From my naive point of view, this seems a very easy problem for screen readers to solve, especially with LLMs: if the link does not describe what it is, include context like the preceding phrase.