Ironic to use AI writing to complain about AI search results to advertise your product. But to the point, being an old school firefox user I still have separate address and search bars. My main search engine on my PC is 'google (web only)' which opens directly to the web tab which only has result links. I'm dreading the day they remove it as a feature.

LLM-written article.

I don't want either, if I'm indeed "searching." But I find that often times I am indeed just looking for a quick answer, and Gemini/Google's "new" search does it fine.

It's one of the few AI features, despite still being shoved in my face, that I actually find useful.

With that said, the worst thing is how search results have degraded significantly since the AI years, even before they added the actual "AI mode."

Google now (and quite a few search features on other services, e.g., Twitter) often returns results that have ZERO relationship to the search keywords I gave -- like an entirely different person when searching for a person's name, which I think should never happen and did not happen when search was still based on a "rigid" algorithm of indexed content. So, I can only assume it's because they have some AI thingy along the process.

I love AI summary and AI mode in Google search. I think it should be up to you (the human) to use your judgement to decide when to do further research by following the links and when to just rely on the AI summary. If I'm searching a TV show by name, I'm generally just looking for an overview, the IMDb rating and a review of that show. If AI summary gets that wrong, it's not the end of the world. I don't bother doing further research. There are a lot of such casual searches I do daily for which AI summary is good enough.

OTOH if I'm looking up an answer to a tax question, I don't just immediately trust the first answer from AI mode. I use it more as a knowledgeable friend who is not a tax attorney and so cannot be 100% trusted, but he/she is giving me useful pointers to go do deeper research and arrive at an answer.

Maybe it’s misplaced nostalgia but google search before google plus ruined the +, and the image search was top tier (not shopping ads) - truly incredible the signal to noise you could find.

Now all the junk comes to the top and the sites you get all have ads and modal popups or sales funnel flows

I disagree strongly with the premise:

> When you search for something, you're usually not looking for a sentence. You're looking for evidence.

There is a long and storied history of Google offering more than just a list of links to go search for, since at least 2012, because a massive amount of people literally are looking for the single answer to a question, whether stated explicitly or implicit in the search term.

Couldn't disagree with the article more. Not wading knee deep through SEO chaff is probably my favorite thing about LLMs. In the rare circumstance that I feel the need to wade through chaff that option is of course still available to me.

When LLM results started being shoved into standard result pages, I immediately added them to my ever-growing Userscript, CSS mods and uBO filters for https://www.google.com/search*.

That said, I DO sometimes prompt explicit web searches via LLM because for certain searches, an LLM can be the best tool for the job. It's almost always cases where it's hard to target terms with inclusions, exclusions and booleans (even if Google hadn't partially nerfed them). This is usually due to English language ambiguities in word meaning and LLMs are actually ideal for that because they already have discrete tokens for "Wear" (as in clothes) and "Wear" (as in tires).

My conflict with Google product managers arises because I'm an advanced search user and I know when an LLM is the best tool for a search and when it's not. And when it's not, LLM results are always cognitive noise that just gets in the way and eats screen-space. I understand that most users may not be able to reliably make that distinction but Google refuses to offer any option for users like me to enable a functional experience, even buried in an advanced menu.

I don't think this gap is just monetization driven. I've worked as a senior group PM in a FAANG-ish tech company and I suspect the problem is related to a deeper issue I saw in many UX designers and product managers - an inability to accurately model the frequency and quantity of ways "automatic" or default options can fail to serve users. It's an insidious issue because it tends to appear as the absence of signal in normal interface usage analytics. And it's hard to detect in focus group and observational studies, unless you aggressively look for it. And they usually don't because there's an innate bias pushing them toward believing the automatic "it just works" solution serves everyone as well as they've designed it to. They insist it must be right because it's exactly the sort of elegant, simple solution their D-School profs graded highly.

Totally get where you're coming from. Sometimes I just want straight answers without all the algorithmic guessing. Do you find yourself using quotes a lot to get more precise results?

I've switched to a local instance of SearXNG, no AI in results and I don't see things I search for in ads on other platforms now.

I agree with that but I think we are on the losing side. This approach probably works for more people than it doesn't work. Also:

"Clicking through now feels like expressing distrust in the tool, rather than just doing your homework."

This is off. Just between us, the "AI" does not have feelings. It does not care if you express distrust in it.

Kagi handles this perfectly, IMO. Defaults to normal search, but invokes an LLM if a `?` is appended to the query.

There will be a learning curve for some, like the person behind this article. Searching in AI mode will mean your query is going to need to be a bit more pointed.

The example given is you want evidence from an original source in Stack Overflow. So instead of just typing a few keywords and digging through pages, your query needs to ask Google exactly what you want and the format you want it in. If that’s a list of Stack Overflow pages, then that’s what you ask for. You can test this out now with AI summaries and a well-written query.

The quality of your results is going to depend on what you put into it. It will probably be annoying for some at first but for those that get it it’s going to be a step up.

You know, I've been investing in reference books since Google started failing to accurately report basic equations and constants.

It's kinda nice actually.

I can't empathize with this mindset at all. I'm in the polar opposite camp. I don't want a list of search articles; just a direct answer to the information I was looking for. I see the problems pointed out by OP but they are being solved away.

I use SearchTool in the Agent program, but when using search engines, I usually use simple keywords and the search results may contain too much noise, but I still use the AI search results as a reference, which is not an either A or B thing for me

I always ignore AI summaries, after having seen just how wrong they can be.

Here's a relevant example: https://news.ycombinator.com/item?id=44142113

Google AI responses are generally crap and annoying. I've gone back to DDG or if I need some context - very specific guidance for claude/chatgpt. Goodle's AI is generally inadequate or wrong without significant clarification.

I hate those AI summaries. Because I dont necessarily agree with suitability and credibility ranks assigned by AI to make those summaries. As author says, there are so many nuances and I usually scan the results page and click that appears more credible first. Not what appears first. I also know which site fills copy with verbiage and which ones give more useful advise - such as in health matters.

Great points..

If you're like me, basically everything other than search results in a search engine is noise. To that end, I direct my Google searches directly to the "web" tab of the results. This chops off news, shopping, video, images, etc from the initial results you see but you can still change to those tabs just as easily as ever. In your browser's settings just find the option to manage search engines, add a new one, and set the url to use to https://www.google.com/search?q=%s&udm=14 and you land on results. No AI misinterpretation of what you're looking for. No confident nonsense that directly contradicts reality sitting at the top of the page. No judgement or chiding (Google's results AI has flat out told me my opinion sucks mid-search more times than I'd like to admit). Just results.

pangram flags this whole post as 100% AI generated fwiw (and it reads that way to me)

why do people trust the AI overview over the Im feeling lucky button?

I've switched back to the pre-llm / AI summaries period when I extensively search for the results and also discover tons of great web resources along the way. The novelty of instant answers from the chatbot is gone. Now I just deliberately ignore the top AI summaries and go straight for the ten blue links.

If I need a chatbot I'll launch the app. Mixing the AI answers in SERP is too much noise for me.

I like "search assist" in ddg.

companies deliberately obscure some information and it helps.

  how much does "<product>" cost  

  what is the phone number for "<company>"

Isn't a search engine for finding information?

"Search" is a ridiculous thing to be doing post-2022. Imagine going to a doctor and asking them a question, and they give you 5 printouts for your to read through to synthesize your own answer. Imagine you asked your spouse a question and they responded "Here's 10 links for you to check out!"

We have AI now and it's doing a mostly incredible job getting us ANSWERS, not SEARCH LINKS. Trying to pretend that links are better is just trying to copy with rapid change.

Quite honestly I'm shocked that Google keeps making more money with search ads because I don't search anymore, I get answers directly from it or ChatGPT without clicking on any links.

Same. AI still makes a lot of mistakes. Like today I was looking for new smoke detectors, and it clarified and then forgot Washington state code multiple times, and then ignored alternatives that were better fit for my problem until I asked pointed questions, and kept recommending solutions that I absolutely didn’t want to consider (or couldn’t because of WA code). If it wasn’t a conversation, it would have gone from a great to a disastrous experience.

i just use chat for everything now. if i need a link, ill ask it to provide one.

well the "AI" is smart because the search is "DUMB" if I google "how long do I fry an egg" I don't need an article with 600 words about the ladys grandmums special egg recipe, but thats how SEO got us here.

Absolutely this.

I left google for chagpt/claude, but have slowly been coming back because these results have become rather useful -- especially when google inspects actual video content to produce replies. This is one of the best outcomes of the AI wave for me: content which was locked away in YouTube videos and forced me to consume it sequentially (without power-reading as I do with articles) is now indexed in search results in some AI summary. It makes mistakes, but it lets me figure out the right videos to watch, and often gives me enough context to find the solution myself. This applies to work, but also to something as simple as game walkthroughs!

As with everything, you need to know when to trust and when to doubt, when to iterate on chat (or chat with it with a smarter assistant like ChatGPT/claude). Hasn't it always been this way (minus the chat)?

I don't understand why people seem to be under this idea that just because it's at the top of a search results page it must be entirely correct, when we have spent decades knowing that you can't trust what's on the internet and you need to have critical thinking. Were you blindly trusting your first-page google results? What about when Wikipedia was starting, were you also trusting it blindly? Do you trust it blindly now?

In fact, when google started stealing the content of websites to display it, wasn't it already producing largely incorrect results because so many hits were SEO-garbage or outdated? (Not that we don't need to address the fact that this steals traffic from websites -- but it's a problem that PREDATES AI and absolutely still needs to be addressed somehow)

I can understand it if they believe the average joe will be more likely to blindly trust this, but that's an education problem -- and one we already had and has perhaps been compounded.

The visceral hate for a a life-changing technology that enables us to do more, faster, never ceases to amaze me. I'm growing so tired of it around my inner circles.

I get that there is a real bubble with real overhype and that some companies are profiting from it and shoving it down everyone's throats. I get that your actual life is worse because of this (you are forced to use shit models to produce more features while being paid the same). I get that maybe your job might change, and that for some people they will actually need to job-hop. This all absolutely sucks and it is a problem for many people, I get that! But that doesn't make the technology as a whole terrible, much less a bunch of its amazing applications. It's this constant wave of negativity, like nothing good will ever come out of AI and we're all doomed to be cornered by the big evil corporations that will run us dry.

It's hard to be rational when you're overcome with emotion and fear, I know, but it's so aggravating to have to interact with someone who can only see the technology itself as negative.