I assume that Kagi Quick Answers are generated with an LLM and the search result previews. I realised that if the search result previews themselves are generic LLM summaries, the Quick Answers might be worse, because they have less specific information to work with.
For example, when I searched aerogard or "off" more effective? because I had to make a quick decision in a supermarket, Quick Answer got to use this Reddit result preview:
Can anyone recommend a mosquito repellent that actually works ...
[Jan 3, 2022] Also worth considering is tropical strength “OFF! ... I'd forgotten that one - a much better smell than Aerogard, IMO, just more difficult to find ...
That information about the smell was really useful for Quick Answer to use. If this search result preview had been procsssed by an LLM, that specific information probably wouldn't be available.
Overall, I think the LLM summaries are occasionally useful, but more often than not, they're worse than the default summaries. Unless Kagi can selectively use LLM result summaries only in good contexts, it would be better to avoid them entirely.