$25/mo is unfortunately a little steep because their lower priced option with 700 searches/month is comically small. On just my work browser I have 4000 Google searches last month. Doesn't count my personal laptop or phone. It's really hard to compete with $0/mo. for Google + AdBlock. I really do like Kagi better but $25/mo. is Tailscale + Notion + Spotify.
To me, the 700 soft limit seems absurd so I'm curious to hear how you interact with the product.
I average 50-100 searches per work day and 10-20 searcher per off day (totalling 1-2.5k per month). But I don't feel I'm using search in a particularly strange way.
For example, if I need to know how the new WidgetBean in SpringBoot works I'll usually end up making ~10 queries related to it before I move on to a new area of research.
E.g., I'll search "SpringBoot WidgetBean release notes" then "SpringBoot WidgetBean examples" or "WidgetBean test double", "WidgetBean in MockMvc test", "WidgetBean PowerMock known issues in MockMvc test", "WidgetBean FooService interation in MockMvc test SpringBoot 3".
Essentially, when I'm searching I go wide to survey the information landscape, then refine my search term as I discover what I actually need to know
This quickly balloons as I'm expected to know many things at $dayjob :)
What does it look like when you search?
Date (UTC) Searches
Apr 28, 2023 23
Apr 27, 2023 11
Apr 26, 2023 24
Apr 25, 2023 4
Apr 24, 2023 18
Apr 23, 2023 7
Apr 22, 2023 1
Apr 21, 2023 6
Apr 20, 2023 7
Apr 19, 2023 10
Apr 17, 2023 18
Apr 16, 2023 1
Apr 15, 2023 14
Apr 14, 2023 22
Apr 13, 2023 7
Apr 12, 2023 18
Apr 11, 2023 11
Apr 10, 2023 13
Apr 9, 2023 4
Apr 8, 2023 15
Apr 7, 2023 7
Apr 6, 2023 6
Apr 5, 2023 7
Apr 4, 2023 12
Apr 3, 2023 19
Apr 2, 2023 11
Apr 1, 2023 15
(UTC) Searches
Apr 2023 311
Mar 2023 495
Feb 2023 313
Jan 2023 434
Dec 2022 386
Nov 2022 396
Oct 2022 425
I typically start with one of the same type of queries as you did above, then I open every interesting-looking page from the first query in a nested tab (I use tree style tabs in my "research" browser) and start reading them, ctrl-clicking again to open interesting links in new tabs nested in a new level beneath, closing pages that was clearly useless and bookmarking in raindrop if I immediately sense that the oage I am on is something I will need again.
Once I have exhausted this tree, at least temporarily, I start a new search - like you do - in a new top level tab anf create a new tree from it.
To be fair this is why the $25 plan exists but if the economics are to be believed is totally unsustainable. For me 1.5¢/search is a really awkward spot. Fixed cost unlimited and GPT cheap pricing of $0.000002/token hit "use without thinking" status but at 1.5/2.5¢ I could actually rack up a rather large bill by accident so it causes me to use the service less.
And so I easily make 50–100 searches per day when I don’t even feel I’m actively searching for anything. And if I'm in a research mode, the count goes beyond that.
I noticed that only after I’ve subscribed to Kagi. After realising that I’ll burn throw the plan’s allowance very quickly, I tried doing most of my “dumb” searches via Google, and more important stuff — via Kagi. But that introduces a cognitive load, and kills the point of using a paid search engine.
Perhaps, I should switch to the $25 plan, but I can’t justify paying $250–$300 annually for a search engine, even though I have nothing against the idea of paying for one. Unfortunately, it’s too much for me. Nevertheless, I’m rooting for Kagi because it ticks all the boxes for me, and provides value. I hope that operational costs go down, and more people could use it.
But I fear they're going to have to get their prices (and relatedly, cost per search) down considerably before I can recommend them to anyone again.
I was fortunate enough to be grandfathered into an unlimited plan until March, but I don't think I know any professionals, e.g. people using a search engine to get work done, which search less than 24 times per day.
How can I know?
The magic is proven by the fact that Kagi mostly respect my queries (or accept my bug reports if I can prove they didn't) despite being built on the shaky foundation of Bing and modern Google.
Also there is some good old fashioned engineering there, like allowing me to pin, prioritize higher or lower or block certain sites.
DDG and Bing (no surprise there) fails at it too, but Google actually works.
haskell megaparsec operators -> should return [0] somewhere high in the result set. Found it on the third page of Brave Search. For comparison, it's the very first result on Google, second page for Bing, and the third page for DuckDuckGo.
Sure, it's a bit obscure, and I don't even write that much Haskell these days. But it's calibrated to determine whether I can rely on the search engine to quickly surface what I need to be productive, and whether I can trust its 'negative results' (I don't see what I want on the results page -> I need to refine my query or take a different approach). The version number returned in the URL also shows how well the engine handles keeping up with versioned documentation pages and aggregating 'link juice' between them.
I try this on every search engine alternative that pops up on HN. Very, very few pass. I find it often works like an adversarial example, yielding completely nonsensical results.
[0] https://hackage.haskell.org/package/megaparsec-5.2.0/docs/Te... (or any other megaparsec-#.#.# version)