When an AI answers from your page, it does not read the whole page and reason over it. It works with pieces. Understanding how gives you the why behind answerability.
First the page is split into chunks, roughly a heading and the passage under it. Then each chunk is turned into an embedding, a list of numbers that captures its meaning, so passages about the same thing sit close together in that number space. The question is embedded the same way. The engine retrieves the chunks whose embedding is closest to the question, and answers from those.
This is why the passage, not the page, is the unit. A page can be about a topic and still lose, if the answer to the real question is spread across three chunks or buried in one that is mostly about something else. It is also why a self contained passage that leads with the answer wins: it lands cleanly in one chunk that matches the question.
Topkay uses the same method, with a pinned embedding model, to score answerability and to search your twin. Same page, same model, same score every time.