A friend sent me a 62 MB PDF last week — a lease agreement, twelve pages, no photos. Her landlord's email had bounced it back three times. She asked me, half-joking, whether she should just print it and re-scan it as a smaller file. That would have worked, actually. It would also have been the wrong tool for the job.
PDFs are strange creatures. A one-page invoice can be under 100 KB or over 20 MB depending entirely on how it was made — not what's in it. This post is about why that gap exists, and how to close it without your document turning into a blurry mess.
A PDF is a container, not an image
The first thing to know: a PDF is not a picture of a page. It's a package. Inside are three kinds of things — text (as actual letters, with fonts), vector graphics (lines and shapes described mathematically), and raster images (photos, screenshots, scans). Each of those has completely different compression rules.
That's why file sizes vary so wildly. A text-only contract might be 40 KB per page. Add one high-resolution logo and it jumps to 400 KB. Scan the whole page as an image instead of typing it, and one page can be 4 MB. Same content, 100× the size.
The four things a compressor actually does
1. Downsample images. A photo embedded at 600 DPI when the PDF will only be viewed on a laptop is 4× larger than it needs to be. Compressors resample it to 150 DPI (print-quality) or 96 DPI (screen-quality) and you literally cannot tell the difference on a screen.
2. Re-encode images. That embedded PNG of a receipt is likely 3 MB. Re-encoded as a quality-80 JPEG, it's 300 KB and still perfectly readable. For photos, JPEG or WebP beats PNG by 5–10×.
3. Subset fonts. A PDF often carries the entire font file even if it only uses 40 letters. Subsetting throws away the unused glyphs and can save hundreds of KB per font.
4. Strip metadata and duplicates. Old edit history, embedded thumbnails, duplicate copies of the same image on repeated pages, XMP metadata — all things a compressor removes without touching what you can see.
Why 'losing quality' usually means step 2 went too hard
When people say a compressed PDF 'looks bad', 99% of the time it's the image re-encoder. Quality 90 is invisible. Quality 75 is fine for most things. Quality 50 starts to show blocky artefacts on gradients and skin tones. Quality 30 makes signatures unreadable.
A good compressor lets you pick. A bad one picks quality 30 for you and calls it 'maximum compression'. If your PDF suddenly looks like a fax after compression, that's what happened.
The safe default for most documents is 'ebook' or 'screen' preset — 150 DPI images, quality 85 JPEG. It typically shrinks a scanned PDF by 60–80% and nobody can tell.
The three PDFs that won't compress
Text-only PDFs already tiny. If your file is 300 KB of pure text, no compressor can help — there's nothing to squeeze. This is a good sign, not a bad one.
PDFs already compressed. Running a compressor twice does nothing the second time. If you got a PDF from a legal firm or a publisher, it's probably already been through this process.
PDFs that are secretly just one giant image per page. These are 'scanned' PDFs. The trick isn't to compress harder — it's to run OCR first (turn the image back into text), then re-generate the PDF from that text. Size can drop 20×.
The privacy angle nobody mentions
Most 'compress PDF online' sites upload your file, process it on their server, and store the result on their disk for 1–24 hours. That's fine for a public brochure. It's not fine for a signed contract, a tax return, medical records, a passport scan, or the M&A term sheet someone just emailed you.
Bluebird's compressor runs entirely in your browser tab. The PDF never uploads. You can verify this yourself: open the browser's Network tab (right-click → Inspect → Network), then drop your file. You'll watch the tool load; you will not watch your document leave. When compression finishes, everything sits in your browser's memory and vanishes when you close the tab.
This isn't a marketing angle — it's a genuine legal difference. Under GDPR, LGPD and India's DPDP Act, a service that never receives your data isn't a 'data processor'. There's nothing for anyone to leak, subpoena, or accidentally train an AI on.
A recipe that works for 90% of PDFs
For contracts, invoices, letters: use the 'ebook' preset — targets 150 DPI, quality 85. Typical result is 40–70% smaller, indistinguishable from the original on any screen.
For PDFs you'll email as attachments: aim under 5 MB (Gmail, Outlook, and most mail servers start choking above that). If the ebook preset lands you at 8 MB, drop to 'screen' — 96 DPI, quality 78. Still crisp on a laptop, might show softness if you zoom to 400%.
For photo-heavy PDFs (portfolios, presentations): resist the urge to crush them. Compress each image separately to WebP first, then rebuild the PDF. You'll get better quality than any generic PDF compressor because you're picking per-image quality.
For scanned PDFs over 20 MB: run OCR first, then compress. A scanned 40-page contract routinely goes from 30 MB to 800 KB this way — because you converted 40 images to 40 pages of text.
One trap to avoid
'Extreme' or 'maximum' compression presets almost always do more damage than they're worth. They downsample images to 72 DPI, drop quality to 40, and produce files that look fine on your monitor but embarrassing when someone opens them on a 4K screen or prints them.
The difference between 'ebook' and 'extreme' is often 500 KB on a 5 MB file — a rounding error. The difference in how the result looks is enormous. When in doubt, use ebook. Your future self, opening the file six months later, will thank you.
