Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Kagi Bloopers – Search Results Gone Wrong (kagi.com)
195 points by embedding-shape 1 day ago | hide | past | favorite | 61 comments




Given the title "Search Results Gone Wrong", I'd like to take this opportunity to try to shame Kagi into fixing the search results for the "More results" feature. This simple feature is broken in that it often gives you repeats of the same initial results it gave. That is, instead of giving you "More results", it gives you a lot of "Same results".

I reported this as a bug about 6 months ago, and was quickly told it was planned to be fixed. But it hasn't been fixed. I checked in again a few weeks ago to see if there was any progress, and apparently they've given up because it is too hard: "Apologies, seems I forgot to update the thread. Unfortunately it is in fact trickier than it looks to dedupe these results. Mainly this is a result of how we work with results from upstream sources, and deduping is heavily complicated by caching issues."

Kagi, you're generally great. I'm usually happy to be a paying customer. But I refuse to believe that deduping a list of URL's is actually too hard for you. Maybe I'm one of the few users who actually cares about searching for web pages, but for my use cases my search results would be much better if you actually gave me more results when I click on "More results". How is this not considered core functionality for a search engine? Please fix this!

Here's the bug report: https://kagifeedback.org/d/7022-clicking-more-results-yields...


Huh. Yeah also in the FAQ they say that the reason why they return so few results is just because their ranking is so fantastic: https://help.kagi.com/kagi/faq/faq.html#why-does-kagi-return...

I feel like that is either hubris (they are overconfident in their ranking) or they have some other reason. Now you bring this up, and it seems to fit with the same kind of thing.

It reminds me of this article which brings up a bunch of suspicious things about search engines, and talks about how weird it is that so many engines limit how far you can go into the results: https://archive.org/details/search-timeline


Somewhat related, Reddit has been broken for me in a very similar way for more than a year now. Whenever I scroll down to load more pages, it will populate with about 80% the same threads as it loaded on previous pages. Over and over, such that by the time I’m on page 6 or so, I will have 6 of the exact same thread.

Really stupid bug that probably only happens with old.Reddit or RES or something. But it’s nice in that it keeps me off of Reddit I guess.


I can confirm that that bug exists even on "new" Reddit.

Common bug for caching lists that are reordering all the time.

Unlikely to be fixed though since it causes you to very quickly skip the repeat content (without having to serve you more than headlines) and see more ads. The "bug" multiplies the value they can get from each post which is a very important metric especially as llm slop has started to destroy perceived value from random posts from strangers (reddits only resource).

I'm not saying they introduced it on purpose the way Google intentionally showed bad search results to encourage a second query but I'm not confident that fixing it will be high on the priority list until it makes people leave the site.


“Pure numbers and French are not compatible”

Yep that checks out


Sixty-ten-eight! Sixty-ten-nine! Four-twenties!

1999 == One thousand, nine hundreds, four twenties, ten, nine.

I studied French in grade school over ten years and I love it. But the way numbers convert into language is wild. I tease it with love.


Switzerland and Belgium got them right!

As a French person who grew up going to a school in Belgium for a bit as a kid, I was quite amused by their numbers.

My thought as a 6 year old was "aw, are soixante-dix, quatre-vingt, and quatre-vingt-dix too complicated for you?"

Even now, while I think the French numbers make objectively no sense (even the countries that do count in 20s are at least more consistent than us), I can't help but find the Swiss and Belgian numbers "cute". Like "Baby's first 70 to 99".

And for whatever reason, I don't have the same opinion about 70-99 in English, Portuguese or Spanish.

Edit: just to be clear, I think my thoughts about it are absurd but they're too deeply engrained and decades old to shed completely.


It’s a well-known phenomenon that with the internet and modern media, large countries’ version of a language can affect the speech of the smaller countries using that language. Think kids in Portugal today growing up using lots of Brazilian words to their parents’ dismay, or americanisms slipping into UK speech. This makes me wonder if any young Vallon French speakers have started to pick up standard French higher numerals.

> Sixty-ten-eight! Sixty-ten-nine! Four-twenties!

https://www.youtube.com/watch?v=8Ze6ZMkT2Z4 :-)


US time: a quarter till 8.

= 775¢

"Four twenties and ten" is better than the Danish "five minus a half, times twenty".

But the history of why we stuck to four-twenties sort of makes it worse.

We were allegedly headed to sanity but l'Academie was like "actually let's stick to soixante-dix, quatre-vingt, and quatre-vingt-dix".


Good grief, it gets worse. It's half third [ordinal] times twenty, ½ #3 × 20.

That is so cursed.

I love it!


Not quite a blooper but I thought it was neat:

I searched Kagi for “veterans day 2025” the other day (on Veterans Day, when I was unsure) and it answered

“= today”


The first blooper seems to forget that time == money.

Just going to drop a quick complaint here that none of these are search results.

(though yes, they are funny)


People searched for something; these were the results. What else would you call it?

If we use the strict definition of organic results in SERP, these aren't the result of webpage indexation, they're the output of widgets and other natural language parsing in Kagi.

https://help.kagi.com/kagi/settings/widgets.html

https://help.kagi.com/kagi/features/search-operators.html#qu...


Those used to be called "instant answers" before every search engine renamed them to "AI overviews".

People prompted the search engine to search for something. These aren’t the results of the search engine searching for something.

More generally, these results aren’t “found”, they are conceived on the spot.


By that definition a 500 page would also be a search result. :-)

Heres one I found: search for "spaceweather" and you get weather for East Derry, New Hampshire. Definitely not space. The results I need are one and two for links below, but a friend pointed out that there is an astronaut (Alan Shepherd maybe?) who lived there which is the only connection to space I can think of for that city.

I guess for "Pop os" it gave a 2004 estimate for the population of the Cocos Islands. https://www.axl.cefan.ulaval.ca/pacifique/cocos-ile.htm

More likely the town of Os in Innlandet, Norway, which was around that population a year or two ago.

Very good! I didn't think of towns. I established that the alpha-2 country code "OS" is unassigned, and went hunting through the smallest microstates.

Is Kagi worth paying for? It's been on my radar for a while.

I’ve enjoyed it. It’s the first time I’ve not been left wanting by a search alternative. If it were to go away tomorrow, I’m not sure what I’d do for search.

The domain-block feature makes it worth it for me. Finally no longer having medium or userbenchmark pollute my search results.

You can self host searxng and block domains - increase priority on some etc. I do it myself.

https://github.com/searxng/searxng


This. Or quora or pinterest or twitter/x or etc etc

I can outright block domains or just adjust their weight. Great for my personal prefs but also huge with the family account and helping keep the BS out of sight for the kids without going full restrictive.


Just be aware that a small percentage of your money would be going to the Russian government: https://ounapuu.ee/posts/2025/07/17/kagi/

I try not to buy from US companies these days, but Kagi is really so good that I make an exception here, despite the US government getting some of my money.

The EU is still buying billions of dollars of fossil fuels and other resources from Russia.[0]

[0] https://www.aljazeera.com/news/2025/10/3/how-much-of-europes...


Nefarious actions by others shouldn't justify your own. We can do better.

It's a matter of scale. Objectively, yandex is a great resource, and Kagi's results would be degraded without it. Pennies per user go to them. The sum of the entire money that has ever transferred from Kagi to yandex is what? 30 seconds of EU oil and gas purchases?

In short: pick smart battles.


So Yandex is one of their data sources? Is that allowed under new sanctions?

also duckduckgo use(d) yandex. not many alternatives in this space

I use duckduckgo and live in a neighboring country, so I know Russian well (thanks, imperialism) and have to search things in it from time to time. It's still good at those queries, so this is just an excuse.

they used yandex up until 2022, and now they use bing and other

My wife and I got the duo package because we do a lot of writing and need citations and sources. Compared to google and DDG it is less noisy and returns fewer spammy pages. We’re giving it a year to see if it is worth it.

Absolutely, yes. It has completely replaced Google for search for me. Good AI search as well if you're into that (but they don't force you to use it!).

I think generally yes. I tried it out for free for a while, found it was substantially better than Google and DuckDuckGo, and paid for a subscription.

Recently it has not had such a strong quality margin, which I suspect is due to the AI slop that all of the search engines are fighting against (due to errors both ways in their detection). I'm hoping this is temporary.

To be clear, I don't use any of their features except search (and domain filtering).


I think so. I switched to it because I have YouTube tv - essentially Google as cable tv provider - and noticed how commercials became too correlated with recent Google searches for comfort. The only time I end up switching back to Google is for looking up local businesses reviews.

If you expect it to be as good as peak google results, no.

If you need something that’s very noticeably better than its competition, then yes.

If you are okay with all of the terrible that comes with using LLM services as a search engine replacement, then probably no.

If you despise the amount of second guessing and source checking required to use LLMs as search tools, then yes.


Have had the family plan for over a year. Worth it for me. Just so easy to customize.

Im happy with it. I have filters which will try to find search results from before 2022, which has greatly improved the quality of results for me

I wasn’t particularly impressed when I have the free trial a whirl - it’s not as bad as DDG, but it’s not anywhere near as good as Google.

Does Kagi have any value in the era of LLMs? My understanding is that it aggregates result from different providers.

Yes: you get reliable source information and don’t get inaccurate summaries. E.g. last week I used Gemini to answer a plant biology question and got two contradictory answers based on minor variations in the wording because it incorrectly relied on blog spam over peer-reviewed articles for the first query.

The initial false answer was baldly asserted by the LLM without sources in the first two paragraphs but some of the phrasing it used was enough to locate the non-authoritative blog content it was apparently laundering. Had it accurately cited sources, it would’ve been easy to see that this random WordPress site saying X wasn’t as authoritative as the PubMed hits saying !X.


You say that as if LLMs are a good thing.

You say that as if a technology can be easily classified as good or bad.

Touché.

Still, I think the down sides of LLMs outweigh the benefits for most use cases.


Kagi assistant is effectively a superset of other LLM chat apps.

Has access to kagi search which is a also a superset of search backends for the assistant


I second the utility of the Kagi Assistant. I didn't think I would use it much but now do so constantly. Especially because ending a regular search query in a question mark will cause the results page to lead with the Assistant answer! It's a delightful way to try both search and LLMs in one UI interaction.

Isn't that what all of the search engines do now by default?

They do it by default, Kagi lets the user control if and when they want LLM results.

If you don't understand the value of a search engine over an LLM, then you're not going to understand the relative value of different search engines.



Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: