reads a file in your browser and reports what matters in practice: whether it declares the format, whether it carries a colour profile, whether every font is inside it, whether it is encrypted, whether anything runs when it opens. It is not a validator, and it says so.
The validator is veraPDF: free, open source, maintained by the PDF Association with the Open Preservation Foundation, and the reference implementation that national archives use. There is a desktop version and a public web demo. There are commercial online validators too, but read their privacy notes before you use one — at least one of the popular ones keeps the files you upload to the free tier. For anything with personal data in it, run the desktop tool, or use the check on this site, which never sends the file anywhere.
And a last practical point: when two validators disagree, the one that matters is the one belonging to whoever asked you for the file.
Why files get rejected
- A font is only named, not embedded. By far the most common reason, and the one people never expect, because the document opens perfectly on their own machine.
- It declares PDF/A and nothing else was done. The claim is not the thing.
- The file is encrypted, often because somebody added a password to stop editing.
- It carries attachments and is being sent as PDF/A-1 or 2, which do not allow them.
- It paints in print colours with no print profile inside to explain them — typical of files that came from a designer.
- The signature was applied first and the conversion broke it.
Nothing is uploaded
Both the converter and the check run on your own machine, inside the browser tab. That matters more here than elsewhere: a document you are filing with a public body is exactly the kind of file that should not be handed to a website. Load the page, disconnect from the internet, and they both still work — which is the simplest proof there is, and it takes half a minute. There is more about how to verify it on the page that explains it.