Why PDF compression works differently from image compression
A PDF is not a picture. A digitally created one is a set of instructions — draw this glyph from this font at this position, stroke this line — and those instructions are extraordinarily compact. A forty-page text document can sit comfortably under 200KB without any compression at all, because nothing in it is stored as pixels.
A scanned PDF is the opposite. Your scanner photographed a piece of paper, and the file is that photograph wrapped in a PDF container. It carries no text, no fonts and no structure — only an image of a page, which is why a three-page scan can be twenty megabytes while a hundred-page book of text is two.
This distinction decides everything about what compression can do for you. A scan has enormous room to shrink because photographic data always does. A digital text PDF has essentially none, and attempting to compress it usually makes it larger. Knowing which kind you have tells you in advance whether this page can help.
How a target size is actually reached
There are two levers, and they behave differently. Rendering resolution controls how many pixels represent each page — high resolution keeps small print sharp and costs a great deal; low resolution is cheap and eventually turns text into grey mush. JPEG quality controls how precisely those pixels are stored, and the artifacts it introduces cluster around exactly the sharp edges that letterforms are made of.
This tool works down a ladder of resolutions from 200 DPI, and at each one first tests whether the lowest acceptable quality would fit. If it would not, no quality at that resolution can, so it drops to the next resolution rather than wasting attempts. Once it finds a resolution where the floor fits, it searches upward for the best quality that still lands under your target.
The practical consequence is that the result reports the DPI it settled on, and that number tells you more about readability than the file size does. Around 150 DPI a document is comfortable to read on screen. At 100 DPI ordinary text is fine and footnotes are marginal. Below about 72 DPI you are looking at a document that a person can identify but not comfortably read.
What you give up, stated plainly
Every page is re-rendered as an image, so a PDF that had real text in it does not any more. You cannot select or copy from the result, Ctrl+F will find nothing, and a screen reader has nothing to read aloud. If the document was already a scan this costs you nothing, because none of that was there to begin with.
It matters when the source was a digital PDF and the recipient expects to search it — a contract, a thesis, a technical manual. In those cases the right move is the standard PDF compressor, which optimises embedded images and leaves the text layer intact, and accepting whatever size that produces. A smaller file that fails its purpose is not a smaller file, it is a broken one.
When splitting beats compressing
A size cap is usually per file, not per submission. When a document genuinely will not fit, dividing it and uploading the parts separately gives you the full budget for each part rather than one budget spread thin across everything — ten pages at 100KB each is a vastly better document than ten pages sharing 100KB.
It is also worth auditing what is actually in the file before compressing it. Scanned documents routinely carry blank reverse sides, cover sheets and duplicate pages, and every one of those consumes a share of the budget that the pages someone needs to read could have used. Removing them is free quality.