By default, we (I work for Google Search). do tend to look for all the words indicated. It's just that sometimes, what seems to be the most relevant documents won't have all the exact words. If you want results exclusively to contain words, that's what the quotes command is for.
> It's just that sometimes, what seems to be the most relevant documents won't have all the exact words
As a user I would prefer to be the judge of what is relevant. If the results aren't what I'm looking for, then I'll modify the query until they are.
Having the search engine make these opaque decisions on what it thinks I meant to search makes it feel like playing a game of Peggle, where I have no idea if what I'm going for will be what I get.
> As a user I would prefer to be the judge of what is relevant. If the results aren't what I'm looking for, then I'll modify the query until they are.
Funny you should mention that workflow, because Google highly values reducing the number of times users have to do that kind of iteration. If they can work out what you actually mean and give you that first time, they count it as a win and stick with that approach.
> Having the search engine make these opaque decisions on what it thinks I meant to search makes it feel like playing a game of Peggle, where I have no idea if what I'm going for will be what I get.
I think on average this kind of "semantic search" works out for the better, for most people, most of the time.
I tend to like precise results too, but I often appreciate Google applying its "magic," so I'm not sure Google running in a "all magic suppressed" mode by default would be better. E.g. I recently searched for help with an issue I'm having with my refrigerator. I searched for "Samsung BLAHBLAHBLAH drawer freezing" where BLAHBLAHBLAH was my fridge's model number. The search results included stuff for other similar Samsung refrigerators, which are all very similar, so the results were quite relevant to my problem.
> The search results included stuff for other similar Samsung refrigerators, which are all very similar, so the results were quite relevant to my problem.
Yes, but sometimes it's explicitly NOT relevant. Different models differ.
What's infuriating is that searching for "Samsung MODELFOO drawer" IN QUOTES doesn't even return the right thing. The spam bots win this race nearly every time. TFA gives several reasons but doesn't address the core issue that searching for something in quotes often only shows unrelated garbage that just happened to be SEOd to hell. To beat this "new solution", a page needs just to include a bunch of pixel images with all the samsung models in the alt text.
> I think on average this kind of "semantic search" works out for the better, for most people, most of the time.
Let's say that maybe before this "magic" was introduced, people needed to refine their search a few times but eventually got what they needed. But now, the majority gets a relevant result on the first time, but a few percent of searches don't go anywhere because you can't find a way to trick the AI and people eventually give up. You've improved your average metric, but arguably the search is worse since a few cases fail completely, when before they took just a bit more effort.
It's very interesting to read the responses to dannysullivan.
HN users need to understand: You are not a representative demographic. If they include results that only include words you typed, Google search will objectively degrade for the majority of non-HN folks. Even I sometimes find what I'm looking for because I didn't know the correct words, but Google presented me with docs with words of similar meaning.
As dannysullivan pointed out, Google is providing you with the tools to do what you want: Just put the words in quotes if it really, really has to include it.
My beef with Google is it returns too many results. I get that by including similar words in the results we'll dramatically increase the number, but I really shouldn't get 1M results for many things I type. Anything after the first 100 or so is almost always completely irrelevant to what I was searching for.
I fully appreciate the fact that my ideal Google is not the majority's ideal Google. I hope my original comment doesn't come across as wanting Google to serve me more than others. I just wanted to add my perspective, and maybe provide evidence that there is a market for a search engine that is more raw.
> I just wanted to add my perspective, and maybe provide evidence that there is a market for a search engine that is more raw.
In this particular case, perhaps just providing your own frontend that transforms the input to what you need with the settings you need probably will suffice. So make a form where this string:
Well google has a billion users, they’re not just optimizng for what you want. Their tests have shown that people generally prefer relevant pages over all three words so that’s what they show, but they’ve decided to include the quote command for people like you.
My original comment does not insinuate that Google should cater to me at the expense of others. I'm hoping just to provide one data point that there is a segment of users who are under-served with the currently available search engines.
I can accept google search doing some trickery where it goes: Okay, I found nothing, or the few things I found seemed irrelevant, however, if I replace one of your words with something similar, or e.g. I take this one word you wrote and stick a space in between (not arbitrarily, also based on 'I observe that this one word is usually typed as two words in all the pages I scanned) - then I do get results.
In other words, not a blanket 'the results for this query suck, but if I just stick my fingers in my ears and completely ignore one your words I get much nicer results so why don't I show you those?' - but a more intelligent presumption.
With, of course, the option to force the search to show specifically what you exactly searched for with no such helpful alternatives applied.
I think what you described is what they already have with the "including results for [similar wording]" version of the results page. They also have a link to see the exact results.
However, as of right now they only offer it for actual typos/spelling mistakes (like your example with the missing space) as opposed to semantically similar wordings.
> It's just that sometimes, what seems to be the most relevant documents won't have all the exact words.
It may seem relevant to google but it's not to its users this is why people are upset. If the result was relevant no one would care that it didn't have the keyword, please listen to your users.
