I guess in the perfect magical world where the algorithm detected "suitability for humans for the given query" then if gpt-3 spat out a bunch of stuff and it ranked highly, it would mean this wasn't really spam, because it was useful to humans, even if it was generated in a spammy way.
Well, it could still be spam, just that is very closely resembles non-spam, but the end result is you didn't get the information you were looking for, just something that was closed to it, but misinformation.
Imagine you're searching for vaccine information. GPT-3 generated spam provides pages that highly look like authentic NIH medical reports and expert opinion, only every piece of factual data was replaced with the spammer's product (let's say a competing Pharma vaccine that is less effective)
Search engines are not arbiters of truth, they can only approximate it via consensus knowledge, and if consensus is distorted by vast amount of spam, well, there's a problem.
Think of it like a blockchain ledger. If someone steals more than 50% of the hashing power, than all bets are off. Well if someone steals a non-trivial amount of a subject area with spam, than consensus algorithms break down, and the 'ledger' for that subject looks to be weighed towards the spammer.
