Given we can write the documents in LLMs and summarise/explain the documents with LLMs one would hope we would go back to written documents instead of videos now.

Instead we've worked out how to get the LLM to make videos.

People are bad at writing clearly what a new startup or product actually does. A video is a richer medium, it can show the thing being used, the UI during use, tell a clearer story with a person using it, some context etc. It also forces them to prioritize more, and have to distill their message down to a certain bounded timeframe.

Neither gives any guarantee. Both static text and videos can be done well and badly. But I usually find it useful to have both. A video is like being guided through their vision of what the thing is (a "push" message). This can be annoying to many nerd types who want to cut through all that and just want to "pull" the info they need, for themselves, at their own fast pace, quickly identifying it and not being spoonfed. But the typical user is not like this, and prefers the spoonfed/edutained approach.

12.4% of the world is illiterate and I'd guess more like 20% would have trouble quickly taking text and turning that into instructions faster than a video could do the same thing.

Yeah because people don't like to read. They do it if they're forced to but no more.