Compress PDF

Almost every oversized PDF is oversized for one reason: pictures. Text is stored as instructions to a font — a handful of bytes per character — and vector artwork is stored as coordinates, so a hundred pages of pure prose is often under a megabyte. A single photograph at camera resolution can be five times that on its own. When a document is too heavy to email, the images are essentially always the culprit.

This explains the result that surprises people most: a text-heavy report barely shrinks no matter which setting you choose. There is nothing substantial to remove. What compression can still do to such a file is small-scale housekeeping — subsetting fonts so only the glyphs actually used are embedded, spotting images that appear on several pages and storing them once, and re-packing the object streams — which typically buys a few percent rather than a few multiples.

The other big win is duplication. Letterheads, logos and background watermarks are frequently embedded once per page rather than once per document, so a fifty-page report can be carrying fifty copies of the same graphic. Detecting and collapsing those duplicates costs nothing in quality and sometimes halves the file by itself.

The levels are not arbitrary strengths of one dial; they are different targets. The gentlest keeps everything at print fidelity and confines itself to the lossless housekeeping above, which is the right choice for anything heading to a commercial printer. The middle setting targets on-screen reading and is the one to use by default — it downsamples oversized images to a sensible resolution and re-encodes them, which is where the large savings come from, while leaving text untouched.

The strong setting pushes the same machinery further, capping image resolution at screen level and accepting visible softening in photographs in exchange for the smallest file that still contains real text. It also enables chroma subsampling, which halves the resolution of colour information while keeping brightness detail intact — your eye is far more sensitive to the second than the first, so this is a large saving for a small perceptual cost.

Extreme is a different operation altogether: each page is rendered to an image and the image becomes the page. It reliably produces the smallest file for image-heavy documents, but everything that made the PDF a document rather than a picture is gone — the text is no longer selectable or searchable, screen readers can no longer read it, and links and form fields stop working. Reach for it only when a hard upload limit leaves no alternative.

Resolution controls how many pixels each image keeps; quality controls how aggressively those remaining pixels are encoded. They fail in different ways, which is why both are worth setting deliberately. Too little resolution makes an image soft and mushy at any zoom. Too little quality leaves the size right but introduces blocky halos around high-contrast edges — most visible on screenshots, diagrams and text inside images.

The practical consequence is that screenshots and line art want high quality and can tolerate reduced resolution, while photographs are the opposite: they hide encoding artefacts well and suffer visibly from lost resolution. When the quality setting is high the tool leaves the choice of encoding to a per-image decision, precisely so that a losslessly-stored diagram is not forced through a photographic encoder that would put ringing into its edges.

A useful way to choose is to work backwards from the destination. Email attachments and web downloads only ever meet a screen, so resolution can come down hard. Anything that will be printed needs the resolution held up and the quality traded instead, because print reveals softness that a monitor hides.

Most compression happens because something has a ceiling: a 10 MB mail server, a 5 MB tender portal, a 2 MB government upload form. The estimate shown before you commit exists for exactly this — it lets you find the gentlest setting that clears the limit instead of over-compressing to be safe and shipping a worse document than you needed to.

If even the strongest lossy setting will not fit, the problem is usually page count rather than page weight, and compression is the wrong tool. Split PDF can divide the document into parts that each sit under the limit, and for scanned documents converting colour scans to greyscale with Grayscale PDF removes two of the three colour channels before compression even starts.