Turn Boxes of Paper Into Searchable PDFs With OCR and a Bot

GuideBy Bot Tutorials · Published Nov 5, 2026 · Facts last verified Nov 1, 2026 · Label: Documented, not yet tested by Bottutorials

Next in this series: Tomorrow: Make Sense of Old Cron Jobs and Shell Scripts With a Bot · The day after: Have a Bot Explain Old COBOL Code, Line by Line · Series hub

Scan old letters and records, make them searchable with free OCRmyPDF and Tesseract, then use a bot to build an index you can trust.

Difficulty: Beginner · Bot used: any general chatbot · Time: depends on the pile; plan an evening per few hundred pages What we tested vs. what's documented: Built from the official Tesseract, OCRmyPDF, and federal digitization guidance linked below. Bottutorials hasn't filmed a full archive run yet.

What the old tech is

Paper is the oldest "old tech" there is: filing cabinets of letters, club newsletters, receipts, and typed reports. Scanners and flatbeds have been around for decades. What's changed is that free optical character recognition (OCR) has become good enough to turn page images into searchable text.

Why it still matters

A scan without OCR is just a picture, and you can't search it. Once there's a text layer, a 40-year-old newsletter shows up when you search for a name. That's what revives an archive: people can actually use it.

The free tools

Tesseract is an open-source OCR engine. Its project page says it recognizes more than 100 languages out of the box (github.com/tesseract-ocr/tesseract).
OCRmyPDF adds an OCR text layer to scanned PDFs, using Tesseract underneath, so the page looks the same but becomes searchable (OCRmyPDF docs).
The U.S. federal Digitization Guidelines site collects guidance on scanning and image quality from federal agencies (digitizationguidelines.gov).

What a bot can automate

A folder and file-naming plan (BOX03_FOLDER12_0001.pdf).
The OCRmyPDF command lines for your operating system, plus a loop script that processes a whole folder.
Summaries and keyword lists of each OCR'd document, so you get an index.
A catalog spreadsheet: file, date, people, places, topic, page count.
Explanations of scanner settings (resolution, color vs. grayscale) in plain terms, linked to the guidance.

What it can't do: read a scan it hasn't been given, guarantee OCR accuracy, or decide what's private.

Step by step

1. Prepare the paper

Remove staples and paper clips carefully, flatten pages, and keep everything in its original order. Order is information.

2. Scan consistently

Use the same settings for a whole batch. Ask the bot to explain the trade-offs and point it to the federal guidelines. Save multi-page documents as PDF.

3. Add the text layer

Ask:

I'm on [Windows/macOS/Linux]. Give me install steps for OCRmyPDF from its official documentation, then a script that runs it on every PDF in scans/ and writes results to ocr/ with the same file names. Don't overwrite originals. Use English plus [other language] if available. Explain each flag.

Check each flag against the OCRmyPDF documentation before you run it.

4. Spot-check the OCR

Open a few results and search for a word you can see on the page. Old typewriter text, handwriting, and faded carbon copies are hard. Handwriting is generally where OCR struggles most.

5. Build the index with the bot

Copy the extracted text of a document (or use your bot's file upload, if you're comfortable with it) and prompt:

Here is OCR text from a 1970s club newsletter. OCR errors are likely. Give me: probable date, a 2-sentence summary, people named, places named, and 5 keywords. Put "?" next to anything that might be an OCR error. Don't fill gaps with guesses.

Paste the answers into your catalog.

Worked example

Say page 1 of BOX03_FOLDER12_0001.pdf reads "Annual Picnic — Riverside Park — June 14 — bring a dish." OCR might return "Annua1 Picnlc — Rlverside Park." A careful bot reply would be: "Probable topic: annual picnic; place: Riverside Park (OCR shows 'Rlverside'?); date: June 14, year not stated." If the bot confidently says "1974" when the page has no year, reject it and remind it of your rule. Over a batch, these corrections show you which pages need a person to transcribe them.

Make the bot check itself

Before you trust any answer in this guide, ask one follow-up:

Go back through your last answer. List every factual claim, mark each one VERIFIED (I gave you the source text), DOCUMENTED (you believe the official documentation says this; give the doc page), or GUESS. Rewrite anything marked GUESS as a question for me.

This takes thirty seconds and catches most invented details: made-up command flags, nonexistent options, confident dates. Then open the linked documentation for anything marked DOCUMENTED and confirm it yourself. Bots are good at structure and wording and unreliable at specifics, and this whole series is built around that difference.

Keep a revival log

Write a short log as you work: date, what you did, the prompt you used, what the bot got right, what it got wrong, and how you checked. It's useful for three reasons. You can repeat the job next year without starting over. Someone else can pick up the project. And you'll see which kinds of mistakes your bot tends to make, which makes you faster the next time. If you share your results, include the log. That openness is what makes a revival project trustworthy.

Glossary

OCR: optical character recognition, turning an image of text into text.
Text layer: invisible, searchable text placed under a page image in a PDF.
DPI: dots per inch, the scan resolution.

What to verify

Every date and name in the index against the page image.
That the original scans are kept untouched alongside the OCR'd copies.
That backups open.

Risks, safety and legal

Privacy. Old records can include addresses, medical details, or financial account numbers. Keep those archives private and don't upload them to chatbots.
Copyright. Scanning your own papers for preservation is common practice, but publishing other people's writing, like newsletters and letters, can raise copyright questions. See the Copyright Office's fair use overview.
Handling. Fragile or valuable originals may need a professional conservator.

FAQ

Can a phone camera replace a scanner?

For quick access copies, often yes. For an archive, consistent lighting and flat pages matter.

Does OCR change how the page looks?

OCRmyPDF keeps the page image and adds an invisible text layer.

Should I upload private records to a chatbot?

No. Keep sensitive archives private.

Sources

Video outline (coming)

Preparing paper
Consistent scanning
OCRmyPDF and Tesseract
Spot-checking OCR
Bot-built index
Privacy and copyright

The narrated video for this piece is in production; this outline is what it will cover.