It's obvious that language follows
rules. This is not because the rules are all written down somewhere - there is no complete
generative grammar for English, and even if there was, no-one could hope to learn it all explicitly.
It's also true that a 'language' such as English, even American English or British English, is not a static unity. As language changes, different speakers will be using different and contradictory rules to produce their utterances, for example, not to mention regional or cultural differences.
We should distinguish between correcting a usage that conforms to no existing commonly accepted rule and correcting one that does. Everyone agrees the former correction is reasonable, but the latter correction may be seen as unreasonable - and this is the point of many arguments founded on the notion that language changes.
Such arguments, supporting a descriptive rather than a prescriptive approach to grammar and recognising the autonomy of language users, need not imply that we can just play Humpty Dumpty and make up any old rules we like - to be valid, a usage must exist 'in the wild'. One criterion for this might be: "is the usage citable, by some group of language users, in explaining or correcting language use?"
The fact that there may be borderline cases, vagueness, even, in this distinction, does not invalidate the distinction itself - which still applies very well, and has utility, in many many cases.