Compress PDF

A PDF is a container rather than a picture. It can hold text as drawing instructions, diagrams as vector paths, and photographs as embedded images — and those three shrink in completely different ways. This is why a tool with a single slider is always guessing on your behalf.

Re-encoding the images is the default because it is nearly always the right answer. The photographs in a report or a brochure are usually most of the file, and compressing them a little further costs nothing anybody can see. Everything else — the text, the fonts, the vector art, the page structure — is copied across untouched, so the document still selects, searches, prints and reads exactly as before.

The lossless tier changes nothing at all. It rewrites the file structure so that many small internal objects share one compressed stream instead of each carrying its own overhead. On a text-heavy document written by a word processor this typically saves a few per cent, occasionally fifteen, and sometimes nothing — which is a real answer rather than a failure, and you get the original back when it happens.

Rasterising renders every page and rebuilds the document out of those pictures. It produces by far the smallest file, and it is the only thing that helps much with a scan. It is also the one option that destroys something, which is why it is last and why the label says so.

After rasterising, the text in your PDF is no longer text. It is a photograph of text. It cannot be selected, cannot be copied, cannot be found with Ctrl-F, and cannot be read aloud by a screen reader — which for a document you are sending to someone else may matter a great deal more than the file size.

For a scanned document this costs nothing, because a scan never contained text in the first place — it was already a stack of pictures, and re-rendering it at a sensible resolution is genuinely the only way to make it much smaller.

For anything produced by a word processor or a design tool, think before choosing it. If the file is going to a person who needs to quote from it, or into a system that indexes its contents, or to anyone who uses assistive technology, the saving is not worth what it takes away. Keep the original either way: rasterising cannot be undone.

Two documents of the same size can have completely different room to move. A twenty-page contract is almost entirely instructions — a few kilobytes of text and a font — and there is genuinely very little to remove; if it is large, the cause is usually an embedded font set or a logo saved at print resolution rather than the text.

A PDF that has already been through a compressor is the other common case. Its images are already at a low quality, and re-encoding them at our default would produce larger files containing worse pictures. Each image is measured individually and kept whenever the new version is not actually smaller, so a second pass over an optimised document correctly does almost nothing.

Images stored losslessly are deliberately left alone. Those are usually screenshots, logos and diagrams, and converting one to a JPEG is how a crisp graphic comes back with smeared edges and coloured fringing around the lettering. The saving is real and the damage is obvious, so that trade is not made automatically.

PDFs are the format people use for the documents they least want to hand to a stranger: contracts, invoices, medical letters, tax returns, identity documents, unpublished work. Every compression here runs inside your own browser, so the file never leaves your device.

You can check that rather than believe it. Open the network tab in your browser’s developer tools and watch while you compress, or disconnect from the internet entirely and compress anyway — it still works, because the code that does the work is already on your machine.