Hacker News Reader

Comment by quasse | Hacker News Reader

Kagi is not a panacea unfortunately. I pay for it and daily drive it to support a Google alternative, but I still have real trouble with my results being full of AI garbage (both image and text search).

As mentioned, product comparisons are a big one but another worrying area is anything medical related.

I was trying to find research about a medicine I'm taking this week and the already SEO infested results of 5 years ago have become immeasurably worse, with 100s of pages of GPT generated spam trying to attract your click.

I ended up ditching search alltogether and ended up finding a semi-relevant paper on the nih.gov and going through the citations manually to trying and find information.

View on HN

That matches my experience. Kagi doesn't surface much content beyond what Google/Bing do. What it does better out of the box is guessing which content is low-quality and displaying so that it takes up less space, allowing you to see a few more pages worth of search results on the first page. And then it lets you permanently filter out sites you consider to be low quality so you don't see them at all. That would have been awesome 10 years ago when search spam was dominated by a few dozen sites per subject that mastered SEO (say expertsexchange), but it is less useful now that there are millions of AI content mills drowning out the real content.

For content that isn't time sensitive the best trick that I have found is to exclude the last 10-15 years from search results. I've setup a Firefox keyword searches[1] for this, and find myself using them for the majority of my searches, and only use normal search for subjects where the information must be from the last few years. It does penalize "evergreen" pages where sites continuously make minor changes to pages to bump their SEO, which sucks for some old articles at contemporary sites, but for the most part gives much better results.

[1] For example: https://www.google.com/search?q=%s&source=lnt&tbs=cdr%3A1%2C...

> For content that isn't time sensitive the best trick that I have found is to exclude the last 10-15 years from search results. I've setup a Firefox keyword searches[1] for this, and find myself using them for the majority of my searches...

OMG. I'm so happy how much AI is improving our lives right now. It really is the future, and that future is bright.

Thanks guys!

Honestly, yikes. And it's not going to be better. Producing AI content is so cheap (and apparently so effective) SEOs are going to milk it forever.

I use Kagi personally every day and my results are definitely not full of AI garbage so would like to better understand your context.

Have you reported any of those issues to Kagi (support/discord/user forum)? We are pretty good at dealing with search quality issues.

Vladimir, when posting things like this, you might want to disclose that you're Kagi's CEO.

Hehe. He really should.

I’m not affiliated with Kagi in any way and my results are not full of LLM generated content either.

Though there is an LLM generated baby peacock pretty high in the image search, and when I go to the website it literally says that it is an example of an AI generated image and not a real baby peacock.

Most of the times I do, but it gets tiring as I post on HN a lot, and here I am primarly a HN user and have been one long before I've become a Kagi CEO. I feel that saying 'we' in the message (as I did) and having a clear disclaimer in my bio is enough.

The UK NHS website is usually pretty good for this so sticking "NHS" in the search terms might help, although I imagine they may not cover non-UK brand names.

"site:" namespace still works on Google... for now.

> I ended up ditching search alltogether and ended up finding a semi-relevant paper on the nih.gov and going through the citations manually to trying and find information.

I've been doing this for years now. The normienet as I call it is nigh worthless, and I don't even bother trying to find information on it.

I also use it daily. One of my favorite functions is being able to boost certain domains and block or downgrade results from other domains. So I boost results from domains I trust which significantly improves my results. They have a page with commonly boosted/blocked/downgraded sites which serves as a good starting point.

Reply on news.ycombinator.com