He tried Google, DuckDuckGo, Brave, and a self-hosted SearxNG. He went back to a paid search engine

A real A/B test against five alternatives

The post is short, but the experiment behind it wasn't: the author didn't just complain about search quality in the abstract, they actually tried Google, DuckDuckGo, Brave Search, Qwant, and a self-hosted SearxNG instance — running their own metasearch aggregator, the maximal DIY option — before resubscribing to Kagi, a paid, ad-free, subscription search engine. That's a meaningfully more thorough test than most "I switched search engines" posts run, and it's part of why it drew 184 comments on Hacker News.

Google came out worst in the comparison: "terrible" results, in the author's words, with the search experience increasingly weighted toward video and image results over the plain text they were actually looking for. "I just want to look up some text," as the sentiment is summarized — a complaint that reads less like nostalgia and more like a specific, current frustration with how much of Google's results page is now occupied by AI summaries, video carousels, and shopping modules ahead of the text link a user typed a query to find.

What actually won the author back

Three things: Kagi's result relevance, its built-in summarize-and-translate tooling, and — the detail that stood out most — the ability to restyle the search interface with custom CSS, a level of customization the author says they "couldn't reproduce even with userscripts/userstyles on other search engines." That's a small, almost cosmetic-sounding feature carrying real weight: it signals a search engine built for users who want to shape their own tool, not just consume whatever layout the vendor ships.

The thread's most substantive tangent asked whether search engines still matter when people increasingly just ask an LLM. One reply drew a clean distinction: "They are relevant to me, because I often look for documents rather than answers. I think the market segmentation is: Seeking documents → search engine. Seeking answers → LLM." Another, with a near-maximal karma-weight standing, put it more bluntly: "LLMs make stuff up all the time. Still. Anybody who's using an LLM for search isn't serious about the answers." A third pointed out the two aren't even separate categories in practice — "LLMs heavily rely on traditional search engines behind the scenes" — most AI answer engines are search engines with a summarization layer bolted on, not a replacement for the underlying index.

The subscription tax nobody in the thread resented

What's notable across the 184 replies is how little pushback there was on paying for search at all — a idea that would have seemed strange a decade ago, when "free, ad-supported" was assumed to be the only viable search business model. Multiple subscribers chimed in with their own reasons for staying: vim-style keybinds for results navigation, opt-in (not default-on) AI features, and search-result curation that lets users boost or block specific domains from their own results. One long-time Kagi user summed up the loyalty plainly: "I feel like half of what I do on HN is talk about Kagi. I switched to Kagi in 2023, and haven't looked back." Whether that holds as Google, Brave, and others keep layering AI answers into free search remains the open question — but for now, a meaningful slice of HN's search-quality-obsessed readership has decided paying directly beats being the product.