Of the four decay mechanisms this is the one currently causing the sharpest failures, because the shift underway is unusually abrupt from the customer side.
The general pattern
A business finds a channel that works and builds around it. Over time the audience moves. The channel still works, technically: it delivers, it is measurable, and the numbers hold for a while because the business is getting a larger share of a shrinking pool.
That last part is what makes it hard to see. Declining reach and rising share look identical in most reporting.
Why the current shift is different
A growing share of consideration now happens inside generated answers, where a business is either named or does not exist. There is no gradual slide down a list. The transition from present to absent is abrupt from the customer side, even though the underlying decay was gradual.
So the usual early warning, a slow decline in position, is absent. The business is fine and then it is not in the answer.
The measurement problem it creates
An answer can be delivered without anyone visiting the site, which breaks the proxy nearly every business relies on. Traffic falls while the business is being recommended more, or holds while it is being recommended less. Neither of those is legible in a traffic report.
What it does not mean
Not that existing channels should be abandoned. A channel with a real audience is still a real audience. The error is assuming that because a channel still works it is still where consideration happens, and those became different questions.
How to detect it
Ask new customers where they first encountered you, and keep asking. It is the least sophisticated method available and it consistently outperforms attribution modelling for this specific question, because it catches the channels no analytics package is watching.
The second method is to check whether the business appears at all in whatever surface the market has moved to, which requires looking rather than inferring.