@MoritzGlantz Quite honestly this pattern is wickedly hard to make work. The tricky bit is that screen readers only focus on one thing at a time, so either you are looking at the editor/edit field which holds your current text, OR you are looking at the ghost text, which generally isn't focusable and might be anywhere from one word to several paragraphs long. VS Code to my knowledge solves this in first-party AI things by providing an "accessibility view" which shows the ghost text in a temporary editor with actions to insert or reject, mostly so screen reader users can look at the ghost text in a detailed manner rather than havinggggggggggggggggg to hear it all at once. I forget how well that is currently being announced but it's in the accessibility docs I believe. I'm not familiar with that extension though, lost track of VS Code for a couple months which means I might as well know nothing with how fast it's been moving
M
