Generally, when you google things, you are looking for either primary or secondary sources to construct and synthesize ideas. Attribution is always important and necessary because one wants to 1) give credit to the right source, 2) understand where ones information is coming from to put it in context of the broader picture of the field. Something like GPT makes it easier to muddy the water by not giving credit to the original source material that allowed it to construct its output, but you can still put the output in some context by referencing your own knowledge (though the process would be more onerous).

For example, someone may construct an argument using a GPT model that animals evolve traits by requiring the need for that trait to exist, i.e., if a crab needs really big claws to eat, the next generations will have big claws. But this is not how evolution works, and we know that from lots of prior research. This is an easy example to both understand and construct a counterpoint against. Without GPT, this person would have needed to look up resources to make their argument, and they would have cited Lamarckism, and if they had we would immediately know where they're getting their idea and fend against it. Simply put, GPT models make it easier to bullshit, and do it tremendously and efficiently.

I don't care if an AI writes something, but if a human uses AI as a co-author, attribution is required and necessary. You should just highlight whatever portion of your writing is you vs. the AI because effectively we are dealing with two separate entities at that point. You are not your AI, and when we will talk in person or offline, it will be painfully obvious.