"PDF" is not one format. It is a family, and several members of it are ISO standards defining what a conforming file may and may not contain. PDF/A is the one most people meet, usually because a court, a repository, or a records department has insisted on it.
The underlying idea is counterintuitive: PDF/A is defined mostly by what it forbids. It is a subset, not a superset — a standard PDF with the risky parts removed.
Why a standard PDF is a poor archival format
A normal PDF is permitted to depend on things outside itself, and each dependency is a way for the document to change or break over decades.
Fonts need not be embedded. A file can reference a font by name and rely on the viewing machine to supply it. Today it renders correctly; in twenty years, on a system without that font, the reader substitutes something else, metrics change, and the layout shifts.
Colour can be device-dependent. "This much red" means different things on different displays with no profile to anchor it.
Dynamic content is allowed. JavaScript, embedded audio and video, and external links. Codecs become obsolete, scripts stop running, and linked resources disappear.
Encryption is allowed. A password protects a document now and locks it permanently once the password is lost — which, over an archival timescale, is close to certain.
None of this matters for an invoice you will forget in a month. All of it matters for a document that must be readable in 2075.
What PDF/A requires
Everything self-contained. All fonts fully embedded. No external dependencies of any kind.
No JavaScript, audio, video, or executable content. Removed both for longevity and because executable content in an archive is a security liability.
No encryption. An archival file must always be openable.
Device-independent colour, via embedded ICC profiles, so colours mean the same thing on any future device.
Standardised metadata, in XMP, so the file describes itself to future systems.
No unresolvable external links. Internal navigation is fine; dependence on an external resource is not.
The trade is deliberate: you give up interactivity and protection in exchange for a document that stands alone.
The conformance levels
PDF/A has versions — PDF/A-1 through PDF/A-4 — and within them, conformance levels. These matter because a requirement usually names a specific one.
- Level B (Basic) guarantees visual reproduction. The document will look right.
- Level A (Accessible) adds Level B plus a tag structure and defined reading order, so it is machine-readable and usable with a screen reader.
- Level U (PDF/A-2 onward) is Level B plus Unicode mapping, so the text is reliably searchable.
Level A is meaningfully harder to produce because it requires proper tagging, which a scanned document cannot have without OCR and structural work. Most institutional requirements ask for PDF/A-1b or PDF/A-2b, which are achievable from any well-made source document.
PDF/A-2 added JPEG 2000, transparency, and file attachments where the attachments are themselves PDF/A. PDF/A-3 permits any file type to be embedded — the basis of the ZUGFeRD and Factur-X electronic invoicing standards, where a human-readable invoice carries machine-readable XML inside it. PDF/A-4 is built on PDF 2.0.
The other family members
PDF/X — print production. Requires CMYK or spot colour, embedded fonts, and defined trim and bleed boxes. Commercial printers frequently require it.
PDF/UA — universal accessibility. Mandates the tagging, reading order, and alternative text needed for assistive technology. See PDF/UA accessibility standard explained.
PDF/E — engineering documents and CAD.
PDF/VT — variable data printing.
These are largely independent, and a file can conform to more than one — PDF/A-2a plus PDF/UA being a common target for accessible archives.
Which to use
| Situation | Format | |---|---| | Emailing an invoice or report | Standard PDF | | A form to be filled in | Standard PDF | | Court filing or public records submission | PDF/A, usually as specified | | University thesis or research deposit | PDF/A | | Records with a legal retention period | PDF/A | | Sending to a commercial printer | PDF/X | | Published for the public sector | PDF/UA | | Everything else | Standard PDF |
Standard PDF remains right for the overwhelming majority of documents. PDF/A is for the ones that must outlive the software that made them.
Producing one
Word — Save As PDF → Options → ISO 19005-1 compliant (PDF/A).
LibreOffice — File → Export as PDF → tick the PDF/A archive checkbox, with the version selectable.
InDesign — via the PDF/X presets, or through Acrobat's preflight for PDF/A.
Acrobat Pro — Save As Other → Archivable PDF, or use preflight to convert and validate an existing file.
Command line — Ghostscript converts to PDF/A with the appropriate PDFA_def.ps definition file.
Note that Google Docs does not export PDF/A. Its PDF export is a standard PDF, and a document exported from Docs must be converted afterwards if PDF/A is required.
Validate, do not assume
This is the step people skip. A file produced by a "PDF/A" export option is not necessarily conformant — non-embeddable fonts, transparency, and unsupported colour spaces all cause silent failures, and you find out when a submission portal rejects it at the deadline.
veraPDF is the free, open-source industry-standard validator and will tell you exactly which clause a file violates. Acrobat Pro's preflight validates and can often fix problems in place. Run one of them before submitting anything that matters.
One thing worth knowing: converting an existing document to PDF/A does not make it a good archival record. A scan converted to PDF/A-1b is a conformant file full of unsearchable images. If the archive needs to be usable, OCR it first.
For everyday work — signing a contract, filling in a form — a standard PDF and a simple editor are all you need. Save PDF/A for the documents that have to survive you.
