Almost all of the size of a large PDF is images. A scanned document is a stack of photographs and compresses dramatically; a report typeset from a word processor is text, vectors and font data, and there is very little there to remove. If a 3 MB PDF barely changes when compressed, it is because it was already close to as small as its contents allow.
Knowing which kind of file you have takes one look, and it predicts the outcome before you try anything.
Which kind of PDF do you have?
Two families, and they behave completely differently.
| Kind | How to tell | Typical saving |
|---|---|---|
| Scanned | Text cannot be selected; zooming shows pixels | Half or more |
| Text-native | Text selects cleanly; zooming stays sharp | A few per cent |
| Mixed | Selectable text plus large photographs | Somewhere between |
The test is to zoom in. Text that stays crisp at 400 per cent is real text — a few kilobytes of characters plus an embedded font. Text that turns into blocks of pixels is a picture of text, and a picture of a page at 300 dpi is several megabytes before anything else is added.
What does browser-based compression actually do?
Three specific things, and it is worth being clear about the boundary.
- Re-encodes embedded JPEGs at the quality you choose, which is where the saving comes from.
- Scales images down where they are larger than the target dimension, because a 4000-pixel-wide scan on an A4 page is far more detail than the page can show.
- Rewrites the file structure using object streams and drops metadata, which reclaims a small and consistent amount.
What it does not do: re-run text through OCR, subset or replace fonts, flatten form fields, or convert vector artwork to bitmaps. Desktop tools do some of those, and every one of them can lose information — a flattened form is no longer fillable, and a subsetted font can break text extraction.
This is also why compressing an already-compressed PDF achieves almost nothing. Once the images inside have been re-encoded at a lower quality, running the same process again finds no headroom and only adds another round of loss — so a file that has been through a compressor once should be regenerated from its source rather than compressed again.
Why do fonts take so much space?
Because a PDF embeds them so the document renders identically everywhere, and a full font file with several weights runs to hundreds of kilobytes. On a two-page letter that can be most of the file.
Subsetting — keeping only the glyphs actually used — is the standard fix and is normally applied by whatever produced the PDF. When it has not been, the file carries the entire typeface including characters that never appear, and that is one of the few cases where a text-only PDF is unexpectedly large.
What actually works when nothing shrinks?
Attack the source rather than the output.
- Rescan at a lower resolution. Two hundred dpi is legible for most documents and is 44 per cent of the data of 300 dpi.
- Scan in greyscale or black and white where colour carries no information. A colour scan of a black-and-white page is three times the data for nothing.
- Resize the images before building the PDF, if you assembled it yourself.
- Split the document. Two 4 MB files often pass where one 8 MB file does not.
The last one is worth remembering because the constraint is usually an email or upload limit rather than storage, and limits apply per file.
Will it blur the scans?
At high quality settings, not perceptibly. The re-encode is applied to images that are typically far higher resolution than the page needs, so scaling a 4000-pixel scan to 1600 and re-encoding at 70 per cent quality usually produces a document that looks identical on screen.
Where it shows is on fine print and on anything you will zoom into. If the document is a legal contract that someone will read closely, keep the original and send the compressed version only where the size limit demands it.
Questions people ask
Does it work on password-protected files? No. An encrypted PDF cannot be read without the password, so the protection has to be removed first by whoever holds it.
Is my document uploaded? No. The file is parsed and rewritten in the browser, and it never leaves the device — which is the main reason to use a browser-based tool for anything confidential.
Why is the first run slower? Because the PDF parser loads on first use rather than on page load. Reading an existing PDF genuinely needs a full parser, so it is fetched when a file is chosen and not before.
What if the tool says there is nothing worth saving? Then the original is the better file. A compressed version that is the same size as the original has lost quality for no benefit, and it is more honest to say so than to hand back a file that is 2 per cent smaller.
Look at what is inside the file before deciding what is possible with it. Compress PDF handles the image-heavy cases, PNG to PDF builds documents that are the right size to begin with, and the image resizer is where to fix the inputs.