Measured, not estimated
Why Your PDF Will Not Compress
Published Last reviewed
Sometimes the compressor reports that your PDF could not be made smaller and offers no download. That is not a failure of the tool. It means the engine produced a file at least as large as the one you supplied, and showing you a bigger file labelled as a success would be dishonest.
There are four common reasons, and all four can be measured. The figures below come from the same reproducible benchmark that produced the compression level comparison, run on GPL Ghostscript 9.56.0.
1. It is already compressed
This is the most common reason and the least obvious one. PDF is not a plain container: text, vector art, and layout instructions inside a PDF are normally stored already compressed. Word, Google Docs, LaTeX, and browser print-to-PDF all do this. There is no loose slack left for a second pass to squeeze out.
The benchmark carries the same twelve pages of text twice, once with its content streams left raw and once compressed the way a real exporter would write them. Both produce an identical output. Only the input differs:
| Same 12 pages of text | Input | Output | Result |
|---|---|---|---|
| Content streams left raw | 39.4 KB | 16.7 KB | −57.5% |
| Content streams compressed | 8.13 KB | 16.7 KB | not smaller |
The realistic version is 8.13 KB and compressing it produces 16.7 KB — more than twice the size. Every level refuses it.
The uncompressed version appears to shrink by 57.5%, but that number describes how wasteful its input was, not how good the compressor is. Any tool advertising large savings on ordinary text documents is quietly relying on this effect. A text-only PDF that will not compress is behaving exactly as it should.
2. The images are already small enough
Image downsampling does most of the work in PDF compression, and each level only downsamples images above its target resolution. If a photograph is already at or below that threshold, nothing happens to it, and the rewrite adds a little overhead instead.
In the illustrated report, the largest image measures 1100 by 760 pixels and comes out at exactly those dimensions under High Quality and Balanced. Both levels are refused. Smallest Size does downsample it, and the same file then drops by 93.9% — the presentation export behaves the same way, falling by 95.3% only once its backgrounds are actually reduced.
So a refusal at High Quality or Balanced is worth reading as information: your images are already modest for their level. Trying Smallest Size is the sensible next step, and it is the one case where it reliably helps.
3. The file is too small to win
Compression is not free. The engine writes structural information, metadata, and font descriptors into the output regardless of how small the document is. Below a certain size that fixed cost exceeds anything it can save.
The interactive form in the benchmark starts at 3.43 KB and becomes 6.87 KB at Balanced. Nothing went wrong; the document simply had less content than the wrapper the engine writes around it. Files of a few kilobytes are already about as small as a valid PDF can be.
4. It has already been through a compressor
Running a compressed PDF through a compressor again almost never helps. The images have already been downsampled and re-encoded, so the second pass finds no headroom and re-encodes lossy images a second time, which costs quality without buying size.
The benchmark includes a document already processed once. Compressing it again at High Quality yields 83.9 KB against an input of 78.3 KB, and is refused. If you have already compressed a file, compress the original at a stronger level instead of stacking passes.
What does compress well
For contrast, the cases that work are consistent and predictable:
- Scanned documents. Pages stored as high-resolution images have a great deal to remove — the scan in this benchmark falls by 86.8%.
- Photographs placed far below their native size. A 4000-pixel-wide image displayed at postcard size is pure waste, and downsampling reclaims all of it.
- Documents that repeat an image. A logo or letterhead stored once per page is merged into a single object.
The pattern is simple: compression removes redundant image data. A document without redundant image data has nothing to give, no matter which level you choose.
What to do when nothing works
If every level refuses your file, the document is already efficient and no PDF compressor will change that. The remaining options are about the document rather than the encoding:
- Split it into smaller files if the destination has a per-file limit.
- Remove pages, embedded attachments, or high-resolution images you do not need, then export again from the original application.
- Re-export from the source document at a lower image quality, which controls resolution before the PDF is ever written.
- Check whether the destination accepts a ZIP archive, which will compress an already-compressed PDF by very little but sometimes satisfies an upload rule.
A file that refuses to shrink is not broken, and neither is the tool. The measured level comparison shows what each level does when there is something to remove, and the Terms of Use list the document features that compression can affect.