Making PDFs Smaller: What Compression Actually Does

A ten-page scan takes up 45 MB while a complete 300-page e-book fits in 2 MB – how is that possible? Once you understand what actually consumes space inside a PDF, you'll instantly see why some documents lose 90 % of their size while others barely budge. This article looks at what a PDF is made of, how lossless and lossy compression work under the hood, what results are realistic – and when a desktop tool like Ghostscript is honestly the better choice than any browser tool.

What Makes a PDF Large in the First Place?

Embedded Images – By Far the Biggest Factor

Almost every large PDF is large because of images. In a scanned document, every page literally is one big photo: an A4 page at 300 DPI measures about 2480 × 3508 pixels – roughly 8.7 megapixels per page. Even stored as JPEG, that adds up to 1–4 MB per page, which is why a ten-page scan weighing 20–40 MB is completely normal. In "regular" PDFs exported from Word or PowerPoint, inserted photos, logos and screenshots are usually the main culprit, too.

Embedded Fonts

To make a PDF look identical everywhere, fonts are embedded into the file. A single typeface costs 20–300 KB depending on its size; a CJK font (Chinese, Japanese, Korean) can add several megabytes. Well-behaved PDF generators embed only the characters actually used (called subsetting) – badly behaved ones embed the full font, sometimes several times over.

Structural Clutter

PDFs accumulate junk over their lifetime: some applications append a new revision every time you hit save (incremental updates) without removing the old one – plus duplicated images, orphaned objects, thumbnails, form data and metadata. This clutter rarely dominates, but it can account for 10–30 % of a file.

Lossless vs. Lossy – the Crucial Difference

Lossless Compression

Lossless means the document looks exactly the same afterwards. This includes deflating content streams (Flate/ZIP-style), packing many small objects into object streams, removing unused objects and stale revisions, and deduplicating identical images and fonts. The catch: text and vector graphics in modern PDFs are already stored compressed. Lossless optimization therefore typically yields only 5–20 % – unless the file is structurally neglected.

Lossy Compression

The big savings almost always come from the images, via two levers:

• Downsampling: the pixel resolution is reduced, e.g. from 300 to 150 DPI. Because pixel count grows with the square of the resolution, this leaves only a quarter of the image data.
• Recompression: images are re-saved with stronger JPEG compression. Going from quality 90 to 60 often halves the size again – with visible but frequently acceptable artifacts.

Together they explain how a 40 MB scan can drop to 4 MB: a quarter of the pixels times stronger compression leaves just a tenth of the original size. Important: lossy compression is irreversible – discarded image detail cannot be recovered by "upscaling" later.

Why a Scan Shrinks Massively but a Text PDF Barely Changes

A PDF exported from Word doesn't store text as an image but as compact drawing instructions: "draw string X in font Y at position Z." A full page of text often needs just 2–5 KB – already Flate-compressed. There is simply nothing left to squeeze. If your 500 KB text PDF comes out at 480 KB, that's not the tool failing – the content was never bloated to begin with.

A scan is the opposite: the letters only exist as pixels inside the page photo. That's why compression looks so dramatic there – and why it's worth running text recognition with OCR if you want to search the scan or copy text later. OCR doesn't shrink the file; it adds an invisible text layer on top of the image.

Quality: Realistic Expectations

Rough real-world guidelines:

• Scanned documents (300 DPI): 60–90 % savings are realistic if 150 DPI is enough for the purpose.
• Office PDFs with photos and screenshots: usually 20–50 %.
• Pure text PDFs: 0–15 % – only structural optimization helps here.
• Already-compressed PDFs: very little headroom. Recompressing repeatedly only degrades image quality without saving much.

What to watch for: in scans, small print suffers first – below roughly 120 DPI, 8-point text turns mushy. JPEG artifacts show up as "ghosting" around letter edges. Always keep the original and inspect the result before you send it off or delete the source file.

Which Tool for Which Job?

How to Proceed, Step by Step

1. Check whether your PDF is a scan or a "real" text PDF: can you select text (for example in the PDF viewer)? If yes, it's a text PDF – expect little savings.
2. Decide what the file is for: screens are fine with 100–150 DPI, print needs 300 DPI.
3. Compress with /tool/compress (or Ghostscript for large batches) and inspect the result at 100 % zoom, especially small print.
4. If the savings aren't enough, images are often not the problem – splitting the document with /tool/split and sending it in parts usually helps more.
5. Keep the original until you've verified the result.

Frequently Asked Questions (FAQ)

Does compression hurt text quality?

Not in real text PDFs: text remains vector-based and stays razor-sharp at any zoom level. Only in scans is the "text" part of the image, and heavy compression will visibly soften it.

Can I undo compression?

No. Image data removed by lossy compression is gone for good. Always keep the original file, at least until you've checked the result.

Why does my file barely shrink?

It most likely consists mainly of text and vector graphics, which are already compressed – or it was compressed before. Either way there is little left to gain; that's normal.

A portal demands 2 MB maximum – can a scan meet that?

Usually yes: a 10-page scan at 100–150 DPI and medium JPEG quality typically fits under 2 MB. For very long documents, switching to black and white instead of color or splitting the file helps further.

Does this work with password-protected PDFs?

Encrypted PDFs must be unlocked before compressing – no tool can rewrite their contents otherwise. With the security tool you can remove the protection (if you know the password) and re-apply it after compressing.