Why compressing a PDF sometimes makes it bigger

Run a scanned contract through a PDF compressor and it might drop from 20 MB to 2 MB. Run a text-only report through the same tool, at the same settings, and it can come out larger than it went in.

That isn’t a broken tool. It’s a consequence of what the two files are storing.

A PDF isn’t one kind of file

The .pdf extension covers two things that have almost nothing in common.

A text PDF, the kind exported from Word, Google Docs, LaTeX or a web page, stores instructions: put this character, in this font, at this position. A full page of text costs a few kilobytes, because “the letter A at x,y” is nearly free to write down. Fonts get embedded once and reused on every page.

A scanned PDF, the kind that comes out of a scanner or a phone app, stores photographs. Every page is one big image, usually a JPEG, in PDF packaging. A single page at 300 DPI is comfortably 1–2 MB, because a photo of text costs vastly more to store than the text does.

Nearly every PDF you’ll meet is one or the other, and everything about compression follows from which one you’re holding.

Why scans compress so well

Phone scanning apps save at high resolution and high JPEG quality. Those defaults were chosen to be safe, not small, so there’s usually a lot of headroom: the image can be re-rendered smaller and re-encoded at lower quality with nothing visibly lost.

While building the PDF compressor we ran a 2.06 MB image-only document through it at 96 DPI and 80% JPEG quality. It came out at 254 KB, 88% smaller, with no visible change at reading size. For a scan, that’s a completely ordinary result.

Why text PDFs get bigger

Now do the same thing to a text PDF. The compressor renders each page as an image and stores the image. But the page it started from was a few kilobytes of character positions, and an image of that page, at any quality you’d accept, is tens of kilobytes.

In the same test, a three-page text document went from 1.9 KB to roughly 66 KB. Thirty times larger. The tool did exactly what it was told to do, and there was simply nothing to save: text stored as text is already about as small as that page can get.

Which is the argument for a compressor that tells you the result got bigger instead of quietly handing back a worse file.

The hidden cost: the text layer

Rasterising compression, the kind that re-renders pages as images, has a second consequence that usually matters more than the file size. The text stops being text.

In the compressed copy you can’t:

• select or copy a sentence
• find a word with Ctrl+F
• have a screen reader read it aloud
• let any system pull data out of it automatically

It looks identical and prints identically. It has become a picture of a document. For something a human is only going to read, that’s usually fine. For anything that will be processed, like a contract, a form, or an application an employer or an authority feeds through their systems, send the original.

How to tell which kind of PDF you’ve got

Two quick checks.

Try to select some text. Open the PDF and drag across a sentence. If the words highlight, there’s a text layer. If your cursor draws a selection box over it like a photo, you have a scan.

Check the size per page, meaning file size divided by page count. Under about 100 KB a page usually means text. Over about 500 KB a page almost always means images.

Or drop it into the PDF to text extractor. If plenty of text comes out, it’s a text PDF. If the tool warns you the document looks scanned, it’s images.

Practical rules

• Scanned or image-heavy PDF that’s too big to email? Compress it, and expect to save 60–90%. 96 DPI is enough for reading on screen; go to 150 if there’s small print in the scan.
• Text PDF that’s somehow huge? The weight is almost certainly embedded images or fonts, not the text. Rasterising will hurt more than it helps. The real fix is re-exporting from the source document with the images downsampled.
• Anything that has to stay searchable or machine-readable? Don’t put it near a rasterising compressor, and keep the original whatever you do.

The compressor here puts that trade-off above the tool and reports the before-and-after sizes exactly, so you can judge the result instead of taking it on trust.