Try to select a single word. If the whole page highlights as one block, it is an image. If nothing highlights at all, the text may be vector outlines. If it selects but pastes as gibberish, the font encoding is broken.
Only the first case is solved by OCR, and OCR will also solve the second. The third has its own answer.
The cursor turns into a hand instead of a text beam, or it drags a blue box across the whole page. Either way the words on screen are not available as words, and the reason matters because the fixes have nothing in common.
Thirty seconds of diagnosis
Three quick actions identify which of the four causes you have.
Try to select one word
Double-click on a word. If it highlights on its own, real text exists. If the entire page goes blue, or nothing happens, it does not.
Try the document search
Search for a word you can clearly see on the page. A search that finds nothing confirms there is no text layer, or that the text present does not match what is displayed.
Select all and paste into a text editor
This separates the last two causes. Nothing pasted means no text. Nonsense pasted means the text is there but its encoding is broken.
The four causes
It is a scanned image
By far the most common. Every page is a photograph and there is nothing underneath it. The whole page highlights as a single object because the object is one image.
OCR is the fix. It reads the picture and writes an invisible text layer behind it, after which selection and search work normally. Accuracy depends on the quality of the scan rather than on the software.
The text was converted to outlines
Design and prepress software offers an option to convert text to curves or outlines, which replaces every character with a vector shape describing its silhouette. It is done deliberately, usually so a printer does not need the fonts.
The result looks like perfect, infinitely scalable text and contains no characters at all. Zooming in reveals crisp edges at any magnification, which distinguishes it from a scan, where zooming eventually shows pixels. OCR is the only route to text, and because the shapes are geometrically perfect it tends to work very well.
The font encoding is broken
Here the text genuinely exists and selection works. The problem appears when you copy it: the pasted result is gibberish, often accented characters or symbols rather than letters.
A PDF font can use any internal numbering it likes for its glyphs. A separate mapping table tells software which actual character each glyph number represents. When that table is missing or wrong, usually from a subsetted font written by a badly behaved generator, the display is correct because the glyphs are drawn by number, and copying is wrong because nothing can translate those numbers back into characters.
An image is sitting on top
Less common and easy to mistake for the others. The page has real text and also has an image or a filled shape covering it, so clicks land on the overlay rather than on the text. A search will find words that you cannot select, which is the signature of this case.
- Whole page highlights at onceScanned image
- Nothing highlights, edges stay crispText as outlines
- Selects fine, pastes as gibberishBroken encoding
- Search finds it, cursor will notImage on top
- Some pages work, others do notMixed sources, merged
What to do about each
The page has no usable text
- Scanned pages, where the text is a photograph
- Text converted to outlines in design software
- Pages rasterised during a flattening step
- Faxed or photocopied documents
The text exists already
- Broken font encoding, where copying gives gibberish
- An image overlay blocking the cursor
- Permission flags that disable copying
- Text in a form field rather than the page
Living with a broken encoding
This is the awkward one, because the information needed to map glyphs back to characters is genuinely absent from the file. Nothing can reconstruct it directly.
Two practical workarounds exist. Try a different reader first, since some make better guesses than others by matching glyph shapes against known fonts. If that fails, rasterise the page and run OCR on it, which throws away the broken text entirely and recognises the visible shapes instead. It is inelegant and it reliably produces usable text.
Give the page a real text layer
Recognition runs in your browser and writes text behind the image. Nothing is uploaded.
Common questions
Narrowing it down.
01Why does my whole page highlight instead of one word?
Because the page is a single image. Selecting it selects the image object rather than any text, since no text exists on that page. OCR adds a text layer and makes normal selection possible.
02The text looks perfectly sharp but will not select. Why?
It was probably converted to outlines in design software, which replaces each character with a vector shape. The result stays crisp at any zoom and contains no characters. OCR is the only way to recover text, and it usually performs well on shapes this clean.
03I can select the text but it pastes as gibberish. What is wrong?
The font's character mapping table is missing or incorrect, so the glyphs draw correctly by number but cannot be translated back into characters. Try a different reader first, and if that fails, run OCR on the rendered page.
04Search finds a word but I cannot select it. Why?
Something is covering the text, usually an image or a filled shape drawn over the page. The text is present and findable, and your clicks land on the overlay instead.
05Why does selection work on some pages and not others?
The document almost certainly combines pages from different sources, such as scans merged with born-digital pages. Each page behaves according to how it was made, so test one from each section.
06Will OCR fix every case?
No. It fixes pages with no usable text, meaning scans and outlined text. It does not fix permission flags that disable copying, and for a broken encoding it works only by discarding the existing text and re-reading the image.