I feel like the headline is torturing the data a little bit. Of the three primary buckets (a click, search session ends, or another search), only the "search session ends" bucket feels like it can really be construed to be not going to the open web. This is presumably because the SERP presented the answer the user was looking for, possibly because of an AI Overview answer, seeing a location on the mini-map, etc.

A search that then results in another search (22%) could either be the user realizing their terms weren't getting them to the right direction so they modify the search, or just changing their mind. The article notes that more research is needed here. Doesn't seem like they are "not going to the open web."

That leaves the biggest bucket (42%) where 79% go to a website, 29% go to a Google property, and 1% go to a paid click destination. The 79% is of course considered the "open web" here, but presenting Google properties as not the open web seems funny. There's no walled garden, login, or payment required to access those destinations, and they're usually presented in search results because of their relevance to the search. Even the 1% paid clicks bucket, while they are ads, technically still go to the open web.

So in general I feel like this is trying to paint a negative picture that the data doesn't really support. If nothing else, the definition of "open web" here feels overly selective, to the detriment of the message. But this is the Internet, so I suppose I shouldn't be surprised that the headline is clickbaity. :)