Hacker News new | past | comments | ask | show | jobs | submit login
[flagged] Tell HN: Google search is getting worse
57 points by nothrowaways on April 2, 2023 | hide | past | favorite | 73 comments
It seems that Google is suffering from gpt influence, and started putting gpt everywhere.

I was searching certain words and the interface drastically changes based on what they think of the word. Then again, the auto completion is apparent hack of some gpt models in a rush.

It really used to be cool and stable, at least for me. Deep learning is not a magic hammer.




This is not even new. Ever since Google ditched the old algorithm and turned to semantic algorithm, search result is completely useless.

For examlle search for USB, Google will give me result of USA and then ask me back if I need infomation about USB.

And it's frequently ditching keywords from the prompt.

Sometimes I want to look for specific comments on some product for instance, I try different vocabularies that people might using exactly in the comment. But then all of these returns the same set of results repeated for 20 pages.

What is the point of Google search engine? Kind of trash to the point that it is now phenomenal.

Last thing is I noticed that image search is not even working recently. The average result is like 100 photos and done. Instead of infinite images in the old days.


This would be fine if they did a good job of semantic search. But they aren't particularly good at that. It's jumping to conclusions, ignoring parts of the query, etc. I find it infuriating when it just ignores part of the query so it can show you some stuff. If I type 3 words, there's a reason for it: they are all important to me.

It's gotten very frustrating to use because it tends to drown out good results with generic results from sites doing SEO optimization. There are some tricks around this. E.g. prefix any query related to AWS apis with stackoverflow or include some package names, use some quotes, etc. If you don't do that, you just get a 100 "this is why we believe AWS is so awesome" marketing to filter through. There are ways to mitigate that but they are quite tedious.

Bing on the other hand has rolled out gpt 4 based search. I've been using it and it's been pretty amazingly good taking over when Google fails me. Which it does frequently.

I particularly like it that I can ask follow up questions. It mixes answering questions and backs up the answers with links. I've been relying on it quite a bit since I got access a week ago. Very annoying that they make you use Edge for that but worth the detour from Firefox when Google is obviously not up to the job. Google is rightly worried about this.


It becomes particularly bad if you're looking for anything money can buy.

Lightest laptop? Here's are 10 lists of laptops that exist in 2023, with zero effort put into distinguishing them. As if SEO spam wasn't bad enough on its own, Google started ignoring your query to serve more of it.

Then there are specific queries that are reinterpreted as generic ones. You get 5 results or so, then it starts dropping keywords on a 2-3 keyword query.

It's infuriating.


I remember when we had explicit syntax for what MUST and MUST NOT be in the search results.

Now the + and - tokens are just a mild suggestion. Wrapping a word or two in quotes sometimes helps, but more often than not just returns no results.

It's funny, in middle school we had a whole unit on constructing search engine queries. It was really quite powerful back then. But I guess it's less profitable for you to actually find what you want without spending 20 minutes sifting through ads


But I guess BingGPT only look at the first 3 results and ignore the remainings. Speaking from my experience. It is kind of returning the same 3~4 results when I ask if there exist some paper about ML in the chemistry.


It's not even the change of algorithms. The results are filled with ads and promoted content, SEO spam, related search terms, images, news, and other nonsense. Long gone are the days when the SERP was a plain text list of the most relevant and useful results. Nowadays the page is designed to keep you as long as possible on google.com, or click through an ad.

Google Search today is only usable through frontends like SearX, and even then the results are no better than any competitor's.


When a site I click on immediately puts up a banner/etc. I back out instantly and try the next result.

Privacy concerns aside, ha ha, I wish a search site could rank results based on how long the user spends on the site. Most of the crap-sites that seem to get on the front page would disappear instantly from the rankings.


Using that metric has its own downsides though, depending on the content.

If I am searching for some particular piece of information, the best site is the one that has that up front, which means I get in and leave pretty fast.

Assuming that a user spending more time on a site is a good thing leads to large amounts of fluff (like with recipes...). But then again, there are some sites where that metric would be helpful.


The repeated results puzzle me, and it happens every single time. Show more results —> literally same results for pages! What’s the explanation for that?


I guess,

prompt: Hey ChatGPT! Here is a list of websites. Make a subset of it that looks like they are on the 4th page in the search result!


Double quotes are your friend. Not excusing how irrelevant the results have become, but when you anticipate this behavior, use the exact phrase method.


Double quotes hasn't worked in a long time. It used to be the case that if you search "foo" "bar" it would only return documents that had the words "foo" and "bar".

I will frequently search for z foo bar z, get documents that have synonyms of foo and bar (and the synonyms will be highlighted in the snippet), then I will search "foo" "bar" and get the same documents except now it doesn't highlight the synonyms in the snippet. Maybe I am in some awful cohort where this feature is intentionally broken to see if it reduces engagement.


>Double quotes hasn't worked in a long time

Same here, but seems to be 30/70. 30% of the time it works.

A while ago I was looking at an odd NetBSD issue and searched using "NetBSD". I got a few results on top, but most of the results on the first page were OpenBSD without any mention of NetBSD.


-"OpenBSD" as a negative quoted keyword in these cases.

Once you get your query dialed in, prepare to take the Turing test.


Make sure your quotes are plain quotes and not smart quotes.

They used to have a bug with those


What I can tell is quoting and verbatim is clearly broken. The accuracy is not up to what it was 5 years ago. For example searching for the exact sentence in a news article, you won't find the article.


They would be, except it ignores them sometimes.


If I search for USB google gives me search results for USB. I think you have a problem.


Searched for USB, not a single USA on the first page of results


It's a mataphor.


It's an "example".


A very bad example.


So false?


I have noticed the quality and usefulness of Google results declining for several years. I don’t think it has anything to do with recent AI developments. If anything, this new generation of AI may finally get Google to fix their results or accelerate its death spiral.

To me, GPT-4 is a breath of fresh air that directly finds what I am looking for and answers complex questions quite well even if I don’t entirely know how to ask them perfectly. People like to talk about prompt engineering to optimize GPT, but I find I can get good results even with very lazy, sloppy input.

On Google, I often have to mess around with quoting certain phrases, using -foo to remove certain types of results, etc. Google is more finicky; you have to adapt to it, because it doesn’t adapt to you. Actually, in many cases Google attempts to adapt to you, but does it poorly, which is worse. Google changes results by location, even if I deny it access to my exact location, and shows barely relevant results from an imprecise location instead of location agnostic results. Google also changes the results based on search history yet it doesn’t really understand how one query is related to the previous query, the way GPT does. There’s also no “New chat” equivalent on Google to start from a fresh slate, short of clearing your browser data.

On top of all that, GPT has a paid tier to better align our incentives and avoid ads. Even the free version has no ads for now. It is a better interface by far.

The fact that I have to log in to GPT is probably the most annoying thing about it at the moment. It does also hallucinate sometimes and it’s too much of a prude. There are concerns about its power and bias, etc. But it is clearly a step-change in technology. One which seems destined to improve, not get worse, as Google has.


It looks like for some reason it really got a lot worse recently - to the point where "regular people" who are far from being proficient with search engines and overall internet usage are starting to notice and complain about it - and it's not a surprise, it's hard to not notice when Google/YT search returns results COMPLETELY unrelated to searched terms.


Please pay for search so ads aren't needed.

Kagi is priced for sustainability and has features like lenses (e.g., search your own list of niche community sites) and prioritization preferences (raise arstechnica, lower buzzfeed) deeply helpful for technologist searching.

https://kagi.com

Friends don't let friends Google Bing on DuckDuckGo.


Kagi will be shutdown soon. The price increases from their data providers (google, bing, etc) and because each search queries a number of providers put them in a difficult situation. Now add in gpt and there market looks a lot smaller.


If you're only searching for family-friendly stuff Kagi is pretty good (anything NSFW yields no result), but there's no way I'd be willing to pay that much money for a search engine.

$5/m for only 200 searches (reminder, 1 page = 1 search. So if you hit "page 2" you just burned yet another search from your quota) is insane. [1]

Love it or hate it but Neeva is also on the private search engine market, and you can get unlimited searches for 50€/y. I find their pricing way more reasonable.

---

[1] https://blog.kagi.com/update-kagi-search-pricing


> (anything NSFW yields no result)

That isn't true at all from my experience. I just tested it and I was able to search for adult videos, etc.

I decided to try Kagi two weeks ago after hearing about it and trying their Orion browser on iOS. So far it has exceeded my expectations. Results have been on par with G/B/D. The new summerizer has been pretty good and I enjoy getting summary of the results at times to save me time (although I've had a few it couldn't summerize or wrong results in past two weeks). I also love the ability to prioritize sites for your future results and the summary ability of individual pages.

I will stick with the $10/mo for near future. It is small price to pay to support their current goals and intentions.


(Kagi founder)

Search has a cost and your searches have historically been paid by advertisers. Recently, also by VC money.

With Kagi it is neither, nobody is paying for your searches but you. Price also should not be the only vector of comparison - quality of results (which we prioritize and are known for) speed and privacy policy should be considered too.

Of course I understand if none of that matters if one’s budge for search is $5/mo. In that case it will have to be VC subsidized search as it simply is not economically viable or sustainable to offer unlimited search (at least of Kagi quality) at that price point.


I'm not rich. I can only put out so many fives and tens a month. I can't justify it when Kagi still falls far short of even pre-lobotomy Google and when Bing Chat consistently gets it right where Google fails.


Three days of Kagi had me never wanting to see a google page again.

Put your money where your mouth is!


this is worse than Google. I don't want to log into an account to search and have the search terms tied to my account


These are low priority objections.

This is a thread about search quality, and neither of your objections lower search quality.

Further, neither of the objections makes any difference in the practical world where Google ties the search terms to your account, and you don't have to log in to search, in the same sense that you don't log in to Google to have your search terms tied to you. After using kagi, then you just use kagi, and they don't save your terms.

Finally, Google has assertions in the TOS you agreed to about what they can do with your searches. So does Kagi. If you do care about search terms tied to you, you're welcome to read those and understand who has concern or liability for what aspects of your privacy.


this also means you cannot use incognito mode to quickly search as session details aren't.


Kagi doesn't save search terms by default: https://help.kagi.com/kagi/settings/privacy.html

Google is a lot better at keeping a profile on you whether you like it or not.


This post is low effort virtue signaling for HNers. Don’t get me wrong, I too believe SEO garbage is winning the current search engine arms race. But this post has no new information, not even an interesting take or even a source or proof for the vague complaints. It’s as interesting as a Reddit thread titled “DAE think republicans are the worst?!?”


I love hating on how bad the modern search experience is -- SEO spam has ruined everything.

But I'm a little confused by the exact wording of this post. Google's Bard is in competition with GPT, isn't it? Is it not? Was there a news item I missed about this?

I don't use Google often enough to have a strong sense of what might be different and what remains the same, but I've just run a couple of searches and don't notice any drastic interface changes?

This is not a complaint -- I just really want to know the new reasons to complain about Google!


I suspect "GPT" is being used by the poster as a generic term for LLM-based AI. It's becoming a generic trademark, similar to "kleenex" and "google" (search). I've seen it used this way elsewhere.


The thing I've really noticed is way worse lately is the "People also ask" section.

I'll Google for something and see the results are not very good, but I'll see in "People also ask" a question that is asking about what I'm looking for, in more detail and better expressed than what I asked.

It used to be that clicking that would almost always reveal what I was looking for.

Now it usually just gives me a result along the lines of all the bad results to my own search.


Good that I'm not using Google or Bing or any of the others directly but through Searx [1] since I have not noticed this downturn in quality. By now the arrangement of tools needed to traverse the 'net in a somewhat sensible manner is starting to resemble a suit of armour.

[1] https://github.com/searxng/searxng


Slightly tangental, I bought a new laptop last week and I was amazed at how bad windows has gotten. Half of the steps in the setup process appeared to be trying to get me to consent to some kind of tracking or another. Installing Chrome meant dismissing 3 or 4 different scary looking popups.


In the past, it was not easy to influence google results with seo bullshit. But now it seems like it gets easier as you see spam sites in every search result. I don't really care if poor quality sites even show up in results. But the issue of spam sites is a complete disgrace.


Yeah, that's one of the main issues. And it's about to get worse with the newer generative models.

I don't think Google ever had a great algorithm and switched to something worse. It's just the SEO actors putting crap out there now, and clickbait. The thing it became worse at was recommending small sites/blogs over big publications/sites.

Back in the day, all they had to face was people trying to spread viruses but those results were super obvious. And they weren't even flagged back then, just users were smart enough to instantly spot it from the result.


I don't think it's that hard. It's that spam sites benefit Google so there is no incentive to do anything about it, as the spam websites may contain Google ads/analytics and contribute to Google's bottom line.

Spam sites generally try to either show you lots of ads, make you click on affiliate links or outright sell you something.

This can all be detected and used as a negative ranking signal, so that all else being equal, a page without advertising/affiliate links/etc would rank higher than a page with those elements. This would allow non-spam content (if it exists and matches the query) to outrank spam content.


Yeah, I think incentives play a part. Similar to YouTube where a cesspool is usually good for them but not for you.

Given Bing and DuckDuckGo don't have the same incentives I wonder if their results are any better.


Bing and DDG have absolutely the same problem.

Microsoft has turned their entire ecosystem (including OS) into an advertising-saturated cesspool.

DDG might be "okay" for a while just because they're still trying to gain marketshare, but the underlying incentives are the same.

A paid search engine is the only solution that aligns the end-user's incentives with the search engine provider's.


Maybe at some point the complexity of their models has grown too complicated to understand and tune.

Google themselves don’t even know how their search works anymore ?


I worked in SEO for 6 years, from 2014 to 2020. I had a network of about 20 sites doing a little over 1M monthly visitors. During this time I religiously tracked search traffic and page rankings with a cumulative volume of about 150M. IMO, the biggest contributor to the decline in site quality is both SEO spam and the tools that google implemented to combat SEO spam.

When I first started the low hanging fruit was these so-called long tail keywords. long, low traffic search terms that represent a specific question within a broader topic.

"how to invert a binary tree with recursion in python" would be a long-tail of "how to invert a binary tree".

A fairly small blog could rank for these lower-volume keywords without much work or investment. during this time, a low quality spam blog would have a large collection of barely-releated articles targeting very specific searches. Their site structure would be something like:

computerlinuxhow2guide.net: - how to invert a binary tree in python - how to quit vim on arch linux - how to make an open world free-to-play video game in fortran - how to install docker on ubuntu 18.04 - how to build a flappy bird clone in react native for samsung phones

Webmasters looking to build a more sustainable business would snipe a few longtails when possible, but their overall content would be topical and more cohesive.

datastructureslut.com: - queues - build a queue using an array - build a queue using a linked list - binary trees - when to use a binary tree - how to invert a binary tree etc..

Google started to pages from sites with a lot of topical relevancy. Covering all the most popular search terms and interlinking these pages would give you a big boost. Affiliate bloggers responded by creating sites like "fishtankexpert.com" and "lawnmowerguru.com".

There are a LOT of these sites. Case in point: the examples above are just the first two things I thought of for $product $authority. and they both exist.

It became more common for larger sites to incorporate longtails into existing articles. Instead of writing a separate article, "how to invert a binary tree in python" could just be a subheading in a larger article. Tools like surferSEO became popular, before long it was common to see 10,000 word articles that covered thousands of search terms.

6 years ago, it'd be pretty common to find a forum about fishkeeping or a personal blog about landscaping on the first page of the search results for these kinds of keywords. But even then a lot of these sites would be abandoned or out of date. That made them an easy target for new webmasters to outrank, even with a limited budget.

The last major factor is profitability. With top 10 style articles, broad searches such as "best fish tanks" would bring in less sophisticated users. We'd often see purchases come in immediately after clicking on an ad. Most sales were one of the top 3 items on our list. Most visitors would skim the headlines, or read the top 1/3rd of the article.

specific Terms like "55 gallon saltwater fishtank" would bring in users who spent more time on your site, were more likely to read multiple articles, and more likely to return to our site at a later date. Overall, longtail traffic was more likely to make a purchase, and they would spend more money. Those sales often came in a day or two later, while broad traffic tended to purchase immediately.

Google never seemed to like longtail traffic. Not only because their pagerank algorithm was easier to manipulate, but also because they were less profitable for ad sales.

When customers could bid on highly targeted search terms that would reduce the number of people bidding on terms with low commercial value. In fact, if you set up an adwords campaign today you won't even be able to see traffic stats for these terms. They are obfuscated until you meet a certain spending threshold.

If you do optimize your bidding to avoid irrelevant searches, you'll get big warnings on your dashboard about "errors" they can "fix" for you so you don't "miss out" on all this "great traffic".

Since niche specific forums or blogs were basically dead, and these terms were almost entirely dominated by SEO spam, and larger "trusted" sites had started covering longtails in their broad content, google started to group topics and redirect searches to larger, more popular results.

The upside is that spam articles like "how to change the tire on a 1994 corolla" wouldn't easily outrank an article that has more general information about tire changing. There will be many popular articles targeting "how to change a tire", and "1994 corolla" is probably not mentioned on any of them. So google will assume that the uncommon portion is irrelevant.

When you have an error code, this strategy falls apart. The most common keywords in that search will be used in many different errors, while the portion specific to your error may only have a few results.

While there is still a lot of spam, these are not the low effort traffic grabs from a few years ago. These are sites that have invested heavily in content, back-links, and behavioral optimization. they'll be acceptable to a good chunk of the people that visit, and their advertisers will be more closely aligned with the results.

who needs search results anyway. If your question can't be answered by an advertiser or the knowledge graph (https://www.google.com/search/howsearchworks/features/) then it's probably not worth knowing.


I generally have to make extensive use of stuff like "-inurl:DONOTSEARCHHERE". I have a long string that I copy paste into the search location bar.


You can set a custom filter in uBlock Origin to do this for you - here's an example filtering out Pinterest garbage:

google.##.g:has(a[href=".pinterest."]) google.##a[href=".pinterest."]:nth-ancestor(1)

https://awesometoast.com/argh-stop-the-pinterest-results/


Pinterest is an odd case. It's not quite like usual SEO spam because it's immensely useful for what it is (collecting images and links while planning stuff, art references, etc). But then they use whatever sneaky tactics they use to flood search, so people who aren't Pinterest users think it's garbage. Pinterest probably doesn't need to do this to rank highly since it's actually useful. I don't know what they're thinking.


Can you share that string please?


uBlocklist does this far more elegantly. It's another one of those tools I forget about until I use a browser without it.


It has been the case for over a decade now. I used to start from page 10 when using other search engines, but Google is not any better now.


I have noticed a decrease in quality results ever since the appearance of SEO. Now that GPT tries to predict what you actually want to search for, and then shows you the websites that have most abused their SEO to beat the algorithm, it is that much harder to find accurate information.


Seo has had it's influence but has really ruined search is the guessing. Google nor respecting the search terms but "trying to be helpful" with symantics that it gets completly wrong. Even when using quotes on full phrases it still tokenises the phrase and doents search for the whole one. The Personalisation is another reason of the declining search relevance. Have to fight the search algorithm everytime to disregard local results or what is popular locally to get to what i really need.


SEO was a big thing in 2002. I built a nice business using it.


What’s the best alternative? It’s not bing. I have ddg as the default and I many times add !g - intend to end up using wuotes or adding specific sites to get more specific info. Most of the time I am using google to search for error messages in linux as part of debugging as a sysadmin.


I use Kagi these days and I'm happy to pay for the service. For error messages though GPT is not bad at all.


How about we bust out the CDRW's from 2005 and deploy that version. I could find anything quickly then.


Ddg is getting better. I find myself using !g a lot less than I used to.


In my experience, it's really just getting worse slower than google is. For the kind of queries that google already failed at years ago, DDG has also gotten worse.


Google hasn’t been a search engine for about a decade if not longer. It’s been a primitive askjeeves style chatbot. An actual search engine is going to allow crafting queries and so far as I know there’s nothing like that on the market since Altavista or early Google.


Yes, google is pretty bad. I scroll down a couple pages to find the useful stuff. And even then it often gets weird and stupid.

What's the good search engine these days?


I haven't been past page 1 of Google in the past decade. For those paying for search engines (eg. Kagi), do you have examples of what you're searching for that you can't immediately find on Google?


This complaint is frequent here, but nowhere else I know of. Examples given (if the are even given) are vague or different from what other people in the thread get for the same search. The only less interesting meme on hacker news is "I won't use anything by Google because they're going to shut it down".


Maybe the internet is just too big now. How can they search over trillions of documents using the same algorithm as before? Probably too expensive.

Let alone how are they affording to index and crawl all those pages, which is still only a tiny fraction of the entire internet.

Maybe search engines are a dying breed, and we'll have to rely on AI and word of mouth for information soon.


If I understand it right, it's not exactly the same algorithm. It's Pagerank, with lots more features and weights than in the past, run first. Then quite a lot of bolted on ML passes afterwards, each specialized for something different. One to suss out low quality content, one to identify low quality incoming links, one to boost the effect of brand/authority, and so on...many many more of these.

Which apparently worked well for quite a long time, but also dangerous over time because you don't really know why something now ranks well, or doesn't. The best you can do is have human raters score various searches, tweak a bunch of weights/knobs, and try over and over again.

Meaning, perhaps it's not that the internet is too big, but that the overall model is too big and disjointed to manage well. Also, if you're on this team, you're competing with the ad side that's continually pushing the organic results down the page, where they matter less. A lot of fussy tuning around "quaint hotel in Paris" seems wasted when the end user is 3 scroll events away from the organic results.


This would be consistent with the internet being "too messy", if not too big generally. There is too much garbage to sift through, and spam trying to game the algorithm.


I don't think it requires more CPU ops compared to GPT, or bigger bandwidth.


GPT was trained on mostly precrawled data and books/code. Go calculate how much bandwidth it takes to crawl the whole internet


You need to do it anyway, if you're in business of internet search. The difference between PageRank, GPT, or whatever else algo is how you arrange results




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

Search: