PDF/A-1b, 2b and 3b which one should you actually pick?

A conversion tool asks whether you want PDF/A-1b, 2b or 3b, offers no explanation, and waits. If a portal or a records policy already named a flavour, the choice is made for you and you can stop reading after the next section. If nobody said, the usual instinct is to pick the strictest option on the assumption that stricter means safer. That instinct is wrong, and it is worth understanding why.

The two things people conflate

"PDF/A-2b" is two independent answers glued together, and almost every confusion about the standard comes from reading them as one.

• The number — 1, 2, 3, 4 — is the part of ISO 19005 the file follows: which edition of the standard, and therefore which PDF features are permitted at all.
• The letter — b, u, a — is the conformance level: how much of the document's meaning, beyond its appearance, has to be preserved.

So a question phrased as "PDF/A-1b or PDF/A-2b?" is only ever asking about the number. The b is identical in both — the same promise, made about a file built to a different edition of the rules. Once that is clear, the comparison becomes small and manageable.

What the 'b' actually promises

Level b stands for basic, and its guarantee is precise: the file will reproduce its own visual appearance reliably, on any conforming reader, indefinitely. Fonts travel inside it, colour is defined absolutely, nothing depends on the outside world.

Note what is not in that promise. Nothing about extracting text. Nothing about searching. Nothing about structure, reading order, or a screen reader making sense of the page. A level-b file is an object guaranteed to look right — which is exactly what most archival requirements are after, and exactly why a scanned page passes validation while remaining an unsearchable picture.

What differs between 1b, 2b and 3b

All three produce a self-contained file with embedded fonts, defined colour, no encryption and no scripting. Those parts are constant. What changes is which PDF features are allowed to survive:

Two rows in that table do most of the work. Transparency is the practical difference between 1b and 2b: any soft shadow, glow, or blend mode in your document has to be flattened to satisfy PDF/A-1, which rasterises those regions and can change how they look. Attachments are the only difference between 2b and 3b — in every other respect PDF/A-3 is PDF/A-2.

The size row surprises people who expect the strictest flavour to be the leanest. PDF/A-1 is pinned to PDF 1.4, which predates the compressed object and cross-reference streams that later PDF versions use to shrink a file's internal structure. Add flattened transparency on top and a 1b file is frequently the biggest of the three. If size matters, compress before converting, never after.

Why stricter is not safer

PDF/A-1 is routinely described as the safest choice because it permits the least. It is worth being precise about what "permits the least" means in practice: compliance is achieved by destroying the features that are not permitted. Transparency is not preserved more carefully under PDF/A-1 — it is removed.

PDF/A-1 was written in 2005 against PDF 1.4, a specification from 2001. Its restrictions describe what PDF could do a quarter of a century ago, not what is fragile. Transparency and layers are not less durable than the rest of PDF; they simply had not been standardised when part 1 was drafted. Part 2 revisited them, permitted them, and remains a full archival standard.

There is one honest argument for PDF/A-1b beyond compliance: it is the oldest and most widely implemented part, so an ancient in-house system is fractionally more likely to accept it. If you are submitting to something built in 2008 and cannot ask anyone, that is a reasonable reason to choose it. "It felt stricter" is not.

PDF/A-3 and the attachment question

PDF/A-3 exists for one purpose: to let an archival PDF carry an arbitrary file inside it. Not another PDF/A file, which PDF/A-2 already allows — any file. A spreadsheet, an XML document, a CAD drawing.

The case it was designed for is the hybrid electronic invoice. Formats such as Factur-X and ZUGFeRD are PDF/A-3 files: a human-readable invoice page for the person who has to look at it, with structured invoice XML embedded inside the same file for the accounting system that has to process it. One document, two audiences, no risk of the two versions drifting apart. As structured e-invoicing obligations have spread across Europe, this has become the most common reason anyone converts to 3b at all.

The corresponding weakness is the reason some archives refuse to accept PDF/A-3 at all. Nothing about the container guarantees anything about the payload. You can embed a proprietary file format that no software will open in twenty years, or a spreadsheet whose formulas nobody can evaluate, inside a file that validates perfectly as archival. The PDF is durable; what is inside it may not be.

How to choose in under a minute

1. Were you told a flavour? Use exactly that one, including the letter. A valid PDF/A-2b file is rejected by a system that asked for PDF/A-1b, and nothing else in this article overrides an explicit requirement.
2. Do you need to embed a machine-readable file — invoice XML, a data file that must travel with the document? PDF/A-3b, and only then.
3. Is the destination an old system, or a rule written around PDF/A-1? PDF/A-1b, accepting that transparency will be flattened.
4. Anything else? PDF/A-2b. It is the modern default, preserves the most of your document, and is accepted essentially everywhere PDF/A is accepted.

One caution about changing your mind later. You can convert a finished PDF/A file to another flavour, but the conversions are not symmetrical: going from 2b to 1b flattens transparency, and going back from 1b to 2b does not restore it — the information was discarded at the first conversion. Keep the original document, not just the archival copy, and regenerate from the source when the requirement changes.

The levels you were not offered

You may have seen PDF/A-1a, PDF/A-2u or PDF/A-2a referred to and wondered why converters rarely offer them. The answer is a genuine limit rather than a missing feature.

• Level u requires every character in the file to map to a Unicode value, so text extraction is dependable. Achievable when the source has real text with sound font encodings; impossible to guarantee when it does not.
• Level a requires a tagged document structure: which text is a heading, which cells belong to which table, what order a page should be read in, what language it is in. That is semantic information about the document's meaning.

A converter receives a page of positioned glyphs. It can see that some text is larger and higher on the page; it cannot know that this makes it a heading, or that a two-column layout should be read down and then across. It can guess — and a confident wrong guess about reading order is worse for someone using a screen reader than no structure at all, because it produces plausible nonsense instead of an obvious gap.

Level a therefore has to come from the authoring tool, where the structure still exists: export a tagged PDF from Word, InDesign or LaTeX, or add the tags by hand in a full PDF editor. If an accessibility requirement has landed on documents that were only ever exported flat, that is a document-production problem rather than a conversion setting, and it is better to find out now than after conversion. Where forms and annotations are getting in the way of a clean archival copy, flattening them first at least removes one variable.

And PDF/A-4?

ISO 19005-4 arrived in 2020, rebuilt on PDF 2.0. It also reorganises the naming: the a/b/u levels are retired, replaced by a single base conformance plus PDF/A-4f for files with attachments and PDF/A-4e for engineering documents with 3D content.

It is a sound standard with thin adoption. Filing portals, deposit policies and retention schedules overwhelmingly still name PDF/A-1 or PDF/A-2, and a requirement written in 2019 does not update itself. Our converter targets the three b-level flavours for that reason. If you are ever offered PDF/A-4, check that whoever receives the file has said they accept it — being early is not an advantage in a format whose entire purpose is that somebody else can open it.

Convert to the flavour you need

Conversion takes about half a minute, with no installation and no sign-up, and uploads are permanently deleted from our servers after two hours.

If the whole idea of PDF/A is new — what the standard is for, what conversion changes in a document, and when it is worth doing at all — start with what PDF/A is and when you need it, which covers the ground this article assumes.

Frequently asked questions