Early testing (clearing some unread shit that I'm never going to do) seems hopeful. NVDA often prepends the word separator and sometimes prepends other details about the message depending on how much focus has needed to move, but focus does seem to land on the righ thing here. @FreakyFwoof

@Scott I'm intrigued to look at the code when I get a chance. I honestly don't know how the AI would have pulled it off without using English strings, which means it won't work in other languages. Language independent options would involvh understanding the document structure of the app, which you can't really do without inspecting the tree and and it seems unlikely that there'd be info about the structure of the WhatsApp DOM floating about on the internet. But as I say, intrigued.