First, work out where the size is
PDFs get large for one dominant reason and two minor ones.
Embedded images. Almost always the answer. A PDF built from ten phone photographs is roughly the size of ten phone photographs, because the images are stored inside it.
Embedded fonts. A smaller, fixed cost — but the reason splitting a PDF can produce files that total more than the original, since each new document needs its own copy of the fonts its pages use.
An inefficient producer. Some software writes needlessly bloated PDFs. Less common than it used to be.
The quick diagnostic: scroll through the document. If the pages are photographs or scans, the size is images. If it is plain text and still large, it is fonts or the producer.
What this site can actually do
Compress the PDF directly. Each page is re-rendered at a lower resolution and re-encoded, and a new document is built from the results. For a scan this is exactly right and often dramatic — the pages were already images, so nothing is lost that you had.
For a document containing real text, it is a different proposition: that text becomes pixels, so it can no longer be selected or searched. The tool detects a text layer and says so before you download, rather than after.
What it cannot do is recompress the images inside a PDF while leaving the text intact. That needs Ghostscript — see below.
Three other routes, which are sometimes better than compressing at all:
Extract the pages you need. If three pages of a forty-page report matter, this is a real and often dramatic reduction — especially if the pages you drop are the scanned ones.
Delete the pages you don’t. The same operation from the other direction. Cover sheets, blank scans and duplicate pages come off cleanly.
Rebuild from compressed images. This is the big one, and it applies whenever you made the PDF from photographs. Compress the images first, then build the PDF again. A document assembled from 400 KB images instead of 3 MB images is roughly seven times smaller, and the process involves no PDF compression at all.
If you need text to stay text
The case this site cannot handle: a text-heavy PDF with a few large images, where the text must remain selectable and searchable.
Rasterising would destroy exactly what you are trying to keep, so the compressor here is the wrong tool and will tell you so.
Ghostscript recompresses the embedded images while leaving text as text, and runs entirely on your machine:
gs -sDEVICE=pdfwrite -dPDFSETTINGS=/ebook \
-dNOPAUSE -dBATCH -sOutputFile=small.pdf large.pdf
/ebook targets roughly 150 dpi, which is fine on screen. /screen goes further and looks noticeably worse in print. /printer is more conservative.
Your operating system’s PDF viewer may also offer a reduced-size export — Preview on macOS has a Quartz filter for it, and most print dialogs can “print to PDF” at a lower resolution.
Both of these keep the document on your machine, which matters: scanned PDFs are overwhelmingly contracts, statements and identity documents. Uploading one to a compression service is the decision worth making deliberately.
What compression does and does not damage
Real text is safe. Text in a PDF is stored as characters plus font instructions, not as pixels. Compressing embedded images does not touch it, and the text stays perfectly crisp at any zoom.
Scanned text is image data. In a scan, the letters are pixels, so image compression softens them. Push it too far and a scanned document becomes hard to read — this is the thing to check before sending.
Always look at the result at 100% zoom before you rely on it.
Quick reference
| Situation | What to do |
|---|---|
| A scan that is too large | Compress it — this is the ideal case |
| You built it from photos | Compress the images, then rebuild |
| Only some pages matter | Extract them |
| A few pages need removing | Delete them |
| Text must stay selectable | Ghostscript, or your OS’s reduced-size export |
| It is text-only and still large | Usually embedded fonts — Ghostscript can subset them |
More on why PDFs behave this way: the PDF format explained.