How to turn a pile of scans into documents you can actually find
Three things stand between a heap of scans and something you can search: the words have to be extracted from the images, each document has to carry enough about itself to be found later, and the whole thing has to survive people adding to it in a hurry. Software handles the first, and habits decide the other two.
What actually makes a document findable?
Three separate things, and most piles have only the first. Worth naming them, because a pile that fails at step two feels identical to one that fails at step three, and the fixes are different.
The words exist somewhere
A photograph of a letter contains no text until something reads it. OCR extracts the words and stores them with the file, which is what makes searching inside a document possible at all.The document knows what it is
Full text alone gets you far and then stalls, because half of what you search for is not written on the page. “The 2023 insurance renewal” asks about a year, a category and a household member, and a scan says none of those things out loud.Adding to it takes seconds
this is where systems dieAny method that asks for a filename, a folder and two tags at the moment of filing will be abandoned within a month, and then the pile resumes growing somewhere else — a camera roll, a downloads folder, a chat thread with yourself.
How do you scan so OCR has a chance?
Most disappointing search results trace back to the capture, not the software. A few habits do most of the work:
| Habit | Why it matters |
|---|---|
| Flatten the page, square to the camera | Skew and curl are what turn clean print into approximate letters |
| Even light, no shadow from your own hands | A shadow across a column reads as a smear rather than words |
| One document per file | Four unrelated letters in one PDF can only ever be filed as one thing |
| Capture every page, including the dull ones | The terms and the account number usually live on the page people skip |
| Photograph thermal receipts early | They fade to blank in a drawer, and nothing recovers them afterwards |
| Keep the original file, not just a compressed copy | Heavy compression removes exactly the fine detail OCR needs |
What is worth naming, and what is worth tagging?
A naming scheme is a promise made by the person setting it up on behalf of everyone who will ever add a file, including themselves at eleven at night. Keep the promise small:
Name by hand the documents you would go looking for by name — the passports, the lease, the will, the vehicle registration. That is usually a couple of dozen files, and a plain 2024-lease-signed.pdf beats any scheme with abbreviations in it.
Everything else — the receipts, the statements, the letters, the school forms — should be findable by what is written in it and what kind of thing it is, which is a job for OCR plus a category rather than for typing. If categories are assigned by hand, keep the list short enough to hold in your head. Twelve categories get used, forty get avoided.
How do you test it?
A pile feels organised right after you organise it, which is the worst possible moment to judge. Try this instead, a week later:
Pick five old documents
Choose them from memory, not by browsing — browsing is how you find the ones that happen to be easy to find.Search for something written inside each one
Searching by file name only tests whether you remember your own naming scheme. Searching by content tests whether the words came out of the images.Have someone else in the household try two of them
this is the one that failsAn archive one person can navigate is a personal skill. An archive two people can navigate is a system. If your partner cannot find the insurance policy, the pile is not searchable, whatever it does for you.
Where does software fit?
Self-hosted document managers do this well and are a genuinely good answer if you enjoy running a server; a scanner that drops files into a watched folder, OCR on ingest, tags you maintain. The cost is that it stays your job — the backups, the upgrades, and the tagging discipline.
Sparky Fetch is the same idea with the maintenance taken off your hands. Photos and PDFs go in by web upload, by forwarding them to the Telegram bot, or through a synced Google Drive folder; each one gets OCR and a category without anyone naming it; and everyone in the household searches the same archive, or simply asks. It is in private, invite-only beta, so the honest thing to offer is a look: the demo is a fictional family's paperwork you can search without an account.
Before committing to anything, the questions in choosing a document tool are worth an afternoon.
Frequently asked questions
Why can't I find anything in my scans even though I can read them?
A scan is a picture of a page. Your eyes read it; the search box does not, because as far as the computer is concerned the file contains pixels and no words. OCR is the step that reads the picture and stores the words alongside it. Until something does that, the only searchable thing about the file is its name.
What is a searchable PDF?
A PDF that carries both the scanned image and an invisible layer of the text OCR found, positioned over the matching words. It looks identical when you open it, and it answers to a search. Most scanner apps and desktop scanning software can produce one; the setting is usually called searchable PDF or OCR.
Is OCR accurate enough to rely on?
For printed pages that are flat, in focus and right side up, it is generally good. It degrades on handwriting, on faint thermal receipts, on tight columns, and on photos taken at an angle in bad light. This is why the phone-camera habits below matter more than the software choice.
Should I name files carefully or use tags?
Name the few things you would go hunting for by name — a passport, a lease, a will. Tag the rest, or let something read them for you. Naming does not scale: a naming scheme has to be remembered by everyone in the household and applied at the moment someone is standing in a hallway photographing a letter, which is exactly when nobody is thinking about naming schemes.
How do I know whether my pile is genuinely searchable?
Pick five documents you have not touched in a year and try to find each one in under thirty seconds, searching for something written inside it rather than its file name — a policy number, a landlord's surname, an amount. If you can, the pile is searchable. If you find yourself opening folders to look, it is a pile with a table of contents you have memorised, and the memory is the fragile part.
Nose to the ground. That is the job. ♡
This guide describes a method rather than any particular product, so it makes no dated claims about other companies' software — check your own scanner app's settings for what it calls OCR. Details about Sparky Fetch describe current behaviour during private beta.