OxygenPDF
compress
optimize
privacy

Why Your PDF Is 38 MB, and What Compression Actually Does

RohmanRohman4 min read
Why Your PDF Is 38 MB, and What Compression Actually Does

Nobody searches "compress PDF" for fun. You hit a wall: the email bounced, the portal rejected the upload, the court's system said the file was too big. The limit is the reason, and the limit decides what you should do, because a PDF is big for one of three reasons, and only two of them are fixable.

The Three Places the Bytes Live

A PDF is a bag of typed objects, and the size concentrates in three of them.

Image streams. Usually 80 to 95 percent of a big file. A single page scanned at 300 DPI in color is about 25 MB of raw pixels. Stored as a JPEG it lands at 300–800 KB; stored as Flate-compressed raw pixels, which some scanners and export tools do, it stays enormous. A photo wrapped in lossless compression is the single most common fixable bloat.

Embedded fonts. The quiet second place. A full Latin font adds 100–400 KB. A full CJK font (Noto Sans CJK, SimSun) adds 4 to 16 MB, which means one font can exceed an entire email limit. Subsetting keeps only the glyphs the document uses: 80 unique characters embed as a ~20 KB subset. You can spot a subset font by its name in the file: ABCDEF+Helvetica; the six-letter prefix is the marker.

Object overhead. Every object carries framing bytes, and every incremental "Save" keeps the old revision's bytes too. A file edited fifty times can carry fifty layers of dead objects. A full rewrite (Save As, not Save) drops them for free.

The Limits That Sent You Here

The walls have names and numbers:

Destination Limit The catch
Gmail 25 MB Attachments are base64-encoded, which inflates them ~33%. The real ceiling is about 18.7 MB of actual file.
Outlook.com ~20 MB Exchange tenants can be configured anywhere up to 150 MB, which is why it bounced for you and worked for your colleague.
US federal courts (CM/ECF) 10–50 MB Each court sets its own cap. Oversized filings are rejected outright.
USCIS online filing 6–12 MB Sources disagree; check the form's own instructions.
State Dept CEAC (visa docs) 2 MB The tightest common limit. A 40-page color scan cannot get there legibly.
Job portals 5 MB The common ATS default for resumes.

Notice the pattern: 2, 5, 10, 25. "Compress PDF" is really four different searches wearing one phrase.

"Under 10 MB" Is a Target, Not a Setting

Open-ended compression takes what it gets. A target is a pass/fail, and reaching one is a ladder, not a slider:

  1. 150 DPI, JPEG quality ~75: the sweet spot for "email it." Halving linear resolution quarters the pixel count, so downsampling beats recompressing every time.
  2. 100–120 DPI, quality ~60: still readable on screen, visibly softer in print.
  3. Grayscale: strips color data on top of the JPEG savings. Fine for scans and contracts, wrong for anything where color carries meaning.
  4. 72 DPI: screen-only territory. Text starts to fuzz.
  5. Split the file: two uploads, two emails. When the ladder runs out, this is the honest answer, not a failure.

No compressor can promise a target in one pass without knowing what is inside. A 40 MB file that is 95 percent photos reaches 5 MB easily. A 40 MB file that is 2,000 pages of text does not, at any setting.

What the $23-a-Month Tool Does Differently

Acrobat Pro's PDF Optimizer is the real thing: separate DPI targets for color, grayscale and monochrome images, a font-unembedding list, JBIG2 for black-and-white scans, and an Audit Space Usage report that tells you which bucket is fat before you touch anything. That diagnostic step is the part worth stealing: compress the thing that is big, not the whole file blindly.

Under half the "free PDF compressor" sites sits Ghostscript, whose presets are the same ladder: /screen at 72 DPI, /ebook at 150, /printer at 300. When a web tool gives you one slider, that slider is usually these presets with a coat of paint.

One honest caveat about JBIG2, the codec that shrinks scanned pages most aggressively: its lossy mode has a documented history of silently substituting similar-looking glyphs, the 2013 Xerox scandal where scanned 6s came out as 8s. For a contract or a filing, prefer settings that keep text as text.

When Compression Cannot Save You

Say it plainly, because most tools will not:

  • Already-optimized files have no headroom. Recompressing a JPEG adds artifacts and can add bytes.
  • Text-only PDFs are near their floor. The content streams are already Flate-compressed; a 500 KB contract might yield 5 percent.
  • Zipping does not work. The content is already compressed, so a zip buys 0–10 percent, yet it is still the most common advice given.
  • Deleting pages does not always shrink the file. Incremental saves keep the old page objects until a full rewrite.

If the file must be under 2 MB and it is a 40-page color scan, no setting preserves readability. Split it, or rescan at a sane DPI.

Compressing One Now

  1. Open Compress PDF
  2. Start at Balanced (150 DPI, quality 75) and check the result against your limit
  3. Step down to Maximum (100 DPI) or Extreme (72 DPI) only if you are still over
  4. Still over? Split the PDF and send it in two pieces

It runs in your browser, so the file never goes anywhere, which matters when the thing you are compressing is a filing or a contract. If the size problem is really a page-count problem, Delete Pages removes what you do not need first.

The right question is never "how small can it get." It is "what is the limit, and what is the cheapest rung of the ladder that gets under it."

Compress a PDF — in your browser, nothing uploaded.

Rohman

Written by

Rohman

I built OxygenPDF because I got tired of uploading contracts and tax forms to random websites. Your PDFs never leave your browser.

Share this articlePost on XLinkedIn

Stop renting your PDF platform.

All 69+ tools free on web. Desktop Pro is $29 once — every desktop tool, the workspace, and batch processing.

  1. $29now
  2. $79after that

14-day money-back guarantee • Works offline • All platforms

We use analytics to understand how our tools are used and improve the experience. No personal files are ever sent.