There are two completely different things a tool can mean by "compress a PDF", and almost every online compressor does the same one. Understanding the difference is most of what you need to know.
The two approaches
The first approach is to rebuild the document more efficiently. Object streams, deduplicated images, stripped metadata. The text stays text, the images stay images, and the file gets smaller without anything changing about how it renders.
The second approach is to rasterise: render every page to a JPEG at some DPI and rebuild the file from those images. The result is dramatically smaller, which is why it is popular. The text is now a photograph of text. You lose search, you lose selection, you lose accessibility, and small type starts to look like mud when printed.
Which one you want
| If you need to… | Use |
|---|---|
| Email a document under an attachment limit | Structural compression first |
| Keep the text searchable and selectable | Structural compression only |
| Publish to the web and file size is irrelevant | Either — or convert to WebP images instead |
| Shrink a scanned document a great deal | Re-scan at a lower DPI, or OCR then compress |
Why PDFs are larger than they need to be
Office applications are generous. They write a lot of small objects, embed the same logo once per page, and store fonts and colour profiles that nobody will ever use. That overhead is often 20–40% of a text-heavy PDF, and reclaiming it changes nothing visible.
Scans are the opposite case. A scanned page is usually one large already-compressed image, so there is very little structural slack to remove. If a scan is too big, the honest fix is at the scanning stage: a 300 dpi greyscale scan of a text page carries far more data than a 200 dpi one, and nobody can tell the difference at reading size.
Metadata is free savings
Before you send a file outside your organisation, check what it says about itself. Author, application, workstation name and timestamps are frequently more revealing than the filename, and removing them costs nothing. A scan exported from a phone can carry GPS coordinates in its EXIF data; a PDF written by a word processor can carry the last person to edit it.
A reliable order of operations
- Remove metadata first. It is instant and changes nothing about the document.
- Run structural compression and look at the real number, not the promise.
- If it still will not fit, split it rather than degrading quality — most attachment limits are per file.
- If it is a scan, re-scan at 200 dpi instead of compressing harder. You will get a smaller, better file than any amount of lossy squeezing.
One thing to check before you trust a converter
Open the compressed file and try to select and search its text. If you can, it is still a document. If the selection grabs a whole block, or if search returns nothing, you have been given images wearing a PDF extension.
Doing it in bulk
Most people discover this problem with forty invoices rather than one. Furtu processes batches locally, so a folder of PDFs can be optimised in one pass without any of them being uploaded — and the results page shows the before and after size for every file, so you can see which ones were actually worth it.