How PDF compression works

What is actually inside a PDF

A PDF is a set of numbered objects with a cross-reference table telling readers where each one lives. The objects that matter for size are images, embedded font programs, and content streams — the drawing instructions describing where each glyph and line goes on the page.

Most content streams are already compressed with Flate, the same algorithm as ZIP, when the file is written. This is why "just zip the PDF" achieves almost nothing: you are compressing data that is already compressed. It also means the text of even a long document is small. A hundred pages of prose is a few hundred kilobytes at most.

Which leaves images and fonts. In nearly every PDF that is uncomfortably large, images are the answer.

The one question that predicts everything

Open the document and try to select a line of text with your cursor. If you can, it is a text PDF: the words exist as text with embedded fonts, and the file is mostly instructions. Expect modest savings — often ten to thirty per cent, sometimes nothing.

If you cannot select anything, and dragging just draws a box, it is a scanned PDF: every page is a photograph of a page, and the file is essentially a bundle of images in a PDF wrapper. Expect dramatic savings, frequently seventy to ninety per cent, because scanners produce images at far higher resolution and quality than reading requires.

A large text PDF is usually a hybrid — a report with embedded photographs or charts. It compresses in proportion to how much of its size is those images.

What a compressor actually does

Downsampling is the big one. A page scanned at 600 DPI holds four times the pixel data of the same page at 300 DPI, and for reading on screen or printing normally, 150 to 200 DPI is usually indistinguishable. Reducing resolution is where the majority of any large saving comes from.

Re-encoding converts images to a more efficient format or a lower quality setting — typically JPEG at a chosen quality. This is lossy and cannot be undone, which is why compressing an already-compressed PDF repeatedly degrades it while saving progressively less.

Font subsetting strips the unused glyphs from embedded fonts. A full font file carries thousands of characters; a document uses a few hundred. This matters most in short documents with several embedded fonts, where fonts can be most of the file.

Structural cleanup removes orphaned objects, deduplicates identical resources — the same logo embedded once per page instead of once — and packs objects into compressed object streams. Individually small, and occasionally startling on files produced by badly behaved generators.

Why yours will not shrink

It is already small. A 150 KB text document has essentially nothing to remove. Compression is not a percentage that applies to any file; below a certain size you are trying to squeeze structural overhead.

It has been compressed before. Lossy compression is not repeatable. Once resolution and quality are down, a second pass has little left to take and mostly degrades quality further.

It is mostly vector content. A CAD drawing or a complex chart can be enormous because it contains hundreds of thousands of individual drawing instructions. There are no images to downsample and the instructions are already Flate-compressed.

The images are already efficient. Photographs exported at sensible resolution from software that knew what it was doing leave nothing on the table.

Choosing a target sensibly

Work backwards from the destination. Email attachment limits are commonly 10 to 25 MB. Government and banking upload forms are frequently far stricter — 2 MB, sometimes 1 MB, and often per file. Print production wants the opposite: 300 DPI images and embedded fonts, so compressing before sending to a printer can be actively harmful.

For a scan that only needs to be readable on screen and legible when printed, 150 DPI in greyscale is usually enough and often turns a 20 MB scan into under 1 MB. For a document with photographs where the photographs matter, compress less and accept a larger file.

If a scanned document has to be small and searchable, run OCR first. It adds an invisible text layer, which makes the document selectable and searchable, and then compression of the underlying image works as normal.