I expect google to ignore what I searched for and change it to something more commonly found. Even setting it to Verbatim did not yield a correct first result:

Kagi used to be so so so good at accurate searching and in the past week or two it really seems like its sadly slipping.

I’ll definitely say the last few days I’ve had more “huh??” Moments, but I have not noticed the generally quality down.

I do think all the anti-AI walls are actually just extortion checks (fuck you Cloudflare and fuck whoever made the OpenAI scraper that repeatedly hits EVERY button and link on your whole website and has necessitated this whole thing) and it’s making legimate scraping tasks miserable.

I haven't noticed anything.

In your screenshot I'm not sure what you're expecting, and how the results don't fit.

The user expects the search for "girls" to return results for that, and not "gurls".

Ah!

With the relative popularity of the more recent song, and the commonality of misspellings either way, I'm not surprised at these results. When searching for a song title, you kind of have to use the artist also, or who knows what you'll get.

Searching for "california girls lyrics" should not return a page full of results for "california gurls lyrics" which is a totally different song.

The default behaiviour of a search engine should not be to assume the user is making a mistake. That's what google does.

It makes sense to me that the relative popularity of the recent song over the 70 year old one, would override the single letter misspelling. This is the result I would expect.

Searching for any song, you kinda need to include that artist. You can't just search for a random song name and expect exactly the version you want.

I cannot accept that assuming the user is making a mistake, when there exists a matching answer, is the correct way for a search engine to work.

I would say that it probably is to assume a correct spelling.
But when you have one set of results that are far more common (a couple orders of magnitude probably) it's reasonable to go with the more popular result.

Okay.
That seems like a stubbornly limited imagination.

I expect a search engine to be equally useful for finding the needle in the hay stack as the hay.

If every time you search for the needle the system gives you the hay, because it assumes you are making a mistake, how do you find the needle?

I completely agree, ideally I want my search engine to return pages containing exactly the words I searched for in exactly the way I typed them.

Though just to note, the words "gurls" and "girls" are both contained in the body of the text of the first result so I would say maybe it's not guessing but picking up on that.

Finding a needle in a hay stack is easy. This isn't.

Your analogy breaks because there is no needle.
Or it's a stack of needles.

The point is, the thing you're looking for and the stuff you aren't are all the same stuff.

In my metaphor the needle is the uncommon search and the hay is the common one. The way google works is to always return the hay and never the needle. If you don't understand how they manipulate the results to achieve this, I'm happy to explain it in more detail.

The point is the thing you’re looking for and the stuff you aren’t are all the same stuff.
It is easy. Just return the result that is the best actual match for my search term, as entered.. Optimizing for the user who enters the wrong search term is bad for everyone involved. There certainly is no case for Kagi to do it.

Manipulation of search results makes the most sense for a search engine that is selling ads, to drive up more placement.

Search results should not be manipulated. Just think about there that leads.

Search results have to be manipulated. Otherwise everything would return porn, or spam. Straight up character string searchs haven't been used online since 1995 if ever.

That makes absolute sense.

Just like a calculator shouldn't assume a "4+4" should be "corrected" to "4+1" just because incrementation is the most common arithmetic operation by far, a self-respecting search engine autocorrect what is in essence a fairly common search just because a more common one exists.

Sure, the words are in the body of the text but the title of the songs are clearly distinct. Maybe its a tricky case but I would explain that it must be ranking on something other than the actual title of the song.

When the search is for a song by title, that should be the primary match criteria, no?

We're not talking about spam filtering. We're talking about not prioritizing the title of a song as written, even in so called Verbatim mode.

As I said, you have a poor imagination. You have no understanding of how any of this works. Your just don't like the magic machine didn't read your mind.

Its busy at work now. I'm done here. Sorry

I'd say an ideal search engine should be as "dumb" as possible in order for it to be as user driven as possible, it's not an "AI assistant" of course so it shouldn't even guess that you're searching for a song by parsing the phrase you sent it, all it should be doing is pulling up all the pages that contain the words you search for and ranking them in some way. (It's like how you wouldn't want a calculator to try to figure out what you're trying to calculate to give you the answer it "thinks" you want.) You're right it's the ranking that's the problem, what's missing is that ideally there should also be advanced customisation for the user to control how pages are ranked or it should at least be transparent to the user; Kagi offers a limited version of this but from where we were 25 years ago I would have hoped it would be a lot more advanced by now (and we have Google to thank for that).