The quality of search results has drastically fallen off for my use cases lately.
It seems like the user categorization they used to do is entirely gone. Previously my searches would often bring stackoverflow or other programming forum posts to the front, now it's an absolute chore to find them if I don't add a specific "site:" or other "shibboleths" to the query.
It feels like they're trying to use language models to add context to the search terms, but because of the weakness of the model, it very often misclassified what my search concerns and shows the least relevant results possible. This also seems to make it much more likely to selectively ignore words from my search queries.
The "must include: <term>" on search results is exceptionally frustrating. I didn't add words to my query for fun, I really do expect every single word to be considered when I make my search.. why would I have wasted time typing it? Do they really expect me "to" "quote" "every" "single" "search" "term?"
Finally.. their "did you mean" query diversions have gotten much worse lately, again from what feels like total misclassification of my search. This happens quite a bit with queries that I used quite often and always previously worked.
They've clearly been making changes, and those changes, for my use case, have essentially ruined what the product used to be for me. I'm now actively trying my queries on other platforms, and as soon as I find one that does even slightly better than google, I'm out.
The funny thing about it is they knew what they wanted (to answer a question vs returning search query results) but didn't have tech like GPT to actually do it well enough to actually work. Scraping yahoo answers to feed their search engine didn't help things.
Now the difference for me on harder queries is Google will return low-quality fake SEO spam sites usually with .it domain and Bing and other chatgpt powered stuff will just make shit up.
If you turn on Google's verbatim search, it works correctly:
- Tools -> All Results -> Verbatim
I've made multiple comments before on this. Regular Google Search will literally ignore parts of your query or outright replace them with junk, Verbatim works like Google Search used to back in its heyday.
They need to let me turn this on permanently it is a night Vs. day better experience. If they keep this up I'll just switch to an LLM alternative.
PS - Verbatim search is NOT the same as double-quoting; it changes multiple rules[0].
Better because it's hard-wired to find pop song titles that consist mostly of stopwords? An above-the-fold Bing result for this search is "I am a total piece of shit". The Google results seem arguably more relevant, even though they are both pretty poor.
Your search query would be bad on most platforms. You're basically searching the index for "pieces", "of" and "me" separately. If that's a song title, you should put it in quotes. If you rewrite your query to do that, it starts giving actual results: https://www.google.com/search?q=%22pieces+of+me%22+is+actual...
I don't know what you expect from Google. I use this trick in DuckDuckGo all the time however and it's great for cutting through riff-raff.
Should they do better? Probably. The thread title is hyperbole though, and I wanted to point out that adding 2 characters to the query puts it's results on-par with the other engines.
Sorry, I forgot to include an interesting wrinkle: if you remove "reddit", Google returns what you would expect (sans reddit results).
I see several replies complaining about the quality of the search query, to which I say 1) almost none of the top results for the original query include any of the search terms, which should absolutely never happen, and 2) see above.
It should be said that this particular instance isn't directly indicative of Google Search's general quality - I'll admit to a bit of hyperbole in the phrasing of the submission. However, it is clearly broken here, and broken in a way that just this side of confounding.
It seems like the user categorization they used to do is entirely gone. Previously my searches would often bring stackoverflow or other programming forum posts to the front, now it's an absolute chore to find them if I don't add a specific "site:" or other "shibboleths" to the query.
It feels like they're trying to use language models to add context to the search terms, but because of the weakness of the model, it very often misclassified what my search concerns and shows the least relevant results possible. This also seems to make it much more likely to selectively ignore words from my search queries.
The "must include: <term>" on search results is exceptionally frustrating. I didn't add words to my query for fun, I really do expect every single word to be considered when I make my search.. why would I have wasted time typing it? Do they really expect me "to" "quote" "every" "single" "search" "term?"
Finally.. their "did you mean" query diversions have gotten much worse lately, again from what feels like total misclassification of my search. This happens quite a bit with queries that I used quite often and always previously worked.
They've clearly been making changes, and those changes, for my use case, have essentially ruined what the product used to be for me. I'm now actively trying my queries on other platforms, and as soon as I find one that does even slightly better than google, I'm out.