Scanned PDF to text with OCR

Why you can't search a scanned PDF, and what OCR adds

Open a PDF that came out of a scanner or a phone app and press Ctrl+F (⌘F on a Mac). Type a word you can see on the page. Nothing. Try to drag across a sentence to copy it: the cursor draws a blue rectangle over the whole page instead of highlighting words. That is the sign you are holding an image PDF.

A PDF is a container. A PDF exported from Word or a web browser contains real characters: the file knows that this run of shapes is the word «invoice», which font it uses and where it sits. A scanned PDF contains one picture per page and nothing else. To the computer, the letters on it are no different from a coffee stain. You can read it, but your PDF reader, your search bar, your file indexer and any screen reader cannot.

OCR (optical character recognition) fixes this by looking at the picture, finding the lines of text, recognising each word and then writing those words into the PDF as a text layer. The trick is that the layer is invisible and placed precisely under the printed words. The original scan is left untouched: same pixels, same stamps, same signatures, same handwritten notes in the margin. What changes is what happens when you interact with it:

• Ctrl+F or ⌘F finds words and highlights them on the page.
• Dragging across a paragraph selects the text, and pasting gives you editable words.
• Your operating system can index the content, so a search for a supplier name finds the PDF in your documents folder.
• Screen readers can read the document aloud.

This is why people talk about making a PDF searchable rather than converting it: you keep the document exactly as it was and add a layer of text to it. If you later want a file you can rewrite, that is a separate step (exporting to Word), covered at the end.

A quick way to tell what you have before doing anything: if you can select individual words in the PDF, it already has text and OCR will not add anything. If you can only select the whole page as a block, or nothing at all, it needs OCR.

How to scan a document so OCR reads it correctly

Recognition quality is decided mostly before the OCR runs. A clean scan reads well on any decent engine; a poor one produces errors that no software setting will fix. These are the settings and habits that matter, in order of impact.

1. Resolution: 300 dpi. This is the standard for text. At 150 dpi the strokes of small letters merge, and «rn» starts to look like «m», «1» like «l», «5» like «S». The common advice that more is always better is wrong here: 600 dpi gives files several times larger and, for normal body text, no extra accuracy. Go higher only for very small print, such as footnotes, the small print on contracts or old newspaper columns.
2. Keep the page straight. OCR reads line by line. A page scanned a few degrees off makes lines drift across each other, and a document with curved lines near the spine is worse. Press the page flat on the glass and line it up with the edge. With a phone, shoot from directly above, not at an angle.
3. Avoid the shadow of the spine. Scanning a thick book leaves a dark band by the binding, and the words in that band are the ones that fail. Open the book as flat as it will go and press gently on the spine, or scan one page at a time with the book turned so the binding sits against the edge of the glass.
4. Clean contrast. Dark text on a light background. Greyscale is usually better than colour for plain documents and better than pure black and white, which can break thin strokes or fill in letters on thin paper. If the paper is yellowed, greyscale still separates ink from paper well.
5. No flash, no reflections. With a phone, use daylight from a window rather than the flash, which leaves a bright spot where the text disappears. Glossy paper and laminated cards are the worst for this.
6. One page per image. Two pages side by side in one image confuse the reading order. Scan or photograph each page separately.

Even with a perfect scan, two things are consistently weaker. Small text (below roughly 8 points) loses detail at 300 dpi and produces more misread characters. Tables are recognised word by word, but the structure of rows and columns is the hard part: the words will usually be there, yet copying a table and pasting it into a spreadsheet rarely lands every value in the right cell. Plan to check tables by hand.

If your source is a set of phone photos rather than a PDF, put them together first: you can convert JPG to PDF in the same app, one photo per page, and then run OCR on the result.

One more caution about photos: do not run a scan of text through an AI image enhancer before OCR. Those models are trained to produce plausible detail, and on a blurry letter they can draw a sharper but different letter. The OCR then reads, with confidence, a character that was never on the page. For text, a straight, well-lit rescan always beats software enhancement.

Free PDF OCR on Mac and Windows, without uploading the file

Many free OCR PDF services work by uploading your document to a server, running the recognition there and sending the result back. For a recipe that is fine. For a payslip, a medical report, a signed contract or a scanned passport, it means a copy of the document has been on someone else's machine.

Airpic PDF is a desktop app for Mac and Windows, and its free PDF OCR runs entirely on your computer. The file is opened, recognised and saved locally; nothing is sent to Airpic or anywhere else. It is free and does not ask you to create an account.

The recognition engine depends on your system:

• On a Mac, the app uses the text recognition built into macOS, the same system feature that lets you select text in photos.
• On Windows, it uses an open-source OCR engine with a model for the Latin alphabet, bundled inside the app, so it also works offline.

What this means for languages: the OCR is built for the Latin alphabet, the script used by English, Spanish, French, German, Portuguese, Italian and many other European languages, with their accents and special letters. Documents in other scripts, such as Cyrillic, Greek, Arabic, Hebrew, Chinese, Japanese or Korean, are outside what it is designed for, and you should not rely on it for them. A page that mixes an English letter with a quoted line in another script will have the Latin part recognised and the rest unreliable.

Two other limits worth knowing before you start. Handwriting is not what this kind of OCR is for: neat block capitals may come through, joined-up handwriting generally will not. And a PDF protected with a password must be opened with that password first; the OCR cannot read pages it is not allowed to open.

Because the app works on the document itself, you can combine OCR with the other things you usually need to do with a scan. Scans are heavy files, so after recognising the text you can compress a PDF to send it by email, and the text layer stays searchable.

Recognise text in a scanned PDF and check it: names, numbers and dates

On a clean 300 dpi scan of a typed document, most of the text will be right. The errors that remain are not spread evenly, and they tend to sit in exactly the places you will search for later. Spend your checking time on these:

• Proper names. Surnames, company names, street names and product codes are not ordinary words, so the engine cannot fall back on common spelling. «Kowalczyk» or «Ostrander» are where you will find one wrong letter.
• Numbers. A digit has no context to help it: «8» and «3», «0» and «O», «1» and «l» or «I», «5» and «S» are the classic swaps. In an invoice or a bank statement, a single misread digit changes an amount or an account number.
• Dates. Slashes and dots in dates (12/03/2024, 12.03.2024) are small marks that are sometimes dropped or read as another character. Check the dates you will need to find.
• Text near stamps, signatures and highlighter. Anything that overlaps the printed words makes those words harder to read.
• Columns and tables. Check that the reading order makes sense when you copy a block: text in two columns can come out interleaved.

The fastest check takes thirty seconds. Think of three words you know are in the document: a name, a number and an unusual word. Press Ctrl+F or ⌘F and search for each. If all three are found and highlighted in the right place, the text layer is aligned and the recognition is good. If a name is not found, search for part of it (the first four letters); if that finds it, the rest was misread, and you know the scan needs a closer look.

A second check: select a paragraph, copy it and paste it into a plain text editor. Read it once. Scattered wrong characters in an otherwise correct paragraph are normal on difficult pages. Whole lines of nonsense mean the page was skewed, too dark or too small, and a rescan will do more than any correction.

Keep in mind what the text layer is for. For searching and copying a few lines, a handful of errors does not matter: you still find the document. If you need the text to be exact, such as for a legal quotation or a figure going into a report, check it against the image every time. The image is the original; the text layer is a helpful copy of it.

How to make a PDF searchable in Airpic PDF, step by step

The app's buttons are labelled in Spanish; the English meaning is given in brackets where it is not obvious. Install Airpic PDF for Mac or Windows, then:

1. Open the scanned PDF. Open it from the app; the pages appear as thumbnails. If you can already select individual words in the document, it has text and you can stop here.
2. Fix the page order first if needed. If the scanner fed pages in the wrong order or one page is upside down, sort that out before recognising: you can reorder and rotate PDF pages in the same app. An upside-down page is read badly.
3. Click «OCR». By default the recognition runs on every page. If you only need part of a long document, select those pages in the thumbnail strip first and the OCR will process just those.
4. Confirm and wait. Each page is recognised on your computer. The time depends on the number of pages and on your machine, not on an internet connection.
5. Save. The new PDF looks identical to the scan. Open it in any reader and test it: Ctrl+F or ⌘F and the three-word check from the previous section.
6. Optional: make it lighter. If the file is too big to attach, use the compression in the app on the recognised PDF; the text layer is kept.

If you need to edit the text, export to Word. A searchable PDF is still a PDF: you can find and copy, not rewrite paragraphs. To get an editable document, run the OCR first (the export needs a text layer to work from), then click «Exportar» (Export) and choose «Word (.docx)». The app writes a .docx with the recognised text, paragraphs, tables and images that you can open in Word, LibreOffice or Google Docs. Simple layouts come through well; complex ones with several columns or text boxes may shift, so check page breaks and tables before sending. The same menu also offers plain text (.txt) if you only want the words. More detail on that step is on the convert PDF to Word page.

That is the whole job: scan at 300 dpi with the page straight and well lit, open it in Airpic PDF, run the OCR, check three words and save. To get started, download the app from the OCR PDF on Mac and Windows page.

FAQ

Keep reading