You screenshot an invoice, hit save, attach it to an email. Your accountant opens it and the numbers look like they were printed on a wet napkin. Meanwhile, every sunset photo you've ever taken at medium quality looks basically fine. Identical compression, identical software, wildly different results.

So what's actually going on?

What compression is actually doing to your pixels

JPEG, the format behind probably ninety percent of photos you've ever seen online, doesn't store every pixel individually. It chops the image into 8x8 blocks, then runs each block through a mathematical process called the Discrete Cosine Transform. That transform describes the block not as a grid of colours but as a combination of frequency waves: slow gentle gradients at one end, rapid sharp transitions at the far end.

Then comes the ruthless part.

Compression throws away the high-frequency information first, because human vision is less sensitive to fine detail than to broad tonal shifts. A smooth gradient across a blue sky is almost entirely low-frequency. A crisp black letter on a white background is almost entirely high-frequency. Compress both at a matching setting, and the sky loses almost nothing noticeable. The letter loses the very information that made it sharp.

That's not a flaw. It's the design intent.

The invoice problem, illustrated

Take two people sending a document to an accountant. Maya saves hers as a JPEG at medium quality. The black text on white crosses three or four of those 8x8 block boundaries, and each block gets its high-frequency edge data trimmed. By the time her accountant opens it, the numbers have a faint grey halo and the decimal points are genuinely ambiguous. Paulo saves his as a PNG. PNG uses lossless compression: it finds patterns and encodes them efficiently without discarding anything. His invoice arrives pixel-perfect. Roughly comparable file size. Completely different outcome.

The content determined everything, not the settings.

Blocking artifacts, explained

That 8x8 grid is why compressed images develop blocking artifacts or mosquito noise. Push compression hard enough and the grid becomes literally visible. Each block gets averaged and approximated independently, so a sharp edge crossing a block boundary gets reconstructed slightly differently on each side. The result is a staircase of tonal jumps in a spot that should be a clean line.

Flat colour with occasional sharp edges, cartoons, logos, text, diagrams, is the worst possible content for JPEG. Photographic images with organic textures and gradual colour shifts are the best. A portrait at aggressive compression might look slightly soft. Apply that same compression to a PowerPoint slide and it looks broken.

Using JPEG on text-heavy screenshots is like fitting a racing tire onto a tractor. Technically it goes on. Nothing good happens next.

This is also why defaulting every saved image to JPEG is one of the small, stubborn bad habits that the software industry has never bothered to fix. The format was built for photographs. It has no business near a spreadsheet.

What quality settings actually control

When software asks for a quality level between 1 and 100, it's adjusting how aggressively those high-frequency components get discarded. Quality 95 keeps almost everything. Quality 60 starts throwing out mid-frequency detail. Quality 20 strips so much that blocking artifacts appear even in smooth skies.

For web photographs, quality 75 to 85 is a sweet spot most professionals land on: file sizes drop dramatically from 95, but degradation stays invisible at normal viewing distances. Below 60, you're gambling.

PNG and WebP use different strategies entirely. WebP's lossy mode uses larger, variable block sizes that adapt to content, which is why it handles mixed material, text sitting next to photos, better than JPEG does at comparable file sizes. If lossless WebP isn't already your default for UI assets and screenshots, there are easy wins going unclaimed.

When did you last actually check what format your screenshot tool defaults to?

For photos being shared, JPEG at 80 is almost always fine. For anything with text, lines, or flat colour, PNG or lossless WebP will save the embarrassment of sending a document that looks like it was faxed from 1987.

Compression isn't magic file-shrinking. It's a calculated bet that you won't miss what gets thrown away. For a meadow photograph, that bet pays off every time. For a bar chart, it's a losing hand before play even begins, and the format knew that when it took your chips.