[aeturnum]

I do think LLMs have their place in search and I think the Kagi approach feels a lot better than Googles'. Kagi doesn't inject LLM results anywhere, but they've been making LLMs accessible in their search interface for a long while - this being the most evolved version of that effort. I am not totally sold on everything they are doing but I hate their integration of LLMs the least.

[lawn]

Kagi's auto summary feature when you add a question mark after your query is absolutely excellent.

It essentially summarizes the top search results for you, leaning in on a strength of LLMs (summarizing) while reducing its greatest weakness (hallucinations).

[aeturnum]

For me, search integration of LLMs also helps bring together the value prop of paid search. It's not just privacy, it's the skillful integration of a premium information processing service (with built in privacy).

[carlosjobim]

> Kagi doesn't inject LLM results anywhere

Kagi has AI generated stubs for some of the search results, probably originating from some of the search indexes they pay for.