What makes a PDF large
Almost always images. A page of text is a few kilobytes; a single photograph at phone-camera resolution is several megabytes, and a scan is a full-page image on every page. A two-hundred-page report can be smaller than a fifteen-page scan, and routinely is.
After images, in a distant second place, come embedded fonts — particularly in documents assembled from several sources, which end up carrying half a dozen near-identical subsets. Then structural waste: revision history from repeated saves, orphaned objects nothing references any more, and uncompressed cross-reference tables.
Knowing which of these your file is made of decides whether compression will help at all.
Two different operations wear the same name
Rebuilding writes the file out again cleanly: shared resources stored once, dead objects dropped, the internal tables compressed. Nothing on the page changes. Text stays text, images keep every pixel they had. It is entirely safe and entirely lossless.
Downsampling re-encodes the images at lower resolution or higher JPEG compression. It is where the dramatic reductions come from — and it is destructive, irreversibly, in a way that shows up when someone zooms in or prints at full size.
The [msm_tool slug="pdf-compress"] does the first. Rebuilding a document is honest about what it can achieve: on a text-heavy or badly-assembled file the saving can be substantial, and on a scan of photographs it will be close to nothing, because the images are the file and they are being preserved exactly.
If a scan has to get much smaller, downsampling is the only thing that will do it, and it should be a deliberate decision with the original kept.
Stripping metadata while you are there
The title, author and keyword fields cost almost nothing in bytes, so removing them is not a size measure. It is a privacy one, and it is convenient to do at the same moment.
Those fields are usually filled in by software rather than by a person, which is how a document ends up naming a colleague who left years ago, or carrying the title of the template it was built from. Clearing them before a file goes outside the organisation is a reasonable default.
How much to expect
Set expectations before running anything, because the disappointment is otherwise guaranteed.
- A document saved repeatedly in an editor can lose a lot, because each save may have appended rather than rewritten.
- A document merged from several sources often loses a useful amount to deduplicated fonts and resources.
- A clean export from a word processor is already tight. A few per cent is a good result and there is nothing wrong with the file.
- A scan will barely move. The pixels are the payload.
A tool that promises to shrink any PDF by ninety per cent is either downsampling your images without making that clear, or reporting a percentage against something other than the original.
Fixing the cause instead
Compression is a treatment applied after the fact. The larger wins are upstream, and they are usually easy.
Resize images before placing them. A photograph destined for a half-page slot needs perhaps 1200 pixels across; the 6000-pixel original behind it is invisible in the output and enormous in the file. Scan at 200 or 300 dots per inch rather than 600, which is beyond what any office printer reproduces. And export from the source application with a sensible preset rather than the highest one, since "press quality" means embedding everything at maximum fidelity for a commercial printer you are not using.
Where the limit comes from
Before compressing to meet a cap, check what the cap actually is. Email systems reject on the encoded size, which is roughly a third larger than the file on disk, so a 24 MB attachment can fail a 25 MB limit. Portals sometimes state a limit per file and enforce one per submission.
And when a file genuinely will not fit, splitting it or sending a link is usually a better answer than degrading it until it does. A document compressed past the point where its diagrams are readable has been made smaller and useless.
Check before replacing the original
Open the output and look at the pages carrying photographs, fine diagrams or small type, then compare a couple against the source at full zoom. With a lossless rebuild they will be identical, which is the point — but confirming it takes seconds and is the habit that saves you the one time a file turns out to have been processed by something less careful than you assumed.