Compress a PDF Without Losing Quality: What Actually Happens When You Compress

PDF compression has a slightly magical reputation — you drop in a 40 MB file, get back a 4 MB file, and everything looks the same. What's actually going on isn't magic. It's a specific set of operations on the internal contents of the PDF, and understanding them helps you know when to expect huge savings and when to stop trying.

The anatomy of a PDF

A PDF is basically a collection of numbered objects. Some objects are pages (which reference other objects for their content). Some are text streams (compressed blocks of text-drawing instructions). Some are embedded images. Some are fonts. There's an index at the end that says where each object lives in the file.

Compression targets each of these differently. And the ratio of image-to-text-to-font data in your specific PDF is what determines how much can actually be saved.

1. Image downsampling and re-encoding (usually the big win)

A PDF from a scanner or a designer often contains images that are much higher-resolution than any reader will ever display. A 4000×3000 pixel photo embedded on a page that's 8.5 by 11 inches at 72 DPI (screen viewing) doesn't need to be 4000 pixels wide — it just needs to be about 600. The extra pixels are pure waste.

Downsampling means resizing the embedded images to a target DPI — usually 150 for "balanced" presets, 100 for "tiny" ones. That alone can shrink a file 5–10x when it's photo-heavy.

Re-encoding means changing the compression on each image. Photos usually get re-saved as JPEG at quality 75–85; screenshots and diagrams as PNG or as JPEG-XL. Even at the same resolution, this can meaningfully reduce size because many source images were saved at wasteful high-quality settings.

2. Font subsetting

A modern font file can be a couple of megabytes — thousands of glyphs, hinting tables, ligature rules. Most PDFs use maybe 200 of those glyphs. Subsetting means embedding only the glyphs the document actually uses.

Well-authored PDFs are already subsetted, but plenty aren't. Word's default PDF export used to embed full fonts, and many CAD tools still do. A 40-page report with 15 unsubsetted fonts can carry 40+ MB of pure font data.

Compression tools re-subset the fonts, dropping unused glyphs. The saving depends entirely on how bad the original was. On a professionally-exported PDF: negligible. On a Word-with-Comic-Sans-and-fourteen-decorative-fonts PDF: enormous.

3. Stream compression

PDF content streams (the actual page-drawing instructions) are supposed to be compressed with the Flate algorithm (essentially zip/deflate). Occasionally you'll find a PDF with uncompressed or badly-compressed streams — usually from a manually-authored file or an old-school exporter. Re-compressing them gains 10–30% on the text portion of the file.

This is usually a minor saving in absolute terms — text-only PDFs are small to begin with. But it's essentially free, so any decent compressor does it.

4. Duplicate object removal

Sometimes the same image is embedded twenty times (a company logo on every page, for instance) because the exporting tool didn't notice. A good compressor deduplicates: the logo becomes one object, and all twenty pages reference it. Similar deduplication happens with color profiles, form fields, and named destinations. Rarely a huge win, but real.

What savings to actually expect

Rough numbers from my own use:

• Text-only PDF (invoice, letter, plain report): 0–10% saving. It's already small.
• Business report with some charts and one or two photos: 20–40% saving.
• Marketing collateral with lots of imagery: 50–70% saving.
• Scanned document (every page is a photograph): 70–90% saving. This is where compression is most dramatic.
• Designer PDF exported at print quality: 50–80% saving on any preset below print quality.

If your PDF is small and you're trying to make it smaller: probably not going to happen. If your PDF is large and mostly text: you'll get a modest saving. If it's large and image-heavy: expect to be pleasantly surprised. Drop it into Compress PDF and see.

Does compression lose quality?

Depends on which compression. Font subsetting and duplicate removal are lossless — the reader sees exactly the same output. Stream re-compression is lossless.

Image downsampling and re-encoding are lossy. That's where the perceived "quality loss" comes from. On a document that will be viewed on a screen (not printed at large size), the loss is usually invisible — 150 DPI is more than a laptop displays, and JPEG at quality 80 is imperceptible on real-world photographs.

The exception is print. If your PDF is going to be printed at large sizes (posters, book pages, glossy brochures) you want higher-DPI images and the "print" preset — or you keep the file uncompressed. For web, email, and desk-scale printing, lossy is fine and the tradeoff is worth it.

When compression doesn't help

Occasionally you compress a 25 MB PDF, get back 24.9 MB, and think the tool is broken. Almost always one of:

• Already compressed. A PDF from InDesign or a similar tool is likely already downsampled and subsetted. There's no fat left.
• Encrypted. The compressor can't re-encode content it can't read. Unlock via Protect / Unlock PDF first if you have the password, then compress, then re-protect.
• Weird stuff inside. Some PDFs embed video, 3D models, or huge JavaScript. Standard compressors don't touch those objects. Whatever is bloating the file isn't what compression targets.

For scanned documents that stubbornly won't shrink, a nuclear option: extract each page as JPG, compress the images aggressively, and rebuild the PDF via Image to PDF. Slower but usually beats any generic compressor by a wide margin.

The mental model

Think of PDF compression as three separate levers: image quality (biggest lever, mostly lossy), font efficiency (medium lever, lossless), and stream packing (small lever, lossless). Most tools apply all three at once. Your file size after compression is a function of how much slack each lever had to pull.

If you're trying to get to a specific size, adjust the preset. If you're past that and still oversized, the file has hit the floor of what compression can do — the rest of the file is signal, not noise.