PDF Compression: What Actually Happens to Your File?
Where the bytes in a PDF really go, why scans shrink dramatically and text documents barely move, and what you trade away for a smaller file.
"Compress PDF" describes at least two quite different operations. Knowing which one a tool is performing explains why results vary so wildly between files.
Where the size actually is
Before compressing anything, it helps to know what is taking up room. In most large PDFs the answer is images. A single page scanned at 600 dpi in full colour can easily exceed a megabyte on its own. Text, by contrast, is remarkably cheap — a page of prose is often only a few kilobytes once the font is accounted for.
The other contributors are embedded fonts, which can add hundreds of kilobytes if a full font is embedded rather than a subset; duplicated objects, where the same logo is stored separately on every page; and metadata, thumbnails and revision history left behind by editing software.
Lossless optimisation
The conservative approach rewrites the file structure without touching page content. It removes objects nothing references any more, merges duplicates, subsets fonts down to the glyphs actually used, and packs many small objects into compressed streams.
Nothing visual changes and text stays selectable. The catch is that savings are modest — often a few per cent — and on a file already produced by good software, close to zero. A file that was exported cleanly has little left to remove.
Lossy, image-based compression
The aggressive approach re-encodes the images, or re-renders each page as a picture and rebuilds the document around it. Reducing a 600 dpi scan to 150 dpi cuts the pixel count to a sixteenth, and a lower JPEG quality setting shrinks it further.
This is where dramatic reductions come from. It is also where the cost is: if entire pages are rasterised, the text layer is gone. The document will look similar but you can no longer select, search or copy from it, and screen readers can no longer read it.
Why some files barely shrink
A text-heavy report generated from a word processor is already close to optimal — the fonts are subsetted, the content is compressed, and there are no oversized images. Compressing it may save almost nothing, and a lossless pass can even add a few bytes of overhead. That is not a failure of the tool; there is simply nothing to remove.
My file got bigger. Why?
Lossless restructuring adds a small amount of bookkeeping. If the original was already efficiently written, that overhead can exceed the savings. Keep the original in that case.
Does compression change the page size?
It should not. Page dimensions are stored separately from content, so a compressed A4 page should still print as A4.
Which setting should I use for a scanned document?
Scans respond best to image-based compression because the pages are photographs. Just be aware the result will not be searchable unless you run OCR on it.