I think it's worth comparing search traffic to language model traffic. There isn't an easy 1:1 comparison (searches vs user messages? searches vs threads / conversations? single LLM queries often compose aspects of multiple searches in a single query) but site / app visits would still be interesting to compare. Google may not be falling in favor of other search engines.
It seems likely that people using special or small search engines are less inclined to leave for LLMs than people who use Google, and that's why their percentage is falling.