Most of the weight in a large PDF is photographs, not text. This finds the images inside the file, encodes them again at a lower quality, and leaves everything else alone — the text stays text, and nothing is turned into a picture of a page. On a three-page file built around one photograph it went from 904 KB to 230 KB. On a PDF that is mostly text there is almost nothing to win, and it tells you that instead of pretending.
The PDF engine (about 500 KB) loads only when you drop a file in, so the page itself stays light. Your file is read on your own machine and never sent anywhere — open your browser's network tab and watch while you use it.
It depends entirely on whether the weight is in photographs. A three-page file built around one photo went from 904 KB to 230 KB in a browser test. A PDF that is mostly text has almost nothing to give, and the summary says so rather than reporting a win that is not there.
No. Only the embedded images are encoded again. Text, vector drawings and the page structure are copied across untouched, so the file stays searchable and selectable. Nothing here flattens a page into a picture.
Anything with transparency, because JPEG has nowhere to store an alpha channel and converting it would silently wreck the image. Also the scanned formats a browser cannot decode — CCITT, JBIG2 and JPEG 2000. Both are counted in the summary rather than quietly skipped.
No. Re-encoding is lossy and the discarded detail is gone, so keep your original if it matters. One thing is guaranteed though: if the result would not actually be smaller, you get your original bytes back rather than a larger file with worse pictures.
These all work the same way — in your browser. Most are free; the ones marked as needing a plan are not.