We redesigned the highlighter eleven times. It sounds absurd for a coloured rectangle. It may be the detail that mattered most in the whole product.
How annotation tools actually fail
Not by being invisible. By making the page unreadable afterwards.
Push the colour, tighten the line height, get seven highlights on one screen, and the text turns into a block of paint. You marked it so you could reread it later. You never opened it again.
Every annotator has this problem, and screenshots hide it. A stronger colour looks more effective in a still image.
What the eleven versions were doing
The first five were about opacity. Too pale and you can't see it, too saturated and it covers the words. We spent a while on fully transparent with an underline, then found that people scroll straight past their own marks.
Versions six through eight changed the approach. Leave the colour, change the blend. Make it read like a thin wash soaking into the paper instead of a bar sitting on top of the type.
Nine and ten were edges. Square corners look mechanical, fully rounded looks like a sticker. We landed on a small radius with half a character of breathing room at each end.
Eleven was colour separation. Four colours distinguished only by hue fail for colour-blind readers, so each one also got its own lightness.
The counterintuitive part
The final version is nearly invisible in a screenshot. A colleague looked at my screen and asked whether anything was highlighted.
Reading a three-thousand-word piece with it feels right, though. Your eye finds which sentences are yours, and the letters are still letters.
That became a rule on our acceptance checklist. Any annotation style gets twenty marks on a three-thousand-word article, then somebody reads it start to finish. If they can't get through it, it fails. Looking good in a screenshot counts for very little against not interrupting the reading.
Related: What local-first costs us · Five hundred highlights, eleven citations