A PDF page looks the same whether it contains a paragraph of text or a photograph of that paragraph. Your eyes cannot tell the difference. The browser can, and that is why selecting a line sometimes highlights the whole page, sometimes produces a column of gibberish, and sometimes does nothing at all.
There are four different situations hiding behind "the text will not select", and they need different fixes. One of them has no fix short of OCR.
What can be inside a PDF page
- Real text. Each character is stored with a font and a position. Selection, search and copy all work. This is what a Word export or a LaTeX document produces.
- An image of text. A scan, a phone photo, or a screenshot pasted into a page. There are no characters, only pixels. Selection selects the image.
- Outlined text. Real text converted to vector shapes, usually by a design tool on export. It looks perfect and prints perfectly, but the letters are drawings, not characters.
- Text with broken encoding. The characters exist, but the font's mapping table is missing or wrong. Copying gives you cid:128, boxes, or random accents.
Tell which one you have in ten seconds
- Press Ctrl+F (Cmd+F on a Mac) and search for a word you can clearly see on the page. No results means there is no searchable text — either an image or outlined glyphs.
- Try to select one line with the mouse. If the cursor is a hand instead of an I-beam, the page is an image.
- Copy a sentence and paste it into a plain text editor. Gibberish means broken encoding; nothing means no text layer.
- For a definitive answer, run the file through PDF to Text. If the output is empty or nearly empty, there is no text to extract.
| Symptom | What it means | What to do |
|---|---|---|
| Cursor is a hand, whole page highlights | Scanned image | OCR the page |
| Search finds nothing, but text looks sharp | Outlined glyphs | OCR at 300 DPI |
| Copy gives cid:### or boxes | Broken font encoding | OCR, or re-export from the source |
| Selection works, copy is blocked | Permissions set by the author | Only unlock documents you own |
| Some pages work, others do not | Mixed document | OCR only the image pages |
The fix for scans: OCR
Optical character recognition looks at the pixels, guesses the letters, and gives you a text file. It is the only way to recover words from a page that has none.
- Check the resolution first. OCR needs roughly 200–300 DPI to be reliable. A scan at 150 DPI will read, with more errors; a phone photo of a screen will read badly no matter what.
- Improve the image before reading it. In Scanned PDF to text, the enhance options apply grayscale and contrast, which helps faint or uneven scans. The sharpness setting controls the render resolution — use Balanced for normal documents and Sharp for small print.
- Pick the right language. An English engine reading Urdu text produces nonsense, and the reverse is also true. For mixed documents, choose the combined pack.
- Check a sample before trusting the whole file. OCR errors cluster around numbers, names and tables — exactly the parts that matter. Read the first page of output before you rely on it.
One honest limitation: OCR here produces a .txt file, not a searchable PDF. Making the PDF itself searchable means writing an invisible text layer back onto the page, which is a different job — desktop tools like Acrobat, ABBYY FineReader and the command-line ocrmypdf do that. If you only need the words, the text file is faster and works everywhere.
Getting better OCR results
- 300 DPI, not 600. Beyond about 300 DPI the file gets huge and accuracy stops improving.
- Grayscale beats colour for black text on white paper. Colour adds noise the recogniser has to ignore.
- Straighten the page. A scan tilted by more than a couple of degrees loses accuracy at the line ends.
- Crop the margins. Scanner edges, punch holes and shadows are read as punctuation.
- One column at a time. Multi-column layouts are read across the gutter and come out interleaved.
- Do not expect handwriting. Modern engines have improved, but cursive remains a coin toss. Typed text is the reliable case.
Why it matters beyond copy-paste
A PDF with no text layer cannot be searched, cannot be read by a screen reader, cannot be translated automatically, and cannot be indexed by a search engine. If you are publishing a document — a policy, a menu, a report — exporting it as an image-only PDF makes it invisible to a large part of your audience. If you are submitting one, expect the recipient's system to treat it as a picture: no keyword checks, no automatic form filling, no copy into a database.
Keep the original file either way. OCR output is a derivative, and every derivative loses something the original had.