How to Scan and File Paper Records So You Can Find Them
There is a box. Bank statements, warranties, tax paperwork, a folder from a house purchase, medical letters, the contents of a filing cabinet someone left you. It has to become digital, and you have a weekend.
The instinct is to start feeding paper into a scanner. That is the wrong first move, and it is why most of these projects produce a folder of four hundred files named scan_0001.pdf that nobody ever opens again.
Scanning is the easy part. The work is everything either side of it.
First: decide what you are keeping
Most of a box does not need to exist in any form. Sorting on paper is far faster than sorting on screen, and every page you discard now is a page you never scan, never process, never store, and never search past.
Three piles.
Keep and digitise. Anything you might need to produce, prove, or refer to: contracts, tax records, warranties still in force, medical history, property documents, qualifications.
Keep on paper. A small pile, and it matters. Some documents have value as physical objects — originals with wet signatures or seals, certificates, anything a body may require in original form. A scan of these is a convenience copy, not a replacement.
Discard. Superseded versions, statements for closed accounts past any retention period, expired warranties, duplicates, marketing.
On how long to keep things: check the rules that apply to you. Retention periods for tax and financial records differ by country and by document type, and they change. This guide is about the mechanics of digitising; what you are obliged to retain is a question for your tax authority’s own guidance or your accountant.
Scanner settings: two choices decide everything
300 DPI. Greyscale. That is the answer for text documents, and it is worth understanding why rather than taking it on trust.
Resolution doubles in two directions at once, so doubling DPI quadruples the data. Colour stores three values per pixel where greyscale stores one. Put those together on a letter-size page:
| setting | values per page |
|---|---|
| 300 DPI greyscale | 8.4 million |
| 600 DPI colour | 101 million |
Twelve times the data — and for black text on white paper, essentially all of the extra describes nothing. 300 DPI is above the threshold where text stays crisp and recognition works reliably, and it is what archive and submission standards generally expect.
Go higher only where detail genuinely carries meaning: engineering drawings, small print you may need to enlarge, photographs. Use colour only where colour is information — a highlighted clause, a coloured stamp, a photograph.
Going lower is a false economy. Below 300 DPI, character shapes lose the detail that separates similar letters, and text recognition accuracy drops off quickly. You will not rescan the box.
Scan in batches, not one document at a time
Load the feeder, scan the whole stack as one file, and separate the documents afterwards. Stopping to start a new file for every three-page letter is what makes this take a weekend instead of an evening.
If your scanner has a duplex mode, use it — and expect blank backs on anything printed single-sided. Those come out later in one pass.
The processing order, which is not obvious
Here is the sequence that works. The order is not a preference: several of these steps destroy the output of others if run in the wrong sequence.
1. Straighten. Fix sideways and upside-down pages first. Everything downstream — reading thumbnails, recognition accuracy, cropping — depends on pages being the right way up. How to rotate PDF pages.
2. Remove the blanks. Duplex scanning inserts a blank behind every single-sided sheet, and they fall on predictable numbers. How to remove pages from a PDF.
3. Crop the scanner edges. Black borders and page-edge shadows are noise that recognition will try to interpret as characters. How to crop PDF margins — note that cropping hides rather than deletes, and will not reduce your file size.
4. Compress — before recognition, never after. Compression re-encodes pages as images, which discards any text layer present. Run it after recognition and you throw away the work. Why is my PDF so large covers what compression actually does.
5. Recognise the text. OCR PDF adds an invisible text layer beneath the picture, leaving the page looking identical. This is the step that turns a pile of images into an archive. How OCR makes a scanned PDF searchable.
6. Split into documents. Now that the batch has text, you can find the boundaries instead of eyeballing them — see below.
7. Rename. The step that decides whether any of this was worth doing.
Steps 1 to 5 operate on the whole batch at once. That is the entire reason for scanning in batches: one rotate, one crop, one recognition pass, rather than forty of each.
Finding the boundaries without reading everything
Once the batch is searchable, splitting stops being a reading exercise.
Almost every document type repeats a phrase on its first page: “Statement period”, “Invoice number”, “Dear”, a letterhead’s address line. Search for it and the list of pages it appears on is your list of boundaries. Search PDF reports each match with its page number.
Then split by those ranges — how to split a PDF. If the batch happens to be uniform, say forty pages of ten four-page forms, splitting every four is faster still; just check the arithmetic divides exactly first.
Naming, which is the whole point
An archive you cannot search is a box you cannot open. The filename does most of that work, because it is what both you and your operating system’s search will match against.
A convention that holds up:
YYYY-MM-DD-type-party-detail.pdf
2024-03-15-invoice-acme-roofing.pdf
2019-06-02-warranty-bosch-dishwasher.pdfDate first, in that format, because it sorts chronologically by default in every file browser on earth. 15-03-2024 does not.
Type second, so everything of one kind groups together when sorted by name.
Lower case, hyphens, no spaces, which keeps names portable and avoids the assorted problems spaces cause in scripts, URLs, and older systems.
Be consistent about vocabulary. Pick invoice or bill and never both — search only finds the word you used, and future-you will not remember which one past-you chose.
Folders should be shallow. Deep hierarchies feel organised and defeat search, because you have to remember the path. One folder per year, or per category, is plenty when the filenames carry the information.
Storage: one copy is not an archive
You have just converted a box that survives a dropped laptop into files that do not.
The rule worth following is three copies, two kinds of media, one of them somewhere else. In practice: the working copy on your computer, a copy on an external drive, and a copy in cloud storage or a drive kept at another address.
A single external drive is not a backup — it is a second thing that can fail, and drives fail. Sync services are not backups either, because a deletion or corruption syncs faithfully to every device you own.
Check that you can actually open a file from each copy, at least once. An untested backup is a belief, not a backup.
For anything you need in ten years
For records with a genuinely long life — property, qualifications, anything legal — there is a format designed for the problem: PDF/A, standardised as ISO 19005 — and what PDF/A is and whether you need it explains what conforming to it costs.
Its requirement is that a document must be self-contained: everything needed to render it exactly, fonts included, lives inside the file. That is what stops a document from degrading as the software around it changes. The PDF Association publishes an accessible introduction to what the profile involves.
The reasoning is the same one behind why a PDF looks different on another computer: documents fail when they expect to find something on the far end rather than carrying it.
Do this on your own machine
A box of personal records is the single clearest case for not uploading anything. Tax documents, medical letters, bank statements and identity documents are exactly the material you would not hand to an unfamiliar service in order to rotate a page.
Every step above runs in your browser — the files are read from your disk, processed in memory, and written back. Nothing is transmitted, and the reasoning is set out in how browser-based PDF processing protects your privacy.
It also means your own hardware is the limit. A three-hundred-page batch is real work for a laptop; recognition especially. Process in chunks of thirty or forty pages rather than fighting one enormous job, and leave the tab open while it runs.
What not to bother with
- Anything you can obtain again in thirty seconds. Most banks and utilities keep years of statements available for download, already digital and already searchable. Scanning those is work you are doing on their behalf.
- Manuals. Almost all of them are online as PDFs already.
- Documents whose original you must keep anyway. Scan them for convenience, but the scan changes nothing about storing the paper.
- Photographs. A document scanner set for text handles photographs badly. That is a separate job with different settings.
The shape of a weekend
Sort the box on paper. Scan in batches at 300 DPI greyscale. Straighten, de-blank, crop, compress, recognise — once per batch. Split on searched boundaries, rename to a convention, store three copies.
The scanning is an afternoon. The sorting and the naming are what make it an archive rather than four hundred files called scan_0001.pdf, and they are the parts worth slowing down for.