Almost nobody compresses a PDF because they want to. They do it because a form has just refused a file, usually after they filled in everything else, and usually with a message that says nothing more helpful than “maximum size 2 MB”. The goal is not really a smaller file. It is a file that is small enough while still being a document someone can read.
Those two goals pull against each other, and the way most people resolve the tension — drag the quality slider down until the number goes green — is the worst available method. It throws away detail in the places you notice most, and it usually does not even produce the smallest file. What follows is the reasoning that gets you a better result in about the same amount of time.
First, find out what is actually big
Every oversized PDF is oversized for one of three reasons, and the correct response is completely different in each case. Diagnosing it takes about ten seconds and saves you from optimising the wrong thing.
Open the file and try to select a word with your mouse. If text highlights, you have a digital PDF — one generated by software, where the words are stored as real characters. If nothing highlights, you have a scan: a photograph of a page, where everything you see is pixels.
That single test tells you most of what you need.
A scanned document is large because images are large. A page photographed by a modern phone can be four megapixels of detail describing a sheet of paper with two hundred words on it. The resolution is real, but almost none of it contributes to legibility. This is the happy case: aggressive compression discards detail nobody was using.
A digital PDF that is large is usually carrying embedded fonts, or images somebody pasted in at full camera resolution, or both. Fonts are not compressible in any useful way — they are already compact, and removing them breaks the document’s appearance on other computers. Pasted images can be shrunk, but ideally at the source rather than after the fact.
A digital PDF that is large for no obvious reason has usually been edited many times. PDFs accumulate: each save can append a revision rather than rewriting the file, orphaned objects linger, and resources get duplicated. A file like this can often be made 30 per cent smaller with no visual change whatsoever, simply by rewriting its structure.
The two kinds of compression, and why the distinction matters
There is a genuine technical fork here, and every tool that offers a “compression level” is quietly choosing for you.
Structural compression rewrites how the PDF stores itself. Duplicate objects are merged, unreferenced data is dropped, and everything is packed into compressed object streams. Nothing about the page changes. Text remains selectable, images keep every pixel, and the document is byte-for-byte equivalent in appearance. The saving is typically 5 to 20 per cent, occasionally much more on a heavily edited file.
Rasterisation is the aggressive option. Each page is drawn exactly as a reader would see it, and that drawing is saved as a JPEG which becomes the new page. Because JPEG discards detail the eye is poor at noticing, the savings are dramatic — 70 to 95 per cent on a scan is routine.
The cost of rasterisation is not really visual. At sensible settings the page looks almost identical. The cost is that the text stops being text. It cannot be selected, searched, copied, or read aloud by a screen reader. Automated systems that extract data from your document will fail. For a scan this costs nothing, because there was never any text to lose. For a contract, a CV or a report, it is a significant and irreversible loss that no one warns you about.
The rule follows directly: if selecting a word highlights it, think hard before rasterising.
Resolution beats quality, every time
When you do rasterise, two settings control the outcome, and people almost always adjust the wrong one.
Resolution determines how many pixels are drawn per inch of page. It scales quadratically: halving the DPI removes three-quarters of the pixels. Going from 300 DPI to 150 DPI cuts the data by about 75 per cent before any JPEG compression is applied at all.
JPEG quality determines how much detail is discarded within those pixels. Its effect is roughly linear and much weaker. Dropping quality from 85 to 70 saves perhaps 30 per cent; dropping from 70 to 50 saves another 25 per cent but introduces visible blocky artefacts around text edges.
So take your savings from resolution and leave quality reasonably high. A page at 120 DPI and quality 80 will be both smaller and better-looking than the same page at 200 DPI and quality 40. This is the single most useful thing to know about PDF compression, and it is the opposite of what the average slider encourages you to do.
Sensible resolution targets:
| Destination | Resolution | Notes |
|---|---|---|
| Screen viewing only | 96 DPI | Comfortable to read, does not survive zooming |
| General purpose | 120–150 DPI | The right default for almost everything |
| Will be printed | 300 DPI | Industry standard for reproduction |
| Archival | Do not rasterise | Keep the original |
The setting almost nobody uses
Greyscale.
Most scanned documents are black text on white paper. A colour scanner records them in full colour anyway, so you are storing three colour channels to represent a page that contains two colours. Converting to greyscale typically halves the file again, on top of whatever the resolution and quality settings achieved, and for a text document there is genuinely nothing to see.
Check first that the document has no colour that carries meaning — a signature in blue ink that proves it is not a photocopy, a colour-coded chart, a company logo whose absence would look odd. If not, turn greyscale on. It is the largest saving available for the smallest cost, and it is buried in an advanced options panel on almost every tool that offers it.
Delete before you degrade
Before touching image quality at all, look at whether every page needs to be there.
Duplex scanners produce a blank page for every single-sided sheet. Fax cover pages, separator sheets, and the scanner’s own calibration page all end up in the file. On a typical scanned bundle, somewhere between 10 and 40 per cent of the pages carry no information at all.
Removing them is completely lossless. Nothing degrades, the remaining pages are untouched, and the file gets proportionally smaller. It is strictly better than compression, and it frequently gets you under the limit on its own.
The same logic applies to relevance. If a form asks for proof of address and you are sending an eighteen-page bank statement, sending the one page that shows your address is smaller, faster, and more considerate of everyone’s time.
A procedure that works
Putting it together, here is the sequence that reliably gets a document under a limit with the least damage:
- Remove pages that carry nothing. Blanks, covers, separators, anything not asked for.
- Check the size. You may already be finished.
- Try structural compression. Lossless, quick, and on an edited document it can be substantial.
- Check again. Still finished, possibly.
- If it is a scan, rasterise. Start at 120 DPI, quality 80, greyscale on if there is no meaningful colour.
- Read the result before sending it. Open the file and look at the smallest text on the busiest page. If it is comfortable, you are done. If not, step the resolution back up rather than the quality.
Two things not to do. Do not compress twice — each pass is lossy and the second one degrades already-degraded output for a much smaller saving. And do not compress in the middle of assembling a document; if you still have merging or reordering to do, do that first and compress the finished article once.
When compression is the wrong answer
Sometimes the file genuinely cannot get small enough while remaining useful, and forcing it produces a document that technically uploads and practically fails.
If you are below 96 DPI and still over the limit, the document is too long rather than too heavy. Split it and upload in parts, or ask whether the recipient needs all of it. Many portals that impose a 2 MB limit will accept several files, and a reviewer would far rather receive three readable documents than one unreadable one.
If the document is going to be printed, do not rasterise below 300 DPI regardless of what the size limit says. A 150 DPI page looks fine on screen and visibly soft on paper, and by the time anyone notices, the printing has happened.
And if the document is legally significant — a contract, a filing, anything that might be quoted or searched — preserve the text layer. Use structural compression only, and if that is not enough, solve the size problem by sending fewer pages rather than worse ones.
The tools referenced here are all on this site and all run in your browser, so you can try several settings on a confidential document without it leaving your computer. Start with Compress PDF; if the file is a bundle you are still assembling, do the merging and page removal first.