Make a Scanned PDF Searchable
Two settings decide the outcome. Leave Text layout on Automatic for ordinary pages and switch to A single column when the automatic mode tangles the lines; raise Render quality from 2 towards 3 when the print is faint or small. Straighten the scan before you start, and treat every number in the result as unverified until you have compared it with the page image.
Files are never uploadedRun OCR on your scan
The pages are rendered and recognised inside your browser — the document is never uploaded, and the English language data is bundled so recognition works offline.
Open OCR PDFWhat the text layer actually is
A scanned page is a photograph. It looks like writing to you, but to a program it is a grid of coloured pixels with no notion of letters, so searching, copying and selecting all fail. OCR works by analysing those pixels, deciding which groups of them form characters, and writing the recovered words into an invisible layer behind the picture.
Because the picture is left alone, the result still looks exactly like the scan. What changes is that its words can be found. That is also why the output is a rebuild rather than an edit: each page is re-rendered as an image before the layer is written, so vector artwork, hyperlinks and interactive form fields belonging to the source file do not survive the round trip.
Accuracy is a property of the scan, not of the tool. Clean, upright, well-lit print from a scanner or a careful phone photo usually comes out very good. Skew is the single biggest destroyer of accuracy, because a tilted line crosses the scan lines the recogniser is building, and small type compounds the problem by leaving fewer pixels to interpret.
Numbers are where mistakes concentrate, for a purely visual reason: nought and capital O, one and lowercase l, five and capital S are nearly the same shape at scan resolution. A machine reading the page cannot use the context a person would. That is why any figure you intend to act on deserves a glance against the original picture.
Step by step
Confirm the PDF really has no text layer
Try selecting a line in your reader. If the words highlight, the document is already searchable and OCR would only add recognition risk. Use PDF to Text instead, which copies the existing text exactly and finishes instantly.
Fix skew and rotation before anything else
A page that is tilted even slightly misleads the recogniser. Straighten it in the source file — Rotate PDF fixes a page that is simply the wrong way up — because correcting it afterwards means re-running the whole job.
Add the PDF and choose the Text layout
Leave Text layout on Automatic for a normal mixed document. Switch to A single column when a page is one straightforward column and Automatic tangles the lines together, and use A single uniform block for a page that is one solid mass of text with nothing breaking it up.
Raise Render quality for difficult pages
Render quality defaults to 2. If the print is faint, small or the scan is low resolution, push it towards 3: the page is rendered at a higher resolution before recognition, which gives the recogniser more to work with at the cost of time.
Let it read the document
Each page is rendered, analysed and rebuilt in turn, and the progress indicator names the page currently being read. A long scan takes proportionally longer, so start it and leave the tab alone rather than closing the lid.
Verify the parts that matter
Open the searchable PDF, search for a phrase you know is on the page, and select a line to confirm the layer is there. Then compare every figure you care about against the picture, especially account numbers, totals and dates.
Keep the file you need
You get a searchable PDF that looks like the original and a plain text copy of everything recognised. Keep the PDF as the archive copy, and the text file when the words are going somewhere else.
Which setting to reach for
Pick the row that matches the page, then check the result. Recognition quality depends on the scan far more than on these controls.
| The page looks like | Text layout | Render quality | Expect |
|---|---|---|---|
| Ordinary mixed document, headings and paragraphs | Automatic | 2 (default) | A text layer in reading order, with larger text found as headings. |
| One clean column that automatic mode interleaves | A single column | 2 | Lines kept in order rather than tangled. |
| One solid block of text, no columns | A single uniform block | 2 | The page read as a single flow. |
| Faint or very small print | Automatic, with layout and orientation detection | 3 | Better recovery of small characters, at the cost of time. |
| Page scanned upside down | Automatic, with layout and orientation detection | 2 | Orientation detected and corrected before reading. |
| Handwritten notes | Any | Any | Mostly not recognised; do not rely on it. |
Mistakes to avoid
Running OCR on a document that already has text
If the words can already be selected, OCR replaces exact text with a best guess and can only make things worse. Check with PDF to Text first; it is instant and lossless.
Skipping the straightening step
A tilted page is the most common cause of garbled output. Fix the rotation in the file before recognising it, then run OCR once on the corrected version rather than twice on a bad one.
Counting on tables keeping their shape
The picture of the table stays exactly as scanned, but the text layer follows reading order, so searching works while copying a table out rarely preserves the columns. Rebuild the grid by hand, or take the figures from the picture.
Letting one bad page spoil a long document
The whole document is processed, and a long scan ties up the tab for a while. For a large file, extract just the pages you need with Extract PDF Pages first, then recognise the smaller document.
Trusting the numbers
Digits are the weakest part of any OCR result because several shapes are nearly identical at scan resolution. Verify anything you will act on against the page image.
Frequently asked questions
What does a searchable PDF contain?
Which language does the recognition read?
Is the PDF uploaded?
Which Text layout should I choose?
Should I raise Render quality?
Why is OCR slower than PDF to Text?
Can I process just a few pages?
Will the output keep links and form fields from the original?
What about handwritten notes?
Tools used in this guide
Send us feedback
Something broken, or unclear? Tell us and we will look at it.