A scanned document converted to PDF/A is fully compliant and completely unsearchable. Both things are true at once, and the combination surprises almost everyone who meets it.
You scan a box of contracts, convert them for the archive, and the compliance check passes cleanly. Two years later somebody needs the one mentioning a particular supplier, searches the folder, and gets nothing. The files are valid. They are also, for retrieval purposes, a stack of photographs.
Why image-only PDF compliance is not the same as usable
PDF/A asks whether the document will render faithfully in future. A photograph of a page renders perfectly faithfully — nothing about it depends on a font being available or a colour profile being guessed.
Nothing in any part of the standard has ever required a text layer. So a scan satisfies every rule while containing no words a computer can read. Our own validator reports zero fonts on such a file, which is correct rather than a warning: there are genuinely no fonts, because there is genuinely no text.
This is worth being clear about because the compliance check gives false comfort. It answers a question about durability, not about whether anyone will ever find the document again.
Why OCR before archiving is the only sensible order
Optical character recognition reads the page image and produces text. Placed invisibly behind the picture, it makes the document searchable while leaving the visual record untouched.
The sequence matters, and it only works one way round.
Recognition adds text, and text brings a font that then has to be embedded to satisfy the standard. Run recognition first and the conversion handles that font like any other. Run it second and you are editing a document that has already been archived, which invalidates the declaration and forces you to convert again anyway. Worse, you may end up with a file that still claims to be archival while no longer matching what was archived.
So: scan, recognise, convert, validate. Our PDF OCR tool handles the middle step, and our PDF/A converter the one after it.
Getting 300 DPI scan quality right the first time
Recognition accuracy is decided at the scanner, and you rarely get a second chance because re-scanning a box of paper is nobody's afternoon.
Resolution. 300 DPI is the widely used floor for printed text and a sensible default. Go higher for small print, faint carbon copies or anything handwritten. Below 300, accuracy drops off noticeably.
Straightness. The single biggest factor after resolution. Recognition engines assume horizontal lines of text, and a few degrees of skew degrades results badly. Deskew before recognising.
Contrast. Faint originals benefit from a contrast increase. Do not push it so far that thin strokes break up, which turns readable characters into noise.
Colour. Greyscale is usually the right compromise for text. Pure black and white loses detail on faint documents; full colour bloats the file for no retrieval benefit unless the colour itself is part of the record.
If the resulting files are too large for an upload limit, reduce them with our PDF compressor before archiving rather than after, since compressing an archival file means modifying it.
Building a searchable PDF/A for records that people can use
Four things we would tell any team setting this up.
Decide how accurate the text needs to be. For finding documents in a large collection, a few percent error is tolerable, because you rarely search on the exact word that failed. For anything that will be quoted or relied on, proofread it — recognition errors in names and figures are confidently wrong rather than obviously blank.
Set the language before you start. Recognition engines use a language model, and pointing one at the wrong language produces plausible nonsense. Mixed-language collections need per-document settings.
Do not expect handwriting to work. Standard recognition is built for printed type. Archive handwritten material for the visual record and add a typed transcript alongside it.
Spot-check the output. Open a few finished files and search for a word you know is on page three. It takes a minute and catches a misconfigured batch before it becomes ten thousand documents.
When the scan is already archived
Inherited an archival scan that nobody can search? It is fixable, just not in one step.
Adding a text layer means editing the document, and editing invalidates the archival declaration. So remove the flag first with our PDF/A to PDF tool, run recognition, then convert back to the level you were asked for. The round trip is covered in editing a PDF/A file.
Keep the original archived copy while you do it. If the collection is a formal record, standard practice is to archive the searchable version as a new item rather than overwriting what is already deposited.
The rule to remember
Converting a scanned PDF to PDF/A makes it durable. It does not make it findable, and no part of the standard was ever meant to.
Recognise the text first, archive second, then check the result — searching the finished file for a word you can see on the page is the fastest test there is. For the background on why the format works this way, see what PDF/A actually is, and for choosing a part, which conformance level you need. If you are assembling a submission that mixes typed chapters with scanned appendices, saving a Word document as PDF/A covers the other half of the job.