Screen reader heuristics are the wrong approach. If you want to use "click here", "download", "pdf", or something generic then use aria-label, aria-labelledby, or some other mechanism to give the additional context to screen readers and other assistive technologies. There's no need for any heuristics other than those in the specifications that the screen readers, web browsers, and web sites agree/conform to.
There may not be a surrounding sentence (e.g. in a "PDF"/"Download" link). The surrounding sentence may not add any/enough additional context, e.g. "You can click here for help." or "To view the article [one of many] you can click here.".
You then run into issues where 1) the screen reader/AT is overriding the ARIA/a11y markup on the page against WAI-ARIA and WCAG standards; and 2) you can end up with different behaviour on each screen reader, in addition to the places they already differ.
It's bad enough when web browsers introduced heuristics on form labels such that "name" + "label" fields were detected as a login form and other situations where bad heuristics were used and the web browsers overrode what websites were specifying.