Compressing a PDF without losing quality

The phrase is doing a lot of work, because two different things get called quality and only one of them is at risk. Text and vector graphics in a PDF are drawing instructions, and re-encoding a file never touches them: a compressed contract has letterforms identical to the original, at any zoom, on any printer. What compression changes is the images, and there it is genuinely lossy — pixels get thrown away and do not come back. So the honest version of the question is not whether quality is lost but whether the loss is visible in the way you will actually use the document, and that turns out to be a question you can answer deliberately rather than guess at.

Truly lossless compression, and its ceiling

There is a lossless path, and it is worth knowing exactly how far it goes. A PDF accumulates junk: object definitions no longer referenced by any page, duplicate copies of the same embedded font, metadata from four editing sessions ago, unused colour profiles, an inefficient cross-reference table. Cleaning all of this up changes nothing you can see, and on a document that has been edited and re-saved many times it can remove a fifth of the file. On a freshly exported document it removes almost nothing, and on an image-heavy file it removes a rounding error — because the images, which are the file, are untouched by definition. If your PDF is ten megabytes of photographs, lossless optimisation will hand you back a file of about ten megabytes. That is not a tool failing; that is the arithmetic of the format.

Where the quality actually goes

Lossy image compression has two mechanisms and they degrade differently. Downsampling reduces the pixel count, and its effect is a loss of fine detail: hair, thin lines, small text inside a screenshot. Its saving grace is that it is predictable — if a page is displayed at about 1240 pixels tall and you keep 1240 pixels, nothing is missing at normal viewing size, and the loss only appears when someone zooms in past 100 per cent. JPEG quality reduction works differently, discarding subtle variation within blocks of pixels, and its artefacts are local: faint ringing around high-contrast edges such as black text on white, and blotchiness in large flat areas such as a sky or a coloured header. This is why the two knobs should be turned together rather than one to its extreme, and why our presets pair them: 1700 pixels at quality 0.80, 1240 at 0.62, 1000 at 0.45.

Screen and print are different targets

A document that will only ever be read on a screen has a hard ceiling on how much resolution can possibly be useful, and it is lower than most people assume. Even on a high-density laptop display, a full-page view of an A4 document is well under 2000 pixels tall, so anything beyond that is invisible unless the reader zooms. Print is stricter: a photograph reproduced at 150 dpi looks soft on paper where it looked fine on screen, and text that was rasterised as part of a scan starts to break up. The rule of thumb is to compress freely for screen distribution and conservatively for anything going to a printer, and if the document will do both, keep the original and send the compressed copy — which costs a few megabytes of storage and removes the trade-off entirely.

A test that takes twenty seconds

Do not judge the result from a thumbnail, and do not judge it from the first page, which is usually the cleanest. Find the busiest page in the document — the one with a stamp, a signature, a dense table, a footnote in six-point type, or a screenshot with interface text in it — and view it at 100 per cent zoom, then at 200. If the smallest text on that page is still readable and edges are not visibly haloed, the setting is right for the document. If it is not, step down one preset rather than abandoning compression, because the difference between balanced and light is often a factor of three in file size for a change most readers never notice.

The failure that looks like success

There is one result to be suspicious of: a tool that always reports a saving. Some files genuinely cannot be improved, and an honest tool says so and hands back your original. Worse, structural rewriting can inflate a small file — in our own measurements a lightweight document grew by seventy-one per cent when run through a linearising optimiser, because the added structures outweighed anything saved. That is why the sensible behaviour is to keep whichever file is smaller and to report "already optimised" when the gain is below about three per cent, rather than presenting a two per cent difference as a result. A compressor that never admits defeat is measuring something other than your benefit.

Fix it upstream where you can

Most quality anxiety comes from files that were bloated at export rather than from compression. A design application offering press quality and screen quality is offering a factor of five, and the press profile exists for a commercial printer. A slide deck carrying a twelve-megapixel phone photograph on a slide that displays it two inches wide is storing forty times the pixels it can show. A 600 dpi colour scan of a black-ink page is four times the data of a 300 dpi greyscale one and no more readable. When the file is yours from the beginning, choosing correctly at export means there is nothing to rescue afterwards, and no quality question to answer at all.

What to keep, and for how long

For anything with legal or archival weight — a signed contract, a deed, a medical image, a scan of a document you cannot re-obtain — keep the original untouched and treat the compressed version as a disposable derivative for sending. For everything else, which is most documents, the compressed copy is simply the document and there is no reason to keep two. The distinction is worth making consciously once, at the moment you compress, because it is much easier than discovering two years later that the only surviving copy of something important is the one you shrank to fit an email.