Why your PDF is 40 MB, and what actually makes it smaller

A PDF is not one thing. It is a container holding text, fonts, images and instructions about where to put them. When a PDF is unexpectedly enormous, one of those is almost always responsible for nearly all of it — and the useful question is which.

Text is basically free

This surprises people. A 300-page novel, as pure text in a PDF, is on the order of a megabyte. Text is stored as characters plus a reference to a font, not as a picture of the words. It costs almost nothing.

So if your document is text and it is 40 MB, the text is not the problem. Something else in the file is, and compression settings aimed at text will do nothing.

The usual culprit: images

A modern phone photo is around 4,000 × 3,000 pixels. Dropped into a document, it keeps every one of those pixels even though it is displayed at maybe 1,200 pixels wide. Ten photos in a report and you are at tens of megabytes for images being shown at a third of their stored resolution.

This is why image downsampling is the single most effective thing you can do to a large PDF. Reducing images to around 150 DPI is invisible on screen and still fine for most printing. It routinely cuts a photo-heavy PDF by 80% or more.

The other culprit: it is a scan

If the document came from a scanner or a phone scanning app, there is no text in it at all. Every page is a photograph of a page. What looks like text is pixels, which is why you cannot select it, search it, or copy it.

Scans are large for the same reason photos are, but worse: they are typically full-page images at high DPI, one per page. A 20-page scan at 600 DPI is a very big file containing very little information.

Scans also compress badly in the wrong mode. Push image quality too low on a scan and the text — which is really just dark pixels — turns mushy and unreadable. Scans need gentler compression than photographs, not harsher.

If you need the text to be selectable, no amount of compression will do it. That requires OCR, which converts the picture of the text into actual text.

Fonts, and the embedding trade-off

PDFs embed their fonts so the document looks identical everywhere. That is the format's whole point, and it costs space — a full font can be several hundred kilobytes, and a document using four weights of two families is carrying a few megabytes of font data before any content.

Subsetting fixes most of this: instead of embedding the whole font, the file keeps only the characters actually used. Most modern exporters do this automatically. Older ones, and some "print to PDF" drivers, do not.

The invisible junk

Editing history, deleted objects that were never purged, duplicate copies of the same image used on several pages, thumbnails, form data, and metadata from whatever produced the file. None of it is visible and all of it takes space.

This is what lossless compression removes. It typically saves 5–20% and changes nothing you can see — no image is re-encoded, no text is touched. It is the safe first thing to try, and on a file that is already efficient it is also the last thing, because there is nothing else to take.

So which compression should you use?

Try lossless first. If the file drops meaningfully, it was carrying junk and you are done with no quality cost at all.

If it barely moves, the weight is in images. Now you have a decision to make, and it depends on what the document is for. A report going out by email can take aggressive downsampling. A portfolio or anything going to print cannot.

Text stays sharp either way. This is worth stating plainly because people fear it: text in a real PDF is font data, not pixels, and image compression never touches it. You can compress images hard and the text remains perfectly crisp and searchable. The exception, again, is scans — where the "text" is an image and does degrade.

When the file will not get smaller

Sometimes a PDF is already efficient. A text-only document exported from a modern word processor has subsetted fonts, no stray objects and no images. There is genuinely nothing to remove, and a compressor that promises to halve it is either going to do nothing or damage something.

That is a good outcome, not a failure. Check the before-and-after figures rather than assuming a small saving means the tool did not work.