Compressing re-encodes the images inside a PDF at a quality level you choose, which is where nearly all the size in a large document actually lives. Scans respond dramatically — a 20 MB scanned contract commonly lands under 3 MB with no practical loss of legibility — because each page is a photograph and photographs compress. A text-based PDF exported from a word processor is a different story: its pages are text and vector instructions that are already compact, and there is very little for a compressor to take. Knowing which kind of document you have predicts the result better than any setting does. The work runs in your browser, so a large file is not limited by an upload cap or a queue, and the document is never sent anywhere. Your original is untouched; you download a new, smaller copy.
Where the size in a PDF actually is
Open the file size question honestly and it nearly always resolves to images. A scanned page is a photograph of paper, typically several megabytes at 300 dpi in colour; forty of them is a large file made of forty large pictures. Compression re-encodes those images and the total falls proportionally.
Text and vector content is already efficient. Glyph positions and drawing instructions compress well by default and there is no equivalent lever to pull, which is why a 2 MB text-based report will not become a 200 KB one no matter what setting you choose. Embedded fonts are the other fixed cost, and stripping them would break the document's rendering elsewhere.
Does text stay selectable
Yes, where it was selectable to begin with. Compression here targets the images; the text layer is left as text, so search, copy and screen readers keep working exactly as before.
A scanned document is the case people get confused about. Its text was never text — it is part of the page image — so it was not selectable before compression and will not be afterwards. Running OCR is what adds a text layer, and it is a separate operation for a separate purpose.
Choosing a quality level
For a scan destined to be read on screen or emailed, a moderate setting is generally invisible: the document is a picture of black text on white paper, which survives compression better than almost any other content. For anything with photographs where detail matters — product shots, medical images, artwork proofs — stay high.
Test with one file before doing a batch, and check the smallest text on the busiest page rather than the document as a whole. That is where over-compression shows first, as softening around character edges, and it is the difference between a file that is smaller and one that is smaller and unusable.