Why Google “Personalizes” Results Based On Obama Searches But Not Romney

Do a search for “Obama,” and you might find your subsequent queries for things like “medicare” seemingly personalized based on your initial Obama search. Do the same for Romney, it is doesn’t happen. Is Google biased toward Obama, as some are now writing? No. Is it the “filter bubble” of personalization out of control? Perhaps not. In fact, it doesn’t even appear to be personalization at all.

The news of how searching for Obama can influence your subsequent search results came out in a Wall Street Journal story today, which conducted a battery of tests to explore the phenomenon. This followed on a tip and similar testing that Google-rival Duck Duck Go conducted. The Journal found that a search for Romney didn’t trigger the same changes.

The bigger story, by the way, isn’t that searching for Obama might influence what someone might see in future searches. Rather, it’s that searching for certain terms — Duck Duck Go likes to call them “magic keywords” — may influence future searches. I’ll explain this more as we go along. But let’s start with the Obama test.

Obama Search Influences Other Searches

It’s easy for anyone to do the Obama test. I’ve been able to confirm it multiple times, and whether I’m signed-in to Google, signed-out or appearing as a completely new person with no prior search history by using “Incognito” mode in the Chrome browser.

Search for “obama,” then search for “iran,” and you’ll likely see a news story about Iran with a message “You recently searched for obama” appearing below it:

You might not see exactly the same news story. But there’s a very good chance you’ll see some type of a news story or link with the “You recently searched message.”

Search for “obama” followed by “medicare” or “gay marriage,” and the same thing happens:

Now do the same thing by first searching for “romney” and then terms like “medicare” or “iran” or “gay marriage,” and this type of customization doesn’t happen.

Why It’s Not An Obama Bias

The first thing to clear up is that this isn’t some type of favoritism for Obama. That’s already starting to spread. The Journal said:

Google Inc.’s quest to guess what we want before we want it has produced an unusual side effect: a disparity in the results the company presents about the presidential candidates….

The search links are altered only for a short period, and there is no indication that Google is intentionally biasing its results….

In the hands of a human, decisions like these might be viewed as biased. For a Google algorithm, they are simply a matter of numbers.

The emphasis is that there’s some type of unique thing happening following Obama searches but that this type of bias isn’t something Google’s doing on purpose. It’s just part of how the algorithm works.

Bias is already a charged word which gets turned into Obama being “priortized” by The Verge:

Google shows personalized search results that prioritize Barack Obama for users that have recently searched for the President, but Mitt Romney isn’t getting the same treatment.

Over at CNET, it’s seen as unfairness:

Google isn’t treating searches related to Presidential candidates fairly, a new report charges.

Every search engine has biases that are a natural consequence of trying to rank web pages based on a wide-range of factors. But by no means are these results somehow doing something favorable for Obama over Romney.

It’s not like when you search for terms like “gay marriage” or “iran” after searching for Obama that you only get positive Obama news or news that somehow “prioritizes” Obama over Romney, whatever that’s supposed to mean.

Nor is Obama somehow getting a special advantage. Consider this listing you get for “medicare” immediately after searching for “obama,” one that does not appear after searching for “romney:”

If you’d previously searched for “obama,” you got this article suggested he — along with Romney — are wrong on Medicare. But if you searched for Romney, the article doesn’t appear, so Romney potentially gets favored by not having something negative about him appearing.

Why It’s Not A Filter Bubble

The next thing to clear up is the idea that this is some type of “filter bubble.” The term, popularized by Eli Pariser excellent book The Filter Bubble, means that each person is trapped within their own little bubble of personalized results, seeing things tailored only to them.

What’s happening here is almost the opposite. In the Wall Street Journal’s testing, about 80% of the people who had searched for Obama and then the subsequent terms saw some of the “You recently searched for” results appearing. Rather than them all being stuck in a bubble, most of them were actually having pretty much the same experience.

The caveat here is that I don’t know that they were all seeing the exact same news stories appearing. Moreover, Duck Duck Go’s testing suggests that the news sources themselves might be different. For instance, while everyone might get customized news links, one person might get a Fox News story while someone else a Los Angeles Times one. But if the results weren’t done at exactly the same time, or without all search history cleared properly, other issues than a filter bubble might be to blame.

If done right, where people all appear to Google as if they were brand new visitors with no search history at all, it’s hard for Google to personalize anything at all in a filter bubble way.

So Why The “Personalization?” & Magic Keywords

If it’s hard for Google to personalize without a search history, how can a single search for “Obama” cause this type of change to happen? Google is using the activity for a significant number of people to influence the “customization” it does for an even larger group.

Going back to the Wall Street Journal article, it says this:

Google said that the Obama-Romney disparity reflects the fact that more people searched for “Obama” followed by searches for “Iran” than the number of people who searched for “Romney” followed by “Iran.”

In other words, because Google has seen a number of people searching for Obama, then searching for these other terms, it has decided it makes sense to insert some Obama-oriented results into those other searches for many people.

This won’t just be for Obama, either. Other words trigger the same type of reaction (try a search for “cars” followed by a state or “travel” by some geographic location — these often work for me).

In the end, rather than producing personalized results, it’s more like ”mass personalization.” A better term suggested by Julia Angwin, the WSJ reporter who wrote the story, as I emailed with her after it appeared, might be “mass customization.” Google is using the previous query actions from some to influence how it customizes results for the masses.

Previous Query & Notification

As a reminder, Google using someone’s previous queries to personalize results isn’t new. Google’s been doing that since 2008. What is new is the idea that the previous search activity of others can influence everyone’s results. Also fairly new is Google’s failure to notify in a consistent manner when such customization is happening or how to turn it off.

In 2008, when Google started using previous queries in a big way, it also rolled out a consistent way to notify people when this was happening, as our story below covers:

Over time, the little “customized” message that would appear dropped from the top of the page to the bottom to having disappeared sometime this year entirely.

Personalized results are supposed to be flagged as part of the new Search Plus Your World system that launched earlier this year, but I find that doesn’t always seem to happen. Meanwhile, I noticed last month that the control icon to switch personalized results on-and-off sometimes seems to disappear. It looks like this:

The “world” icon is supposed to turn off personalization easily. When I mentioned not seeing this to Google last month, I was told I might not be seeing it because I was perhaps part of one of the many tests that Google’s always running.

Perhaps, and I do see it now when logged in. But for those logged out, if they don’t want personalization or customization to happen based on their search history, this switch is definitely not showing, making it complicated to get more “normal” results.

Of course, it’s also important to remember there are no “normal” results. Google, like Bing, will tailor results based on your geography. That means everyone will see something slightly different. But still, for those who want a more “pure” search experience, Google could — and should — make it easier to get that.

Postscript: See our follow-up article, Of “Magic Keywords” & Flavors Of Personalized Search At Google.

Related Articles

Related Topics: Channel: SEO | Features: Analysis | Google: Personalized Search | Google: Search Customization | Google: Search Plus Your World | Top News

Sponsored


About The Author: is a Founding Editor of Search Engine Land. He’s a widely cited authority on search engines and search marketing issues who has covered the space since 1996. Danny also serves as Chief Content Officer for Third Door Media, which publishes Search Engine Land and produces the SMX: Search Marketing Expo conference series. He has a personal blog called Daggle (and keeps his disclosures page there). He can be found on Facebook, Google + and microblogs on Twitter as @dannysullivan.

Connect with the author via: Email | Twitter | Google+ | LinkedIn



SearchCap:

Get all the top search stories emailed daily!  

Share

Other ways to share:
 

Read before commenting! We welcome constructive comments and allow any that meet our common sense criteria. This means being respectful and polite to others. It means providing helpful information that contributes to a story or discussion. It means leaving links only that substantially add further to a discussion. Comments using foul language, being disrespectful to others or otherwise violating what we believe are common sense standards of discussion will be deleted. Comments may also be removed if they are posted from anonymous accounts. You can read more about our comments policy here.
  • http://steveplunkett.com @steveplunkett

    Danny,

    (points to elephant in the room)

    The results will change and be biased as they should be.. @barackobama is the president of the united states.

    http://www.google.com/search?hl=en&newwindow=1&safe=active&spell=1&q=president&sa=X&ei=Ce-XUNezIZSuqAGf5ICgBg&ved=0CB4QvwUoAA&biw=1280&bih=675&changed_loc=1

  • danny76

    Maybe one day someone will take a hard look at Google Trends, and try to explain why they never showed “Benghazi” as a top search trend when volumes clearly dwarfed all their “hot searches” at the time.

  • http://twitter.com/hashtagdeals #HashTagDeals

    The bias — during an election year — is that Obama’s name is on the page multiple times. Go back in time 12 months and append Ron Paul or some other candidate’s name to half the posts. It will make a difference simply by being on the page.

  • Jenksy

    Once more, Danny, you leave the obvious floating about as if it were an anomaly, which is — by definition — either biased, or shit (read: CLULESS) journalism.

    Google’s algorithms are rooted in numbers. Algorithmic numbering is is based on quantity (as all mathematics is). Should a greater NUMBER develop for any given idea, it will express itself in Google’s engine as A..) More important, B.) More True, and C.) More Authoritative.

    Do you address this at all…?

    NOPE.

    Loathe as I am to admit it, others, in other posts have called you out, questioning whether or not you have ever ranked a website. At this point, I must wonder the same.

    Danny, my friend, at this point I am forced to conclude that you do not know your ass from your elbow concerning how search engines operate.

    Do us all a favor and attempt to rank a site that does not deserve to rank on page 1 Danny — you, and all the rest of us will learn much over the course of the process.

  • http://searchengineland.com/ Danny Sullivan

    Carl, first, I’m hoping to get Google to explain more. But the popularity in searches doesn’t equal the popularity in second searches. IE, lots of people who search for Obama may go on to search for Iran. Lots of people who search for Romney might not.

  • http://www.gabrielweinberg.com/ Gabriel Weinberg

    Hey Danny, a few comments on the above based on our study.

    –The results that are inserted do generally seem more Obama-friendly than not. The most common source inserted was Huffington Post and we also saw other left-leaning orgs more than right-learning or neutral orgs. Again, like you I think this is a simple consequence of the algorithm, e.g. it is likely these sources cover Obama more, but it is nevertheless true.

    –Even if they were completely neutral sources about Obama generally, I still think that biases things because of availability bias, which is a powerful form of bias documented again and again. The more you see something, especially in a research context, the more you’re comfortable with it, likely to recall it, etc.

    –In our study people did search at the same time, and we saw a lot of variation across the users in these inserted-results that appeared to be because of personal signals (but again, impossible to tell). Some people never got it, some more than others and in different places. So I agree magic keywords in general seem like mass-customization, but as they seem to be applied in a personal manner they create their own filter-bubble.

    –I agree with you not being able to opt-out is particularly troublesome. On that, most people — especially outside the US and on Google Apps — did not have a toggle. And even when they did have the toggle, it appeared to do less than signing out did. It *seemed* (and I have no idea for sure) that the toggle was more about turning off data from connected personal accounts (that have the face icon next to them) rather than other personal signals (inferred personal accounts, search history, magic keywords, etc.).

  • http://searchengineland.com/ Danny Sullivan

    Thanks, Gabriel. I’m looking forward to exploring this more. It’s confusing why people would get different news sources for the same search if they were logged out, had no prior search history, etc. None of them should have had any personal signals inserted that way, other than their location and their direct previous search. It sounds like in the WSJ study they were all logged out/no history but in yours, some might not have done this. But I’ll follow up separately to understand more.

    In terms of bias, all search results are biased — even Duck Duck Go has biased search results, as a consequence of how your own search algorithm makes decisions on what to sort through and how to rank.

    I’m trying to help people understand, I suppose, that bias doesn’t necessarily mean favoritism. In a highly charged election year, I see people reading this as Obama somehow getting some favoritism over Romney.

    Do a search for “iran,” then do a search for “syria,” and you’ll see your search results being influenced by Iran. Is that a favoritism for Iran? Of course not — but that’s the same thing happening with the Obama search.

    For whatever reason, Google has decided there’s a strong correlation between Obama and certain follow-on searches, just like there’s a correlation between Iran and Syria. It’s not the same thing as favoring Obama or Iran, but some may misinterpret as so.

  • http://www.gabrielweinberg.com/ Gabriel Weinberg

    It is indeed confusing, but that’s what appears to be happening. Yes, our first experiment was a bit muddled as to the off state. However, our second experiment has an off state that is strictly private browsing, and people searched at the same time. We hope to share those results soon, but it is clear that people saw different links, orderings, and news results (though it had no magic keywords involved).

    It would be great if you did a series of your own studies. I’d be happy to share what we learned in terms of setting them up.

    The favoritism argument is not that magic keywords = favoritism, but that inserting so many Obama-results without inserting the Romney-results amounts to inadvertent favoritism. That is, obama being a keyword and not romney is where the favoritism comes in — again unintentional.

  • Fodbold

    Mitt Romney = some rich guy that wish to be president.

    Obama = has currently been the president for 4 years.

    The two men are not equal in any aspects of either fingerprint on politics and thereof derived influence on any political matter or in ambition to do so. Maybe Romney is currently, like this week, polled as “as popular” as The President. Maybe he has an opinion about this or that, but Obama have 4 years in the oval office actually being The Man.

    Shouldn’t that reflect on a search engine result?

  • Dom Casas

    As far as I know a query for Obama will give millions of results compared to Romney. Google is not bias it’s just that Obama has been that popular in the last 4 years we can’t blame Google for showing you relevant search term about Obama “Obama This Obama That”.

    If Romney will be the next President then he’ll have the same popularity as Obama on Google search result. I’m from the Philippines that means I am not in favor to any of them.

  • Pat Grady

    you’re talking like Romney searches are some arcane, little searched thing. people have been digging in for details, deep details. think highly of you, but this smells funny to me. you’re conjecturing more than usual, have always appreciated your data / fact / evidence based reporting. i can’t help but think that disclosure would be healthy here (i voted for Gary J).

  • http://searchengineland.com/ Danny Sullivan

    If you’re asking who I voted for, it was Obama. As for Romney searches being “arcane,” I never said that. I never said that Romney wasn’t a popular search or wasn’t somehow not as popular as Obama.

    I’ll I’ve said is that for some reason, Google seems to be looking at certain words (Obama, cars), sees a high percentage of searches that occur right after those and seems to think it makes sense to give the subsequent searches in some cases a flavor of the previous search.

    Are they doing it to be biased toward Obama? If so, it’s a pretty stupid way to “favor” him. I already gave one example where the result, by not appearing for Romney but appearing for Obama, potentially helps Romney.

    Go search for Iran. Now go search for Syria. You’ll see a “you recently searched for Iran” link appear, influenced by the fact you just searched for Iran. Is Google doing that to somehow favor Iran over other countries?

    Search for “utah,” then search for “used cars.” You should see a “you recently searched for Utah” link. Is Google somehow favoring Utah over other states?

    I don’t see, from what I’ve explored so far, that this is all part of some secret plot by Google to somehow manage to influence people to like Obama on the off chance they search for him and then later search for things like Iran or gay marriage.

    What it is, however, is an interesting twist to how Google is deciding to alter search results for a wide variety of terms — those magic keywords if you will — beyond what someone initially entered.

    You bet, I’m working to discover more about why Google does this and when it happens.

  • keaner

    I also agree, this post seems very bias towards Obama and Google having done nothing wrong. Good article though.

  • http://steveplunkett.com @steveplunkett

    Danny,
    Obama has more .gov information.
    lol.
    =)

  • http://www.v2interactive.net/ Josh

    Insert political comment here that does not relate to SEO nor marketing.

  • Pat Grady

    others gave 10 down arrows to Danny’s answer here? look, i asked for disclosure and got it. Danny’s explained things further and is looking at more data. my experience is that when you challenge someone this directly, it is rare to see a reply filled with grace, understanding, analysis, but also devoid of becoming defensive. really not sure what others are negative on here. thank you Danny, i look forward to learning more from your reporting.

  • http://searchengineland.com/ Danny Sullivan

    Pat, I could leave a comment saying that Google is the most evil company in the world and should be abolished, and it would get 10-20 down votes from the same small group of people who intensely hate Google and have decided I get to be a proxy for their upset. It really doesn’t matter what I write.

    Your question was fair enough, and I was happy to answer it. I really didn’t see this as some type of political story. The WSJ saw that angle and ran with it, and that makes sense. But as a search marketer, was was far more fascinating (and important) was the idea that certain “magic keywords” are going to influence the subsequent searches we do — not just for us, but for everyone.

    That’s a new thing we’ve really not seen before with Google. We knew that previous queries could alter what an individual might see but not that previous queries could alter what everyone might see.

    I was trying to help clear away some of the cutter about “bias” and “filter bubbles” to highlight that more important (to me) point — this is a new ranking factor. And it’s definitely one I’m trying to get further answers to from Google. I just couldn’t get them in time for this story, and I wanted to cover what we could tell so far as a starting point.

  • Arne van Elk

    I would like to know which magic words trigger mass customization after doing a Romney search. If ‘Obama’ followed by ‘Iran’ triggers some form of customization, which search terms do the same for Romney?

  • jayyepeeen

    I think some of the users here are being a little myopic in their views. One should also take into account the history and global results, not just America’s. I am not from the States and I didn’t really know who Mitt Romney was until 2 months ago. But everybody knows who Obama is and he had 4 years in the oval office and it’s no surprise that he’s more searched about -overseas or in America. Maybe it had to do with the total number of hits and not number of hits per week. Just my two cents!

  • danny76

    For the two people that voted my comment down thus far, can you please tell me why you did so? I’d really like to hear your logic.

  • http://twitter.com/ronin_tan The Ronin

    the question is how many queries would Google consider as trending, what if you can influence the search for a specific keyword then most probably that people within that specific demographic will get search results relevant to the keyword you had planned to influence on.

  • http://twitter.com/ronin_tan The Ronin

    hmm so I guess with your comment here, I can imply that what Romney should have done is to have an expert panel of search engine marketers, studied carefully results from Obama searches wrote a lot of press or articles about the same topics and publish.

    This way his presence becomes more viable and related to the search terms thus affecting results.

  • danny76

    Danny – do you care to comment on why Google Trends totally omitted Benghazi as a “hot search” phrase when it’s search volumes clearly outpaced the other phrases they dubbed a “hot”?

  • Farhat Bakhsh

    Not surprising at all.Least to give the explanation of algorithm is more depressing! wish google had been more rational in its personalisation so had not end up infuencing the result of the most crucial decision of its own land.

Get Our News, Everywhere!

Daily Email:

Follow Search Engine Land on Twitter @sengineland Like Search Engine Land on Facebook Follow Search Engine Land on Google+ Get the Search Engine Land Feed Connect with Search Engine Land on LinkedIn Check out our Tumblr! See us on Pinterest

 
 

Click to watch SMX conference video

Join us at one of our SMX or MarTech events:

United States

Europe

Australia & China

Learn more about: SMX | MarTech


Free Daily Search News Recap!

SearchCap is a once-per-day newsletter update - sign up below and get the news delivered to you!

 


 

Search Engine Land Periodic Table of SEO Success Factors

Get Your Copy
Read The Full SEO Guide