If it's automated, it's always GDPR.
If it's not automated, it's GDPR under the condition that the data is part of a filing system.
So if you order your stone tablets alphabetically by their title, it's GDPR, but jumble a sufficiently large pile of stone tablets and you're in the clear.
In practice, this means that if you have a warehouse full of unordered boxes full of unordered forms, then somebody exercising their right to be forgotten cannot force you to go through every single box to see if there is data in there.
Conversely and frustratingly, if you have one gigantic folder of digital media, then you are technically required to actually go through that data, although I've heard of cases argued with authorities where this can be forgone in cases where it would be extremely uneconomical.
I'm still anxiously waiting for the first big decision on e-mail, for example. In large-scale corporate environments, good luck identifying every email containing personal data of a particular person, should that person ever exercise their right to be forgotten.
(Edit: no idea why you got downvoted, you raise an important point)
Finding out where that line is is probably going to be an interesting academic exercise which will result in lots of fines that could have been avoided easily: if you don't have a right to process certain data in an automated way pretend you don't have that right at all to stay safe.
After all, once the data is sufficiently disorganized to be searched efficiently it is also sufficiently disorganized to keep it secure and a data leak of disorganized data would be just as big an issue as gathering the data itself.
I don't recall the source at the moment, but one convincing argument I've heard was that an amount of disorganized data that you can organize given a few hours time would probably be treated as equivalent to organized data in the eyes of the authorities, otherwise you'd have a trivial loophole.