Ten megabytes sounds like a lot of text. It almost never is. A typical 10 MB PDF holds one or two page-sized images — a scan, a phone photo, a screenshot exported at full resolution — plus a thin layer of text and fonts wrapped around them.
Where the bytes live
A PDF is a bag of objects: text, fonts, images, metadata, and a cross-reference table that indexes everything. A compression tool can only work on the parts it understands. Here is the honest map for a browser-based compressor:
| Inside the PDF | What object-stream compression does |
|---|---|
| Title, author, producer, dates | Stripped completely |
| Cross-reference table, small objects | Repacked into compressed object streams |
| Embedded images (scans, photos) | Nothing — left byte-for-byte |
| Fonts and page text | Nothing — reorganized, not re-encoded |
That last row is the whole story. If your 10 MB file is a scanned contract, the 9.8 MB of image data stays 9.8 MB.
What the PDF compressor actually does here
The PDF compressor opens the file with pdf-lib inside your browser — nothing is uploaded. It clears the document metadata (title, author, subject, keywords, producer, creator), resets the creation and modification dates, and saves the file with object streams enabled. Those two changes are real but modest: kilobytes, usually.
It also refuses password-protected PDFs, so unlock one first if a file mysteriously fails to load.
The 5% test
Compress the file and look at the reported savings. If you get 2–5%, the weight is image data and no metadata trick will rescue you. The fixes are upstream: rescan at 150–200 DPI instead of 600, export the source document with downsampled images, or crop pages before scanning. Shrinking a scan after the fact means re-encoding the images, which a metadata-stripping compressor deliberately does not do.
Compress once, read the percentage, and stop. If a 10 MB scan comes back at 9.7 MB, believe it — the problem is the scan, not the PDF.