Why your PDF is large, and what can be done about it
Almost every genuinely oversized PDF is oversized for one reason: it contains images at a far higher resolution than anything needs. A phone photograph of a document is a twelve-megapixel image, and a scanner set to 600 DPI produces roughly four times the data of one set to 300. Ten of those in a file makes a document of thirty or forty megabytes made almost entirely of pixels.
A text-based PDF is a completely different animal. A fifty-page contract exported from Word might be 200 KB, because the text is stored as characters plus a font rather than as a picture of characters. There is very little fat to remove — the file is already close to as small as its contents allow.
This distinction determines what you should expect. Scanned and image-heavy documents routinely compress by seventy to ninety percent with no meaningful loss of legibility. Text documents often cannot be usefully compressed at all, and a tool that claims otherwise is either doing nothing or degrading your file to prove a point.
How this tool compresses, and the trade-off it makes
Each page is rendered at a chosen resolution and re-encoded as a JPEG, and those images are assembled into a new PDF at the original page dimensions. This is extremely effective on scans and photographs, because it targets exactly the thing making them large.
The trade-off is explicit and you should know it before you click: text becomes part of the page image. It stops being selectable, it stops being searchable, and screen readers can no longer read it. For a scanned document that was already a picture of a page, you lose nothing you had. For a text-based PDF, you lose something real — which is the other reason not to compress documents that were not large to begin with.
If the tool tells you the file is already optimised and hands back your original, that is not a failure. It means rendering the pages produced a larger file than the one you started with, which happens routinely with text documents, and returning the bigger file would be actively unhelpful. It also means this particular method has nothing to offer that file, not that the file is incompressible in principle.
Choosing a compression level
Light renders at around 200 DPI with high JPEG quality. Text stays crisp under moderate zoom and photographs keep their detail. Choose it when the document will be printed, when someone will read it closely, or when it contains fine print that has to survive.
Balanced renders at 150 DPI. This is the right default for almost everything: comfortably readable on any screen, prints acceptably at normal sizes, and typically produces a file a fraction of the original. Most documents being emailed or uploaded should use it.
Maximum renders at 96 DPI with aggressive JPEG quality. Use it when a hard size limit is the only thing that matters and the document simply has to fit. Small text will soften noticeably and detailed images will show compression artefacts, so check the result before sending it somewhere that matters.
Hitting a specific size limit
Upload forms are unforgiving about this, and the limits are often much lower than people expect — 2 MB, 1 MB, sometimes 500 KB or even 100 KB for photographs on government portals. Worse, many of them reject an oversized file without telling you what the limit was. The practical approach is to start at Balanced, look at the resulting size, and step up to Maximum only if you are still over the number you were given.
If Maximum still leaves you above the limit, the problem is usually page count rather than page quality, and no amount of further compression will fix it gracefully. Split the document and submit the relevant pages, or reduce what you are scanning. Compressing a fifty-page scan down to 100 KB produces something technically within the limit and practically unreadable, which fails the actual purpose of sending it.
One habit worth forming: keep the original. Compression is lossy and irreversible, and the compressed file is a copy made for one recipient with one particular limit. The next person who asks for the document may want the readable version, and regenerating it from a heavily compressed copy is not possible.