Digitise a paper archive
Goal
Turn scanned boxes of paper into a searchable archive: clean pages, a text layer, documented properties, a way to ask questions across all documents, and, if your archive requires it, a PDF/A version with a verified report.
An office is closing and 200 scanned files are going to storage. Someone must be able to find "the 2019 supplier contract" without opening all of them.
Starting state
- Scanned PDFs (one per document), not password-protected.
- You know each document's language.
- A decision on the target: searchable (this recipe's core), and/or PDF/A (optional last stage).
Chain
Rotation → Deskew → Searchable → Compressor → Extract Metadata → (Converter → PDF/A Validator) → PDF KB
upright level text layer shrink inventory archival standard (optional) ask across them
Step by step
1. Rotation and Deskew
Fix sideways pages with Rotation (Selected pages), then run Deskew on the file. It straightens small tilts, but its output is page images and does not keep text. Why first: recognition quality depends on it.
2. Searchable
Set the Document language, keep Deskew image on, tick Remove background/watermark for stained paper, keep Output format on Searchable PDF, press Transform. If it asks you to split a very long file, use Splitter. See Searchable.
3. Compressor
Medium is the usual choice; use Low for documents that will be printed. Why before archiving: the text layer enlarged the file. See Compressor.
4. Extract Metadata
Upload the files; read the panel or the CSV (extracted_metadata.csv) for title, author, creation date and producer. Use it as an inventory. The app only reads; if the author fields are wrong, the book cannot edit them. See Extract Metadata.
5. PDF/A Validator (only if an archive demands PDF/A)
Press Validate PDF/A on Detect automatically. Expect a file produced by Searchable to be reported not valid: the Validator only checks and never converts. To get a PDF/A, use Converter with a PDF/A target, then validate again. The whole route is explained in Archive-ready. See PDF/A Validator.
6. PDF KB
Upload up to 10 searchable PDFs (100 MB) per upload, create a knowledge base, and ask questions with page citations. See PDF Knowledge Base.
Why this order
Shape the pages, then read them, then shrink, then describe and check. The inventory and the PDF/A check describe the file you will actually store, so they come after anything that rewrites it. PDF KB is a convenience on top of the finished text and does not alter your archive.
Expected result
Searchable, reasonably small PDFs; a metadata CSV listing them; optional PDF/A versions with a clean report; and a knowledge base you can question.
Check it worked
- Ctrl+F finds a word you can see in three random files.
- Search for a known name returns a marked-up copy (no result on a file that has it means a recognition error).
- The CSV has a line per file and no failures in the summary.
- Validator verdict and counts (if used); the JSON report is saved with the file.
- PDF KB statuses are "Done"; none shows "No text (scanned?)".
Variations
- Extract the text in bulk: Extract Text after Searchable gives one .txt per file plus a combined file.
- Keep both files: keep the PDF/A as the master and the searchable PDF for daily use.
- Multi-language archive: run Searchable per language group.
- Repeat for hundreds of files: see Make it repeatable.
What can go wrong
- PDF KB retention is about a day: download the wiki .zip, since expiry deletes the knowledge base (rebuild from the PDFs).
- Recognition errors in names and numbers: spot-check; AI correction can also change words.
- PDF/A "not valid" after Searchable: normal; convert and re-validate.
- The Validator cannot fix: it reports clause, page and object only.
- Scanned originals are not archived by these apps; keep them.