Dumping Google’s enshittified search for Kagi

For all of these things, Kagi has responded quickly and correctly. The time to service a query feels more or less like Google’s service times; according to the timer at the top of the page, my Kagi searches complete in between 0.2 and 0.8 seconds. Kagi handles misspellings in search terms with the grace expected of a modern search engine and has had no problem figuring my typos out.

Holistically, taking search customizations into account on top of the actual search performance, my subjective assessment is that Kagi gets me accurate, high-quality results on more or less any given query, and it does so without festooning the results pages with features I find detractive and irrelevant.

I know that’s not a data-driven assessment, and it doesn’t fall back on charts or graphs or figures, but it’s how I feel after using the product every single day for most of 2025 so far. For me, Kagi’s search performance is firmly in the “good enough” category, and that’s what I need.

Kagi and AI

Unfortunately, the thing that’s stopping me from being completely effusive in my praise is that Kagi is exhibiting a disappointing amount of “keeping-up-with-the-Joneses” by rolling out a big ‘ol pile of (optional, so far) AI-enabled search features.

A blog post from founder Vladimir Prelovac talks about the company’s use of AI, and it says all the right things, but at this point, I trust written statements from tech company founders about as far as I can throw their corporate office buildings. (And, dear reader, that ain’t very far).

The short version is that like Google, Kagi has some AI features: There’s an AI search results summarizer, an AI page summarizer, and an “ask questions about your results” chatbot-style function where you can interactively interrogate an LLM about your search topic and results. So far, all of these things can be disabled or ignored. I don’t know how good any of the features are because I have disabled or ignored them.