Lossless vs Lossy PDF Compression, Explained Properly
One rewrites the file's plumbing. The other trades quality for size. Choosing wrongly wastes time — or ruins a document.
In short
- Lossless compression rewrites the file's structure; the visible document is untouched.
- Lossy compression re-encodes the images — or turns whole pages into images.
- Which one helps depends on where your file's bytes are, not on how big it is.
"Compress a PDF" describes two operations with almost nothing in common. One is housekeeping. The other is a deliberate trade of quality for size. Choosing the wrong one wastes time at best and destroys a document's usefulness at worst.
Where the bytes actually are
A PDF is a container of numbered objects: page descriptions, embedded fonts, images, annotations, metadata, and a cross-reference table pointing at all of them. Break a typical file down and the weight sits in predictable places.
| Document | Where the size comes from | What can reach it |
|---|---|---|
| Word export, 20 pages | Fonts and structural overhead | Lossless |
| Report revised 30 times | Stale incremental updates, orphaned objects | Lossless |
| Colour scan, 10 pages | Ten large JPEG or JBIG2 images | Lossy only |
| Photo brochure | High-resolution photographs | Lossy only |
| Slide deck | Repeated template graphics plus screenshots | Both, partly |
This is the whole decision. If the weight is structural, lossless compression removes it for free. If the weight is pixels, only re-encoding pixels will help, and re-encoding pixels always costs something.
What lossless compression does
Producers write PDFs carelessly. Word, scanner drivers, and export plugins each have habits that leave a file larger than it needs to be, and repeated editing compounds them. A lossless pass cleans up four things:
- Object packing. Thousands of small objects are grouped into compressed object streams instead of sitting individually with their own overhead.
- The cross-reference table. Every save can append a new table while leaving the old ones in place. Rewriting produces one current table.
- Orphans. Deleted pages, replaced logos, and abandoned annotations often remain in the file — invisible, still counted.
- Uncompressed streams. Page content stored as plain text gets deflated.
Nothing visible changes. Text stays selectable, images keep their exact encoding, fonts stay embedded, links keep working, and print output is identical. The catch is the ceiling: once the structure is clean, there is nothing left to reclaim, and a second pass will save zero.
What lossy compression does
Lossy techniques go after the pixels. Serious PDF software does this selectively — downsampling images above a resolution threshold, converting them to JPEG at a chosen quality, sometimes converting colour scans to greyscale. Browser-based tools, including the shrink tool here, take a blunter route: render each page as it would be displayed, save that rendering as a JPEG, and rebuild the document from those images.
The blunt approach has one large advantage — it works on any PDF, however strangely constructed — and one large cost: everything that was structured data becomes appearance.
| Lossless | Lossy (page re-render) | |
|---|---|---|
| Typical saving on text PDFs | 5–40% | Often negative |
| Typical saving on scans | 0–5% | 80–95% |
| Text stays selectable | Yes | No |
| Searchable and screen-reader friendly | Yes | No |
| Links, bookmarks, form fields | Preserved | Lost |
| Can hit an exact target size | No | Yes |
| Reversible | Nothing to reverse | No |
The counter-intuitive case
Running a lossy tool over a clean text document usually makes it bigger. A page of text is a compact set of drawing instructions referencing an embedded font; the same page as a 1600-pixel JPEG is a few hundred kilobytes of photographic data. If a shrink attempt returns a larger file, that is the reason, and the original is the file you want.
Choosing, in one paragraph
If the document contains real text and you need it smaller, compress losslessly and accept whatever you get. If the document is scans or photographs, or you must be under a hard limit today and can live without selectable text, re-render to a target. If the document is a contract, a form, or anything someone will need to search, quote, or read with assistive technology, do not rasterise it — split it or send a link instead. And if you are unsure, run lossless first: it takes a second and it can never leave you worse off.
Each tool reports the before-and-after size, so you can see which technique your file responds to.