How to OCR a scanned PDF into text you can actually quote
To OCR a scanned PDF into text you can rely on, recognise it once on purpose, check the result page by page rather than trusting one accuracy figure, re-run only the pages that came out badly, and keep the original scan untouched so every quote can be checked against the page image it came from.
A scanned PDF is a stack of photographs. Until the photographs become characters, nothing can search it, quote it or translate it — and an AI that answers questions about it anyway is answering from nothing.
Why "98% accurate" is the wrong number
OCR tools love a single accuracy figure. For a book you plan to quote from, it is close to useless.
Errors are not spread evenly. A clean scan of a four-hundred-page book can be near-perfect on 390 pages and badly wrong on ten: the page with the skewed photograph, the fold-out table, the chapter where bleed-through from the reverse side came out as text. Averaged over the book, those ten pages vanish into a reassuring percentage. Read individually, they are where your quotation will come from three weeks later — and a sentence that is 70% right does not look like garbage. It looks like a sentence.
So the useful question is not "how accurate was it?" but "which pages should I not trust?"
Step by step
1. Decide which document is worth recognising
Recognition costs time and, with AI-based OCR, money. Running it silently over a whole folder spends both on files you never meant to read. Start with the one book or paper you actually need.
2. Run it in a way that survives interruption
A long scan takes a while. You want progress kept per page, so that closing the laptop at page 230 does not throw away pages 1–229.
3. Check the result page by page
This is the step most workflows skip, and the one that matters. Look for a per-page record and a list of flagged pages. Spot-check the flagged ones against the image. A good check looks at the shape of the result — empty pages, runs of nonsense, text far too short for a full page — so it works whatever language the book is in.
4. Re-run only the bad pages
Fix the ten pages that failed instead of the four hundred that did not.
5. Keep the scan as the scan
Do not overwrite the original with a re-encoded "searchable PDF" unless you have to. Keep the image as the source of truth, so any quote can be checked against the page it came from.
6. Then use it like any other document
Search it, highlight it, ask it questions, translate it. The recognition was the expensive part; from here on it should cost nothing extra to come back to.
How MindMax handles a scan
MindMax recognises a scan as a scan the moment it opens: it shows the page images, and the AI actions say why they are waiting instead of answering. Recognition is a button you press on that document. It runs in batches, keeps completed pages if interrupted, gives every page its own quality record and lists the pages that need review, so you can re-run just those.
The PDF itself is never modified. The recognised text is kept on your computer, matched to the file by its contents rather than its name, so renaming or moving the file loses nothing. It opens as a readable column with page markers; quotes from it open on the page image. From then on the book works like any other — questions with checked quotes, an AI reading pass, search across the folder, whole-book translation — and nothing is recognised twice.
Where it goes wrong
- The scan sets the ceiling. Skewed pages, tight gutters, bleed-through and low-resolution photocopies all cost accuracy. No processing recovers what the scan did not capture.
- Handwriting and marginalia are unreliable. Treat them as hints, not text.
- Dense tables and multi-column layouts are the hardest case: the characters can be right while the reading order is wrong.
- Long books are not instant or free. Recognise the book you need, not the shelf.
More on OCR for scanned PDFs and the step-by-step scanned PDFs guide. The definition is in the glossary under OCR. Download MindMax and try it on the worst scan you own.