Hacker News new | past | comments | ask | show | jobs | submit login

If I had to tackle the notion of over-personalization in ~5 minutes, I'd say:

- If someone prefers to search Google without personalization, add "&pws=0" (the "pws" stands for "personalized web search") to the end of the Google search url to turn it off, or use the incognito version of Chrome. Personalization tends to be a nice relevance improvement overall, but it doesn't trigger that much--when it launched, the impact was on the order of one search result above the fold for one in five search results.

- personalization has much less impact than localization, which takes things like your IP address into account when determining the best search results. You can change localization by going to country-specific versions of Google (e.g. search for [bank] on google.co.uk vs. google.co.nz), or on google.com you can click "change location" on the left sidebar to enter a different city or zip code in the U.S.

- We do have algorithms in place designed specifically to promote variety in the results page. For example, you can imagine limiting the number of results returned from one single site to allow other results to show up instead. That helps with the diversity of the search results. When trying to find the best search results, we look at relevance, diversity, personalization, localization, as well as serendipity and try to find the best balance we can.

I saw Eli Pariser's talk at TED and was skeptical, although I did enjoy his example of Facebook starting to return only his liberal friends because he only ever clicked on the links his liberal friends shared. I had a number of concerns browsing through Pariser's book, but I would encourage anyone interested in these issues to pick up a copy; it's a thoughtful read.

I agree, I don't believe that over-personalization is an issue. (As I already said in another comment.)

But here's why some people probably don't like personalization: It's invisible. There is nothing on the results page that tells you whether your results are personalized or not. Sure, you can look at whether the browser is in incognito mode or you can look for some parameter in the URL, but these things require that you already know about personalization.

In contrast to personalization there are various indicators that a page is localized. The most prominent is obviously the language of the text. As soon as all the search results are in my local language it is very obvious to me that I got localized search results. I can also detect localization by looking at the Google logo on the homepage (localized versions have the country name in grey text below the Google logo), by looking at the language of the Google interface, by remembering the domain name I used to access Google and by looking at the sidebar on the left that even displays a guess of my location on the city-level.

There are no such indicators at all for filtered/personalized results. Every user around me starts with the same version of Google results. That's how everyone got acquainted with Google in the beginning. Same results for everyone. There is no reason for a user to question that until you see the differences by comparing search results, which most users won't. Someone who doesn't happen to work at Google or didn't hear about the "filter bubble" can't know that search results will start to diverge from vanilla results over time.

So personalized Google results violate the principle of least astonishment.

While I can't speak for other people, I think that the concerns that some people have about this are rooted in the fact that you can get trapped in some sort of feedback loop without ever knowing. If you could see that your results are personalized you could compare them to unpersonalized results and decide for yourself which you like better.

"But here's why some people probably don't like personalization: It's invisible. There is nothing on the results page that tells you whether your results are personalized or not."

Sorry, this is just not the case: we do provide indication on Google's search results page for personalized results. Here's a couple links that talk about how we surface whether results have been personalized: http://www.google.com/support/websearch/bin/answer.py?answer... is our support page and http://searchengineland.com/google-now-notifies-of-search-cu... is an article on Search Engine Land from 2008 when we started surfacing information on how results were personalized.

Here's a simple demo. Do a search in Chrome incognito mode and go to the bottom of the page. You won't see a link that says "View customizations." Now do a search in regular Chrome and check for that link. When I did a search for [matt cutts] in regular Chrome, I saw the "View customizations" link, clicking the link gives this message:

"Search customization details: matt cutts

When possible, Google will customize your search results based on location and/or recent search activity. Additionally, when you're signed in to your Google Account, you may see even more relevant, useful results based on your web history. The following information was used to improve your search results for matt cutts:

Web HistoryOne or more items in your Web History were used to improve search results. Manage Web History Remove Web History from my Google Account If you're curious, you can see what a search for matt cutts looks like without these improvements. The 'More details' link on your search results page can be used to display this page for approximately 30 minutes, after which it will no longer show this page."

In other words, not only can you tell whether a search results page was personalized, you can click a link right on the search results to see exactly what criteria were used to personalize the results. And that page has a clear link to run the search again without personalization.

As I mentioned before, personalization is typically a minor effect in Google's search results and it's almost always an improvement. But for people who are worried about potential "over-personalization," we do provide easy ways to see when a search was personalized, why it was personalized, and do the search again without personalization.

Thank you for the clarification. This comes as a surprise to me. I did not know that.

In my defense, I couldn't know about the "View customizations" link because I do have web history turned off, so apparently I never saw any personalized search results. After reading the DuckDuckGo page I expected that everyone's search results get personalized, especially if I am logged in with a Gmail account.

It's obviously not your fault that I didn't know about that, but, on the other hand, you can never expect from a user to know the contents of any help page. Clicking on "help" links is not what most users do. (imagine smiley face here, I don't dare to do that on Hacker News)

Additionally, I think that the "View customizations" link is a bit misleading, because usually customizations (in terms of software) are not automatic. At least I would expect that customizations are something that I do.

Also, the link seems to be placed at the bottom of the page, which means that 99% of the users are probably blind for it. (I can't verify where it is actually placed, because I don't see it.)

After all, I am thankful for the great search results that Google offers. Thank you for your hard work.

Happy to discuss this, jannes. Your points are well-taken: when we first launched the ability to see why/how results were customized, we added a link at the top-right of the search results (the Search Engine Land article has a snapshot from those days).

But there's another guiding principle that things on the search results page need to "earn" their pixels. Since personalization is a second-order effect and very very few people ever cared enough to click the link and get more info, eventually that link made its way to the bottom of the search results.

I'm sorry, but a "View Customizations" link is nowhere near as clear as a simple statement like "These search results have been personalized." right on top of the search results page, where you can't miss it.

But the real solution is to use a search engine that does not track you. Even better is to use it in such a way that it can't track you (ie. through a Tor proxy, while taking other reasonable precautions).

In case you missed where I said this below, we did launch a message on the top-right of the search results page. The Search Engine Land article had a snapshot, but here's a direct link to what it looked like: http://www.flickr.com/photos/searchengineland/2717951328/

Over time, we saw that people didn't seem to notice/care about the message and corresponding link much, so it eventually migrated down to the bottom of the search results.

It's interesting that the response to people possibly not noticing the link was to make the link less noticeable.

Instead, you could have tried to make it more prominent by (for instance) moving it to the upper left rather than the upper right of the search results, right above/below the ads.

Another issue that might be interesting to explore is to what extent users really understand what search customization is, and whether they'd care more or less about it being done automatically once they understood it better.

I have a feeling the vast majority of them probably wouldn't care, and take the attitude of "do whatever it takes to make the results you return more relevant, and I don't really care how."

If the flickr image is the actual size, then no wonder it was not noticed. No matter how much I customize all my interfaces -- and with the increasing pixel count of displays -- interfaces are constantly populated with immutable 8 point fonts. Any font less than 14 points is fine for 1985 and VGA displays; but not anymore.

After knowing that there was a "View Customizations" link, it took me > 1 minute to find it. It is in the most unintuitive place where 99% of the time I don't even scroll to. Sorry this is in no way advertised.

Google also filters special terms like bittorrent in instant search. This is part of that bubble and people don't even realize it. Thats the point, that most people won't notice, not that the views are not there.

Its like experts-exchange.com putting content below the long footer of the page, yea they can claim its there but many won't notice.

There's also this http://i.imgur.com/PMD5U.png which I hadn't noticed before. I just clicked through DDG's links and it tied guns and Obama for me, pretty cool.

Here is a slightly different question, how do you get "no country redirect" to stick reliably?

I cant begin to describe how annoying it is that I am presented with a different language when travelling to a different country. All I ever want is Google in English but it keeps going back to a localized search regardless of the many times that I choose "google in english"

Good question. I just saw an expert in the hallway and asked him. The basic answer is that your preference is stored in a cookie, so the preference would be forgotten if you're clearing cookies. If you still have the same cookie and yet the "no country redirect" isn't sticking, that's a bug we could dig into.

By the way, I asked why the "no country redirect" isn't stored with your Google account rather with a cookie. The main reason he gave was that whether to do a country redirect is one of the first things we decide, and it's faster to use a cookie for that than to go looking up the user's account setting. Or at least, cookies have been faster up until this point. Hope that helps explain things.

I regularly see this without clearing cookies, so yes, I would consider this a bug.

There seems to be no discernible pattern as to why it works in one session, suspend laptop, go somewhere else and it stops working. Or it'll work twice in a row and when I return location a (both in the same foreign country), it stops working.

Interesting. HN isn't ideal for debugging, but if you wanted to send cookies or IP addresses (e.g. via Twitter), I could see if someone could look into it.

Use: http://www.google.com/ncr to disable. NCR=no country redirect. I've set this as my homepage and not had any issues

FANTASTIC TIP! This drives me (and 3 co-workers who will want to buy you a beer) insane - I travel a lot, and use proxies - so google is constantly swapping languages on me.

Though it makes me less worried about them collating all my personal information if they can't figure out I didn't suddenly learn to speak German ;)

There does not seem to be a way to use Google SSL + NCR. E.g.


does not seem to be a valid option. Is there a way to pass NCR in a param like ncr=1? Or the fact that Google SSL is used implies use of NCR?

EDIT: It looks like &ncr=1 does work but I'm not sure if this is equivalent to no country redirect.

Right now, this is what I have set up (which also disables personalization)


Cant easily do this as i am currently in country. Can you shoot me a contact email (mine is HN username @ gmail) and I can send you something when I'm on the road again and can send you cookies/IP

It's a shame how Google completely disregards your HTTP Accept-Language header. If they didn't, this would be much less of a problem.

This annoys me more than anything - not just google, but for a large and growing number of sites. They completely ignore your browser settings and select language based on IP address. I installed the google international search plugin from mycroft which has solved my problem with google - but still suffer the myriad of other sites that ignore my browser's config.

I think in the past we saw a lot of people with their Accept-Language header set wrong, which is why we haven't used it. But we've been having a good discussion internally about the "my language won't stick" issues raised on this thread.

How practical would it be to initially trust Accept-Language but also prominently display a link to change to the language detected through IP geolocation (with an easy way to hide/decline the link)?

I suppose you know about http://www.google.com/ncr ? It works some of the time.

Some of the time it doesn't, e.g. now when Google has custom logos I'll get a search term in Dutch (I'm in The Netherlands) when I click on it, even though I'm using google.com in English when doing so.

Yes, I know about about /ncr but its hard/impossible to set that on a mobile device

Also, it only works sometimes.

As yet another aside, has anyone else noticed how it is pretty much impossible to set SafeSearch to off without being logged in?

Trying to set the cookie just does not seem to stick

If you want, you can set "&safe=on" or "&safe=off" at the end of the url.

I reported this localization problem years ago. The best way is to take language from browser headers, like a lot of other sites do.

For some reason responsible googlers were ignoring my proposal.

I don't want to jump so many hoops just to do one damn search. Personally, I hate the country specific personalization, and perhaps it's useful to most people in my country rather than showing more US results, but I'm really not interested in those types of results myself. I wish I could just check a box in my Preferences and then be able to see the universal search results.

I know I can use /ncr at the end, but I'd rather not have to do that all the time, and I think you can't even make that the default search for the Omnibox in Chrome, which means I have to give up Omnibox, which I love using, in order to get away from personalization. And even then, I think it just means I won't see my country specific results, but it probably still personalizes my search results through other types of signals.

So fine, don't make universal search the default way to search, but just give me a checkbox so I can turn it on when I want to. I want to see the best results, period - not the best results for me (or whatever Google thinks are the best results for me).

"I want to see the best results, period - not the best results for me"

But aren't you a part of the relevance equation? The ideal results for a search like [bitcoin crash] should be different for a Japanese-speaking searcher in Tokyo vs. a German-speaking searcher in Munich vs. a bitcoin expert vs. a programmer trying to diagnose why compiling bitcoin is crashing vs. my Mom who has never heard of bitcoin before, right?

But why on earth do you persistently try to present German search results to an English speaker in Munich? This has been peeve #1 for years, across all Google services.

Good question; one person is annoyed by this endlessly a few offices down from me. It's hard to make sure that things are handled consistently sometimes, but if you use google.com/ncr or the "Google.com in English" link at the bottom of the home page, that should help. The "Language tools" link to the right of the search box should also let you set a cookie with your language preference.

And yet for some reason, the search link from Google Toolbar seemed to ignore that cookie and certainly ignored my account preferences, sending me to the localized search page regardless.

Short version of long story: I used to bounce my net traffic through an ssh tunnel to a hosted VM. The VM was moved to a new machine in south-east asia. Having Google Toolbar constantly send my searches to the localized engine, despite the cookie selection, my being logged in and my preferences being clearly set, was more of a day-to-day annoyance than having my traffic piped across the Pacific twice.

I appreciate that you went and asked someone for the clarification that this issue is about the preference being set in a cookie and not in user settings, but this doesn't solve the problem for many of us.

For YEARS, on a weekly and sometimes daily basis, always logged-in, always with preferences set to english, it is infuriating to routinely end up receiving results that conflict with your explicit language settings.

What, concretely, do we have to do to get someone at google to push a change from using cookies to a real user-setting to fix this absurdity?

You said they did it for speed. Giving me completely incorrect results in a language I can't even read 1 millisecond faster than giving me results that I actually care about is a win? This is over-optimization.

Exactly, and it's cross-device, cross-service. What about browser built-in search? Or mobile? Or not wanting cookies or having to log in? I see my browser having a preferred language setting...

Just add a new search engine with the special url and make it the default as opposed to the pre-baked search engines.

You cant easily do this on a mobile device.

Personalization tends to be a nice relevance improvement overall

I agree completely. Reading that page, my first thought was "if (not saying I don't) Egypt is a place I probably don't plan on going to, why should I waste time looking at those links in my search results?"

Overall, I think personalization (wow, this isn't a word?) reaches its own form of market efficiency. If the personalization algorithms are bad, then people would shy away from the search engines that provide them. However, if they make sense, and return the most relevant results most of the time, then that's saving us a lot of time.

Do you have a pointer to a publicly visible reference to the definition of some of the other query string parameters Google uses? Just thought I'd ask, given the context. I've noted various references people have cobbled together, but perhaps there's something a bit more... "canonical". Thanks for the above.

http://code.google.com/apis/searchappliance/documentation/61... is a reference for our search appliance, but a lot of parameters are in common. I'm not aware of any other official Google doc available externally, but if you search for [google search url parameters] the top 3-4 results are all quite good.

If someone prefers to search Google without personalization, add "&pws=0" (the "pws" stands for "personalized web search")

"Append a cryptic query param" is a terrible user interface, and it's a little silly to suggest this as a solution. Make it an option real people can discover and use.

when it launched

Just curious, why add this qualification? Is it significantly different now?

the impact was on the order of one search result above the fold for one in five search results.

That statistic needs a little clarifying. Query terms frequency follows a power law distribution, doesn't it? So if you're counting each individual term, the fact that 1 in 20 have altered results could very easily still mean a majority of actual searches are altered. And depending on how you calculated the 'one search result above the fold' number, it could very easily still mean that when a page is altered, it's altered significantly.

More interesting would be knowing these stats for just the fat head of the query term distribution.

Just wanted to say thanks to Matt_Cutts for addressing many relevant issues/questions here on HN. Your responses are appreciated and not always taken for granted.

A much better alternative is https://ssl.scroogle.org , which doesn't have personalization since google can't tell scroogle users apart. It also has benefits like having a bit of privacy while you search.

I think Scroogle hits Google from a relatively small set of IP addresses. Be aware that Google is probably trying to localize for those IP addresses, so your results could be less relevant (in the same way that if you searched through a proxy in Germany, you'd be more likely to get results with a German emphasis).

Personally, if I have a search that feels a bit sensitive, I just hop into incognito mode in Chrome. Control-shift-N is an easy shortcut to open an incognito window in Chrome. And don't forget that you can use https://encrypted.google.com/ to do a search via SSL as well, which provides an encrypted tunnel between your browser and Google.

Your reccomendation is hilarious because it still does not give the user any privacy. Neither private browsing mode nor SSL will give the user any privacy from google, which is the real concern. You still log searches and IP addresses for an extremely long time.

One solution that actually does solve this is to use Scroogle or to access google through Tor.

EDIT: I feel that i should clarify that of course SSL is a very important feature to have for a whole slew of reasons and that i'm glad Google supports it, just that it wasn't relevant to my point.

"Your recommendation is hilarious because it still does not give the user any privacy."

I'm a big believer in prioritizing actual issues over perceived issues. Using SSL search prevents bosses, ISPs, and governments from sniffing your queries, and I consider those to be the largest threats to your privacy. Some ISPs sell their customers' query data and surfing patterns, for example. In contrast, when the Department of Justice tried to subpoena two months worth of user queries, Google resisted that challenge and won in court. Having worked at Google for 11+ years, I know that my colleagues care a great deal about our users' trust and privacy and work to protect it with features like SSL search, two-factor authentication, warnings when sites might be hacked or hosting malware, etc.

If you're that worried about Google, don't use it, but if you still want Google results but with as much anonymization as possible, I would choose Tor+incognito-Chromium instead of Scroogle for your searches.

> ...the real concern.

Ironically, "the real concern" is personalized. The married guy searching for "hot local hookups" doesn't care what google knows, he just doesn't want it to show up in his browser's history. The junior high student reading The Big Book of Mischief in the computer lab doesn't want his school's network monitors to find out. The dissident in Tunisia doesn't want his government monitoring his Internet usage.

There's only a small set of privacy-conscious Internet users who should be concerned about Google, whether or not they're as impregnable a bastion of privacy as their employees might claim.

I had to stop using encrypted.google.com because it does not provide a link to switch to image search.

Which in turn, incidentally leads back to DDG : ).

Crazy good stuff has been going on with DDG the latest couple of months, I'm just glad I've finally found a search provider who takes these issues seriously!

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