Is a phone camera good enough?

For ordinary print, yes, if you get the lighting and angle right: even light, no shadow across the page, camera parallel to the sheet. A flatbed scanner is more consistent and much faster per page for large volumes.

One PDF per document or one big PDF?

One PDF per document, almost always. A single enormous file is slow to open, awkward to share and impossible to organise later. Group by document, not by scanning session.

From a stack of paper to a searchable archive

Scan at 300 DPI in greyscale, fix orientation and margins first, recognise once, then name files so you can find them without opening them.

Digitising paper goes wrong in predictable ways: files nobody can find, scans nobody can search, and a folder of scan001.pdf through scan247.pdf. The work is mostly in the decisions, not the tools.

Decide what you are keeping first

Sort the paper before scanning anything. Most stacks contain a lot that does not need to exist digitally, and scanning it costs time now and clutters search later.

Sort into: keep and digitise, keep as paper only (originals with legal weight, which you scan and retain), and discard. Do this at the table with the paper in front of you — it is far faster than doing it on screen afterwards.

Capture

Resolution: 300 DPI. The right compromise for ordinary print. Go to 400–600 only for small print or documents with fine detail you may need to read closely.

Colour mode: greyscale. Colour roughly triples file size without helping recognition. The scanner's black-and-white mode makes an irreversible thresholding decision in hardware and destroys faint text. Greyscale keeps your options open. Use colour only where colour carries meaning — a chart, a coloured stamp, a photograph.

Both sides. If you are handling the paper anyway, duplex is nearly free. A missing reverse side discovered later means handling the whole stack again.

Straight and flat. Align pages in the feeder. If you are photographing pages, hold the phone parallel to the sheet rather than at an angle, and make sure no shadow — including your own — falls across it. Perspective distortion cannot be corrected the way simple rotation can.

Fix the pages before recognising

Do this in this order; each step makes the next better.

Rotation. Sideways or upside-down pages ruin recognition. Fix with Rotate PDF.
Cropping. Cut away scanner borders, black edges from photographing, and damaged margins with Crop PDF. Less noise means fewer spurious characters.
Grouping. If the pages of one document are spread across several files, combine them with Merge PDF — see combining scanned documents for the ordering traps.
Removing blanks. Duplex scanning produces blank reverse sides. Strip them out.

Then recognise, once, with OCR PDF. Doing it after the fixes rather than before means you recognise the cleaned-up page, and you never have to redo it.

Naming, which matters more than anything else

An archive you cannot search by filename is an archive you will not use. Adopt one convention and never deviate.

A pattern that works:

YYYY-MM-DD_party_type_reference
2026-03-14_acme_invoice_INV-4417.pdf
2024-11-02_hmrc_correspondence_selfassessment.pdf

The leading ISO date sorts chronologically by default in every file manager. The party comes next because it is what you usually remember. The type narrows it. The reference lets you match against another system.

Three rules: use the date on the document, not the date you scanned it; never use spaces or accents, which break across systems; and never abbreviate inconsistently — pick invoice or inv and stay with it.

Folders and search

Keep the folder structure shallow. Two levels — year, then category — is usually enough, because with good filenames and a searchable text layer you will find things by searching, not by browsing.

Deep hierarchies feel organised and then quietly fail: the moment a document belongs in two places, the structure starts lying to you.

Storage

Have more than one copy. A digitisation project is exactly the situation where a single drive failure erases weeks of work and the paper is already gone.

The conventional rule: three copies, on two different kinds of media, one of them off-site. For a personal archive that can be as simple as the working copy, an external drive, and an encrypted cloud backup.

Test a restore once. An untested backup is a hypothesis.

Size

Only compress if you need to. A 300 DPI greyscale page is typically 100–300 KB, so a hundred-page document is manageable. If a specific file needs to be emailed, compress that copy and keep the original — do not compress the whole archive to save space on a drive that costs less than the time you are spending.

Realistic throughput

With a duplex feeder scanner, a few hundred pages an hour is achievable, and the bottleneck is preparing the paper — removing staples, unfolding, sorting — not scanning. With a phone, count on a couple of hundred pages in a session before it stops being tolerable.

Recognition in the browser runs at roughly one to four seconds per page. For a large archive, run it in batches of a few dozen documents rather than one enormous job.

Privacy

Personal archives are full of exactly the documents you would least like to upload: medical letters, bank statements, tax correspondence, identity documents. Everything in this workflow runs locally in your browser here — the background on how that works explains the mechanism and its limits.