Compress PDF, on this page, means redraw every page as a JPEG image and wrap those images in a new PDF. Reduce PDF size is the same action. It is the reliable way to shrink a scan or a photo-heavy file without sending the document to a server. It is a poor way to shrink a contract that is already a few dozen kilobytes of real text, and the form says that before you start. The original file stays on your disk. The download is a second file. If the second file is worse, delete it.
The quality tradeoff is the whole product. A JPEG is a picture. Once a page is a picture, the letters are pixels. You cannot highlight a sentence, a screen reader cannot read the words, and a link that was a real PDF link is gone. A signature image is still visible if it was visible on the page, but it is no longer a cryptographic signature. Choose this when the recipient needs to look at the pages, not when they need to copy a paragraph out of them. If you need the text to stay text, this is the wrong tool, and no setting here will change that.
The size menu has four choices. "Just make it smaller" uses one moderate JPEG setting and does not chase a number. "Try for 200 KB" and "Try for 100 KB" walk a short list of settings from clearer to blurrier and stop at the first result that fits under that many kilobytes. Custom accepts a target from 20 KB to 10240 KB. Fitting is best effort. The clearest step that fits is the one you get, so a file that fits easily is not crushed to mush. If nothing fits, you get the smallest attempt and a sentence that says the target was missed. The page never labels a 400 KB file as 200 KB. A photograph is a different job: compress image resizes a JPG or PNG toward a kilobyte target without turning a document into a scan.
A target of 200 KB is a common inbox and form-upload limit. A page of dense text, or a colour photo, may not reach it without becoming hard to read. The compress PDF to 200 KB page opens with that target already selected so you do not have to hunt for the menu. The 100 KB page does the same for a tighter limit. Both pages use this engine. Neither page invents a guarantee. Photo scans with large empty margins shrink well. A one-page vector logo sometimes grows, because a logo stored as a few drawing commands is smaller than a photo of that logo.
The work stays in the browser. After you press the button, the page loads pdf.js to draw each page and pdf-lib to embed the JPEGs. Those libraries are stored with this site. They are not fetched from a file host that would see your PDF. The worker that draws the pages is the pdf.js worker shipped here, including the standard font files a text PDF needs in order to paint letters before those letters become pixels. Up to 25 pages are accepted. A longer document should be split first. Memory is the reason: every page is a bitmap for a moment.
How to read the result. The note shows the new size and the original size in kilobytes, rounded. Two hidden fields hold the exact byte lengths, which is what a test can check. If the note says the file grew, do not replace your original. If the note says the target was missed, the download is still the smallest attempt this list can make. Opening it will show whether the blur is acceptable. Going further would mean a smaller picture than the floor already used, and this page stops rather than handing you an unreadable grey rectangle and calling it a success.
Related jobs are separate on purpose. To put several PDFs together first, use merge PDF, then compress the single file if you still need to. To turn photographs into pages, use JPG to PDF. Compressing those pages afterwards is what actually shrinks camera photos. The reduce PDF size page is this form with the other common heading and the same warning. Nothing you drop here is stored, and there is no account. A file you would not email to a stranger is still a file you should not leave in a shared downloads folder after you finish.