A 50MB PDF that's mostly text is doing something wrong. Modern PDFs balloon because of two things: images embedded at full camera resolution, and font subsets that were never trimmed. Once you understand what's inside, you can shrink dramatically without anyone noticing.
What's actually inside a PDF
A PDF is a container. Open one with a hex editor and you'll see a few different kinds of streams:
- Content streams — the actual drawing instructions for each page. Tiny.
- Embedded fonts — typically 100KB-2MB per font, sometimes more.
- Embedded images — usually the bulk of the file. A single 12MP photo at full res is 3-8MB.
- Form data, annotations, attachments — small unless someone went wild with comments.
When you "compress" a PDF, you're almost always squeezing the images. Fonts contribute less and can rarely be reduced safely.
Three levels of compression
Lossless
Removes unused objects, deduplicates resources, recompresses streams with better algorithms. The output is visually identical to the input. Savings: usually 5-20%, sometimes more for messy files.
This is always safe. Run it first on anything before reaching for heavier tools.
Image downsampling
Images embedded at 600 DPI for a screen-only PDF are pure waste. Most screens are 96-150 DPI. Downsampling images to a target DPI shrinks them dramatically:
- 300 DPI — print quality. Photos still look great in print.
- 150 DPI — screen quality. Looks fine on monitors, slightly soft when printed.
- 96 DPI — email/preview quality. Visible softness up close.
Downsampling from 600 → 150 DPI quarters the image data and cuts most PDFs by 50-70%.
JPEG recompression
Many PDFs embed images as lossless PNGs even when JPEG would do. Converting to JPEG at quality 80 typically cuts image bytes by another 5-10×. Quality 80 is visually identical to quality 95 for photographs.
Combined with downsampling, this is where the big wins come from.
When compression hurts quality
Three scenarios where you should compress carefully:
Scanned documents
A scanned PDF is just pages of images. Aggressive JPEG compression introduces blocky artifacts around text edges — readable, but ugly. Use moderate compression (quality 80) and leave DPI at 300 for scanned text.
Diagrams and screenshots
PNG screenshots of code, terminals, or diagrams have sharp edges and limited colors. Converting them to JPEG introduces visible ringing around text. Most good compressors detect this and skip JPEG conversion for images with sharp edges, but verify.
Watermarks and logos
If a logo was embedded as a high-res transparent PNG, JPEG conversion will replace transparency with white and destroy it. Keep these as PNG and only downsample.
How to actually do it
The Compress PDF tool offers three preset levels:
- Low — lossless cleanup + recompress streams. ~10% smaller, no visible change.
- Medium — adds 150 DPI image downsampling + JPEG quality 80. ~60% smaller. Recommended for most use cases.
- High — adds 96 DPI + JPEG quality 70. ~80% smaller. For screen-only PDFs where size is critical.
Almost everyone should use Medium. It's the sweet spot for email-sized PDFs that still look professional when opened.
What "lossless PDF compression" actually means
You'll see services advertising "lossless PDF compression up to 90%". This is misleading. True lossless compression saves 5-20%. The "up to 90%" headline number comes from running aggressive lossy steps and counting them as lossless because PDF readers don't visually flag the change.
If a tool claims 90% lossless savings, it's downsampling and re-encoding images. Check the result against the original — you'll see the difference, even if it's subtle.
After compression
Always verify:
- Open in Adobe Reader and skim a few pages at 100% zoom. Image artifacts show up most at native size.
- Print one page if the PDF is print-bound. Many issues only appear on paper.
- Check OCR text if the PDF is searchable. Aggressive image compression sometimes degrades the underlying OCR layer.
TL;DR
- PDFs are bloated by images, not text or fonts.
- Lossless = 5-20% smaller, no change.
- Medium = 60% smaller, looks identical.
- Verify a printed page if you're compressing a doc that will be printed.
- Don't trust "90% lossless" claims.