This used to be solved by allowing queries like `Class Inheritance +ruby' to require results to include "ruby". They killed this for Google+ by changing it to quotes, so `Class Inheritance "ruby"' but now they interpret even those. When I use Google, which is less and less, I am not looking for a fight with a computer to express my intent, I'm looking for the answer to a question. That never seemed to be an issue until recently.
I work for Google Search. If you put a word or a phrase in quotes, we will only find things that have that exact word or phrase. Nothing has changed in this. When it happens that people feel it fails, it's often that they don't realize we've matched that word or phrase appearing in ALT text or text that's appearing in a less visible part of the page -- or in a few cases, the page might have changed since we indexed it.
> If you put a word or a phrase in quotes, we will only find things that have that exact word or phrase.
I'm sorry to tell you, but this is flat wrong. I commented[1] about this a few months back with a random phrase as an example. I see it often in my day to day also.
I searched for "eggzackly this" and all 10 results on the first page contain the phrase, although most have punctuation in the middle.
Looking at all 22 results (without opening the "omitted results" section or image results), the phrase became harder to find off the first page, but I tended to find it in the source code, or the DOM, or the cached version of the page, or by disabling JS (sometimes requiring a combination of those techniques). I found it in 21 out of the 22. The one page I couldn't find it in (the dolls one) looks like a frequently updating page, and it was highlighted in the snippet, so it looks like it was there at the time of crawling but not now; the cache isn't there for that page.
Full disclosure I work at Google, but not on Search.
Punctuation matters. I explained this in another response, and I should have mentioned it as part of my response here. But to repeat, these are typical reasons why it might seem that quoted search isn't matching when it is:
1) text appears in ALT text
2) text is not readily visible on a page (maybe in a menu bar or small text)
3) there's punctuation ("dog cat" will match "dog, cat"
4) page has changed after we've indexed it (so view the cached copy, if available)
In the [eggzackly this], you found matches of those words separated by punctuation -- which we interpret as a space, so the phrase is matched.
I wish we'd fix the situation with punctuation. I get that's confusing. But that's not a new change; quotes have operated that way for ages.
The article has been updated with a response from Danny Sullivan, the person you're responding to, which is worth a read.
I just tested "eggzackly this" and every result on the first page contains the string "eggzackly this", albeit usually with different punctuation ("Eggzackly. This [...]", "Eggzackly, this.", "Eggzackly! This [...]")
I'm sorry but this has absolutely changed. I'm not sure why but quite often we are suggested results in queries that ignore quotes. The engine is even telling us that if omitted those terms.
We don't have control over this and it's very frustrating.
We haven't changed anything. Promise. Honest. Not at all. But we definitely want to look into any cases where people feel this isn't working, so actual examples (if people are comfortable sharing) will really help.
What you're talking about is probably a case where there's a quoted word or phrase as well as other words that aren't quoted. In such a case, we're going to absolutely look for content that matches the quoted parts. That's a must. The other words, we'll look for them, but we'll also look for related words and sometimes, we might find content that doesn't match one of them.
Because those other words aren't quoted, we'll tell you if we find a match that seems helpful but doesn't contain those non-quoted words. That's what the message is about. But it should never be telling you we omitted a quoted word or phrase because we won't -- with one exception.
If there's literally nothing on the web we know of that matches a quoted word or phrase, then we're not going to show anything at all and say we couldn't match any documents.
I tried to find a counterexample and I couldn't! I believe you that quotes really are working. What's confusing though is that a quoted word or phrase often doesn't show up in the Google results snippets. This is certainly the reason why people think quotes aren't being respected.
Though, why does enabling "Verbatim" (in tools) on a search reduce the number of search results if all my terms are already quoted? Enabling Verbatim often does it make it feels like my queries are interpreted more literally, but if quotes are already being respected I don't understand how Verbatim would reduce the number of search results.
I agree, it would be easier if it were in the snippet. That's something we're looking at. I believe it used to, but sometimes the quoted part might not have been the best overall snippet to use. But as said, we might revisit that.
On the counts -- basically, it's all really rough estimates. We make a rough quick count, you go deeper into the page, we make a fresh estimate. It can change, and it doesn't always make sense and personally, I'd hope we just get rid of counts because of this, perhaps more confusing than helpful.
Interestingly I tried my pet peeve search, and it worked for the first time this year! It is for a specific recipe, and I search for 'ocau slow cooked balsamic beef'. I have had to manually find it in the archive of the overclockers.com.au forum for the past year, as Google seemingly forgot it existed no matter what search terms or operators I used.
The main difference is I am using Firefox on Manjaro and not my historically typical environment for searching. Normally I would either be using Chrome on Windows or Chrome on iOS if I am in the kitchen.
I will play around on some other devices and see what evidence I can find.
P.s. I'm being referred in for a role at Google currently, is the Search team only in a specific area like Silicon Valley or is it a global team? Most of the jobs in Australia seem to be commercial facing, not product facing.
Glad to hear that works! Search has teams around the world. I'd suggest if you see something relevant, apply even if it's in a particular location. Remote work has changed a lot things.
Hey, thanks for much for this comment. I've personally experienced this and looked myself for an example when I first read your comment, but I couldn't find one.
Today I stumbled upon one. Here's a broken example query: linux next hop "[::]"
This is not the case in my experience. I type a query with some parts in quotes and often get lots of results that have in small letters at the bottom something along the lines of “does not include <word in quotes>”, with no in bold highlighted part showing the phrase in the page context. This was not the case in the past and google made sure the word I put in quotes is absolutely mentioned somewhere
I’m guessing this happens when there are less results matching my phrase
I would love if you or anyone who ever has this happen can share an example, if you're comfortable doing so. We'll debug. But if you quote something, we shouldn't show anything but that which matches the quoted material.
Now, if you quote something and put in other non-quoted words, then we'll look for stuff that matches the quoted part and the other things are optional. So when you see that strikeout message, it means basically "We found this page that has the exact words you quoted, and it probably has one or more of the other words or related words you didn't quote, but heads-up, it doesn't have one of those non-quoted words at all."
And we do this because sometimes there might be a useful page that doesn't contain all of your optional non-quoted words.
Totally agree it would help if we did a better job bolding the sections of a page where the quoted terms apply. Often we do, but sometimes the snippeting won't include them if there's better text to describe the page overall. But we're looking at maybe improving here.
Unless it's changed in the past 56 minutes, the SERP is pretty up-front that it can't find any pages with all three of those.
It literally says `No results found for "kitchen boss" "g320" "disassembly".` right at the top, and then shows you the (properly explained) results for the non-quoted version of the query.
Mobile doesn't show that, at all. It also doesn't show that bit if you use the brand name "kitchenboss" instead of "kitchen boss", which I originally intended. My phone might have auto-corrected that without me noticing.
Anyway, with the terms I posted earlier, on desktop it says:
Results for kitchen boss g320 disassembly (without quotes):
It looks like there aren't many great matches for your search
Then it goes on to show me lots of unrelated results, instead of showing me the 'not many' great matches for my search. If my search doesn't turn up many good results (let alone great), that fine. Show me what you've got, let me decide how to broaden my search to get more. All of the unrelated results being shown are marketing spam sites. That's not helpful.
On mobile and desktop, I'm getting a message saying that when you search for ["kitchen boss" "g320" "disassembly"] that we have no results. And there's not much we can do if there are no pages we find that match all those words you required be present. There just aren't the pages.
What we can do is try searching for all of the words, so perhaps you'll find something useful that way. But even we can tell when doing this that the results might not be what you want, which is why that automatic warning about maybe these aren't great matches.
IE: we can't show you what we've got for a query where there are no exact matches. It's impossible. We can show you what we got if we don't require all the word be present. And we can tell you what we're doing. And you always have the choice to restart the query in another way if you don't like that.
Now here's something else. You probably used the quotes because I'm guessing you figured it was better to tell us exactly what to do than trust us to look at all the words and analyze the context and so on and see if we could make matches generally. If you had done that, just searched for [kitchen boss g320 disassembly], then the first web page result is the instruction manual for your sous vide machine. It has cleaning steps, which I'm guessing also might be what you're after? (It looks like it's just take off the outer casing).
Those results, doing it directly like that, are different than when we gave them to you after your quoted search failed. and that's probably because when you gave us quotes, and there were no matches, we might have tried some stricter matching to keep closer to the original requirements rather than use our general ranking.
To wrap up: maybe don't try the quotes at first. It's totally fine to type in a long natural language query like [how do I take apart a kitchen boss g320] and if you do that, the instruction manual is right there.
I added the quotes because I wasn't getting what I wanted without them. I don't need the user manual; I already have that and I'm not just trying to clean it. I wanted to find a guide to fully disassembling it, like the videos you find for phones when you want to replace a cracked screen yourself. Knowing that there are no matches at all is good information. (Disappointing, but good.) Trying to give me results for a different query, apparently assuming that I accidentally used the exact-match-only quotes, is not useful and kind of condescending. I think that's what leads people to think the quotes don't work the way they used to. Finally, the wording about not many good matches makes it sound like there are some good matches, but they've been mixed in with these other irrelevant matches. That's probably just a wording issue on the message, trying to soften the "We can't find anything" result.
Thanks for the additional information. It really helps understanding the situation.
Showing results for non-quoted words isn't intended to be condescending nor an indication that we think someone made a mistake. Apologies that it comes across that way.
We think we're clearly saying there are no matches with the "No results found" message. If we stopped there, the page would have nothing else. And I get that for you -- and perhaps others -- that might be preferable, a further reinforcement that there's nothing out there.
For others, not so helpful. They potentially might give up with no real way of going forward, when just losing the quotes perhaps could get them useful information.
And I get the trade-off concerns. I've seen many comments here that things done to support less "pro" users are annoying. And yet, we do need to find a way to support everyone. I think that's why we've probably gone with the message about no matches with and showing the quotes.
Perhaps we should consider making that an option -- "Would you like to try this search again without quotes."
As for the message about not good matches, we always try to show the best stuff we have first. That's the point of our ranking. The warning is meant to indicate that even though we'll list the best we have, for that query, none of it is particularly helpful -- not that here are a bunch of results, and there's some good matches mixed in with poor ones. Ranking that way would make no sense.
But it might be also that we're so close to it -- that we always try to rank the most useful stuff first -- that we didn't consider the interpretation you had. So thank you again, it's really helpful to get that feedback.
Sorry the information doesn't appear to be out there. I hope if you disassemble it on your own, you'll post the info out to the open web. I'm pretty sure you'd end up ranking well for that and helping others who might have a similar need.
Weird. On a computer, `"kitchenboss" "g320" "disassembly"` returns exactly six results, all of which appear to include the quoted terms. Plus the "It looks like there aren't many great matches for your search" message at the top. Which sounds like exactly what you want. I wonder why it's different on mobile.
So, besides the possibility that you are flat out wrong (as another commentator claims you may be), let's assume that you're right.
1. This is still horrific UX/UI
2. The culture internally at Google seems to have a "we know better than the users" attitude in all things.
3. Query rewriting is a horrific technique in general with almost zero value to life/society outside of fixing spelling errors. Whatever your A/B testing says about it's purported utility is polluted by Google's own dark patterns and political whims of the managers who run the internal search organizations.
It helps to actually know better than your users if you want to take the attitude in number 2. I don't believe that you or Google knows better than it's users, for many reasons previously enumerated in this thread and others.
One day Google search is going to be displaced and it's current utilization of query rewriting techniques will be one of the fundamental reasons for this.
You should take the absolutely massive amount of recent criticism and the fact that users repeatedly claim it's happening in the face of your claim it's not seriously rather than literally blaming the users writ large for a problem that is fundementally with the behavior of Google search.
I routinely see queries with quoted keywords where results don't have them highlighted in the snippets on the results page (but do have other, non-quoted keywords highlighted!).
That doesn't mean the quoted words weren't present in the content. It just means our system didn't think creating a snippet around those words was the most relevant snippet. Which I get, in some cases, actually would be better. It's something we're looking at.
Your system for snippet creation is so bad that according to you it's created a situation where many, many users believe that quotes don't work because of how bad it is.
Please fix it, like now. Displace your teams current sprint priorities, or the anti-google search backlash will turn into a situation where in 2025 Google is on the defensive for search market share from a Phoenix rising yahoo or something like that.
The fact that anyone at Google ever okayed this behavior at all in the first place is simply rage inducing and you should see that with the magnitude and persistence of the "actually it really doesn't work bro" kind of comments.
If a snippet doesn't contain the words I searched for, I don't click it, because I assume Google has fucked up the search again and given me some irrelevant thing. If that is what has actually changed in the last few years making people think results are bad, the snippet algorithm, please revert that. I want to see the exact context that the words I typed appear in the page.
In some cases? I'm literally asking for a specific word to be included no matter what; how could it be irrelevant to the quote used to describe a page?
Snippets we show tend to focus on full-sentences or enough context to describe what a page is about or relevant to a search. If we have a match that's strongest simply because a quoted term appears in ALT text or some obscure menu item, that probably doesn't generate a compelling snippet in how we normally would measure things. But given that for someone doing that quoting, seeing the quoted area might be the most important thing. So the regularly snippet process isn't as helpful -- and it's something we'll look at.
I don't have any recent information on how google search works, but years ago it looked at the expertise level of the searcher. So newbies received newbie results, advanced searchers received advanced results (and more visibility into filtering functionality). Today... they're hiding the advanced features and also seem to be reducing personalization of results to save compute resources. It's horrible.
You: Class Inheritance +ruby
Google: searching for "cash inheritance..."
I work for Google Search -- we never operated like this. We don't know that someone is somehow a "newbie" vs and "advanced" searcher and change (nor did change) the results somehow.
