Just to add another perspective and use case to the mix, I use SearXNG extensively for research. Aggregating results is really useful for literature review because it saves so much time and reduces the chance that I miss relevant research.
You can randomize usage of public instances (“what about trust extension?”) but I’ve recently moved to using a single trusted instance.
Compartmentalization circumvents any worries, for me. Although the information is non-identifying anyway, the activity is already transparently reflected by my publicly available information, so fingerprinting is a non-issue.