PDF Text Layer Explained: How Searchable Scans Actually Work
Two PDFs can look identical on screen and be entirely different documents underneath. In one, the page's content stream contains text operators: select a font resource, set a position, show a string of character codes, and the reader draws the glyph outlines from an embedded font. In the other, the content stream places a single large image object covering the page, and every letter is just an arrangement of pixels inside that picture. The first has a text layer. The second has an image layer. Everything from search to copy to conversion depends on which one you are holding.
A text layer is semantic and resolution independent. The characters exist as data, so a reader can extract them for search, a screen reader can voice them, and the outlines stay crisp when you zoom to four hundred percent because they are being redrawn from vector descriptions each time. It is also compact: a page of prose stored as text and a subset font is a very small amount of data. An image layer knows nothing. It has no letters, only tone values, so zooming reveals pixels and searching finds nothing because there is nothing to find.
Scanners produce the second kind, which is why a freshly scanned contract cannot be searched. Optical character recognition fixes this, and the way it does so is elegant. Recognition analyses the image and produces characters together with the coordinates where each one was found. Those characters are then written into the page as a real text layer, positioned to line up with the picture and drawn in invisible text rendering mode. The scan you see is still the original image, pixel for pixel, but underneath it sits a transcript aligned to the page geometry.
That two-layer structure is why a searchable PDF behaves the way it does. Search hits highlight in the right places, selection follows the visible lines, and copied text comes out as words, while the document still looks exactly like the paper it came from. It also explains the limits: the hidden layer is only as good as the recognition, so a faint fax or an unusual typeface produces a transcript with errors that stay invisible until someone searches for a word that was misread. The image is truth; the text layer is a helpful interpretation of it.
The most useful consequence concerns compression. A tempting way to shrink a PDF is to render every page to an image and rebuild the file from those pictures, which does reduce size but silently destroys the text layer, turning a searchable document into a stack of photographs. That is why compress on PdfWill only recompresses the embedded images and leaves text and fonts intact. It is also why PDF to Word needs real text to work with, so a scan should go through OCR first, and why converting a text document with PDF to Image should be a deliberate choice rather than a default.