@vcbuzz #VCBuzz A3.
I take it we are talking about optimising content/pages/sites for more advanced algorithms etc.
In which case,
pretty much the same approach as for human written rules, statistically observed (human/machine), stochastic/heuristic etc.,
by deduction.
>>>

@vcbuzz >>>
As is often the case, we don't "know" what inputs are used (we can hazard guesses though - based on pre-existing knowledge/claims, papers/research, logic/observations/testing, gut/hunches etc.).
But we can see what seems to rank,
for a given term/language/location etc.
>>>

@vcbuzz >>>
We also don't generally know "how" they work it out,
but we can see what is common (present or not, order/sequence/frequency etc.).
We can then test and see whether there is a reaction, or not.
Not the best, nor fastest, approach,
but still the only one we really have
>>>

@vcbuzz >>>
(The alternative is to build your own systems,
and attempt to emulate/replicate ... which is no easy nor small task - and not guaranteed to be "true" (we might miss X, substitute Z with Y etc. - and get highly similar (but technically wrong) results).