Of “Magic Keywords” & Flavors Of Personalized Search At Google
When is personalized search not personalized search? A recently discovered shift on how Google may alter your search results based on what you — and others in aggregate — previously have searched for may have you wondering how to answer that question. To understand the latest development, I think it’s helpful to go back and […]
When is personalized search not personalized search? A recently discovered shift on how Google may alter your search results based on what you — and others in aggregate — previously have searched for may have you wondering how to answer that question.
To understand the latest development, I think it’s helpful to go back and review the “flavors” of personal search that Google has, flavors that often all get mixed together. Let’s dive in.
Personalization Based On Geography
Google has long had personalized search, where various factors are used to provide results tailored to the individual.
In fact, Google really hasn’t had “normal” results for years, since for over a decade, it has personalized based on geographic location. These days, it’s gone well past being country-based and works to the city level. What someone sees in one city can be radically different than someone in another.
For example, here’s what I get for a “zoo” search:
Local listings for my area top the page, followed by the two major zoos nearest to my home (it would be nice to see the Santa Ana Zoo top that list, as being the closest to me, but the LA and San Diego zoos are larger yet still nearby).
Personalization Based On Search History
Google has also personalized using your search history stretching back for years, if you’re signed in and have enabled its web history feature.
Even if you’re signed-out, Google may personalized your results because it keeps a 180 day record of what a particular browser has searched for, linked to a cookie in that browser.
Consider this search for Windows 8 below (you can click to enlarge):
The left side shows the results I get when I searched with personalization turned off, both either by using a the Chrome browser in a fresh “Incognito” session, so Google has no history associated with the browser, or using the button Google provides for signed-in users to turn off personalization.
The right shows my personalized results. The arrows point out the key difference. The main Microsoft Windows site, which wasn’t in the top ten results for Windows 8 for my unpersonalized results, suddenly moves up to number three. That causes a page that was at number three, about Windows 8 and Windows RT, to move down to number four.
Why the change? There’s a clue right underneath the listing:
The arrows points to a message where Google is telling me that I’ve visited the main Microsoft Windows page five times that it knows about. How does it know this? I’ve been logged into Google, done searches and selected that site at least five times that it has recorded.
The fact that I keep going back to this site is almost certainly causing Google to think that I like it. So, when I do certain searches, Google’s personalized search algorithm gives the page boost. That’s why it makes it into the top ten results.
This type of personalization is the oldest form of personalization that Google does (and Bing does something similar). It stretches back to Feb. 2007, as we covered then:
It was greatly expanded in Dec. 2009, as our two stories that month covered:
- Google Now Personalizes Everyone’s Search Results
- Google’s Personalized Results: The “New Normal” That Deserves Extraordinary Attention
Personalization Based On Personal Sharing
Other things beyond search history have also influenced the type of personalization covered above, where new results might appear or the overal rankings might shift. Among these were social signals. Who you know, what they’ve shared or what you’ve personally shared could also have an impact.
But earlier this year, those types of signals grew dramatically in their power and presence and also got largely lumped under a new name, Search Plus Your World. To illustrate this, consider this search for Windows Phone:
As before, unpersonalized results are on the left, personalized on the right. On the right, you can see there’s a new link — one with my picture next to it. It’s a post about Windows Phone 8 that I made today on Google+. Google’s Search Plus Your World algorithm is slanted toward showing you your own content — a perfect “ego search” reward mechanism. So, I see this, while others don’t.
Personalization Based On Social Connections
You can also see above that I also get a list of images that are related to Windows Phone that have been shared by people I’m connected to on Google+. Adding this new content means something has to go — on the left, those three arrows pointing down show that those listings all have been dropped from the top half of the page (and some out of the first page of results entirely).
Now look here, for a better example of the impact social connections can have on results:
That’s further down the page of results. Again, on the left, the non-personalized ones. On the right, the three arrows point to results that are all appearing because of my social connections.
The first is a CNET article, almost certainly appearing because I’m “friends” with CNET on Google+. The second is from the Typekit Blog, appearing because someone I know shared it. That’s what the fourth arrow on the right is indicating, who exactly that was (thanks, Jeffrey Veen!). The last is an article by Harry McCracken, probably appearing because I’m friends with Harry in various ways, probably most importantly for Google+.
By the way, if you still haven’t gotten it into your head why Google+ is important, this is it. If people become friends with you (or your company), that can be the number one factor trumping all others about why you might rank well. Links? Title tags? All those various SEO factors? A Google+ connection can beat them.
Personalization Based On Previous Query
I mentioned personalization based on search history earlier. One unique variation of this that Google has long used is to look at what you searched for just before you did another search to see if it makes sense to use both terms to refine your results.
Called “previous query refinement,” here’s a good example of it. I searched for “flag,” then I searched for “california,” and Google showed me this:
See the listing with the message “You recently searched for flag” that the arrow pointed at? It’s for a page on Wikipedia about the California flag, even though the search I did was just for “california” and not about flags at all. But Google, seeing I’d just searched for “flag,” is guessing that I might find it relevant if it mixes in a match for both “california” and “flag.”
Google has done this for years. It was introduced in 2007 for ads and 2008 for unpaid listings, as we covered back then:
- Personalized AdWords: Google Ads You See Influenced By Previous Searches
- Google Looking At Multiple Previous Queries To Tailor Search Ads
- “Previous Query” Refinement Coming To Hit Google Results
- Google Now Notifies Of “Search Customization” & Gives Searchers Control
Mass Personalization Based On Everyone’s Previous Queries
This leads to the new factor that emerged this week. Google-rival Duck Duck Go discovered that for certain words, previous query refinement would happen for a wide range of people, rather than on an individual basis.
The Wall Street Journal, in an article this week, spotlighted the most notable example of this. Searching for “obama” would cause searches for topics like “iran” or “medicare” to change. Searching for “romney” didn’t cause this type of change.
Here’s an example, where a search for “medicare” after “obama” brought up a listing that combined both terms:
The fact that Obama acted as what Duck Duck Go calls a “magic keyword” but not Romney caused some small debate on whether Google was somehow being political biased. I don’t think that’s the case, and I explain more in my story on all this from earlier this week:
Why this happens for Obama and not Romney isn’t a consequence of Google trying to favor Obama but down to its search algorithm looking at all the things people search for, followed by the subsequent things people search for and trying to decide if it makes sense to inject personal query refinement for everyone.
Consider this search for “amazon” that I did right after a “hunger games” search:
You can see that in a search on “Amazon,” I’m getting a Hunger Games listing included among the rest, even though I never used the words “hunger games” in my current search. That’s previous query refinement kicking in.
Try the same thing for a search on “Redshirts,” a great sci-fi novel by John Scalzi (required reading for anyone with even a passing familiarity of how the redshirts on Star Trek always die). The refinement doesn’t happen. Why not?
Redshirts isn’t a magic keyword. Google isn’t biased in favor of The Hunger Games over Redshirts. Google just probably hasn’t seen enough data to make it feel confident that it makes sense to insert a Redshirts link into a “pure” Amazon search.
What Makes A Magic Keyword?
As I said, previous query refinement has been around for ages. But Google tells me this new flavor, where data from everyone is used to help influence when it triggers for others, began as a test last year (and says the messaging of “You recently searched for….” started in April 2011). Currently, Google says it happens for about 0.3% of all searches.
Google also doesn’t really consider this personalization, in that while the refinement is influenced by a person’s previous query, whether it will actually shows depends on activity overall. It’s similar, Google says, to the familiar spelling corrections that Google shows, the former “Did you mean…” or the current “Showing results for…” corrections that it does:
That’s also why, Google tells me, that it doesn’t offer a way to easily turn this type of refinement off (you can, if you really try, as I’ll get to in a bit).
What exactly causes a word or phrase to have a magical ability to influence future searches? Google won’t say much, other than it’s linked to query popularity over time and that it’s especially designed to surface recent navigational queries that might be related to a recent search.
To use the Hunger Games as an example, Google seems to see a lot of searches for both that and Amazon, and it probably has lots of data also where people are specifically searching for both together (hunger games amazon). It can look at the top things people pick from that combined search and make some guesses about what it should likely insert as a previous query.
In fact, look at the top unpersonalized results for “hunger games amazon” and “california flag” below
In both cases, the top result for the combined term is the result that got inserted as the previous query refinement link. It’s not always the top result, however, as this shows:
The bottom line for search marketers is that this means that if you manage to rank well for a combined term, and if one of those terms is a “magic keyword,” you’ve got a shot at doing well for that single term.
Impact On Searchers: Probably Good, But More Control Needed
For searchers, this is probably a useful feature. However, I do think Google should provide an option similar to what it does for signed-in users, so that this can easily be disabled.
It’s also unclear when this is happening. I’ve spent time talking about how we originally had previous query refinement for individuals. That means if I searched for a word, then another word, I might see a refinement while someone else might not.
The newly-discovered shift makes is far more likely everyone now sees these types of refinements. But I can’t find examples where previous query only works for me as a individual but not also happening when I’m signed out (and thus doing a type of “everyone” search).
The bottom line here is that previous query probably isn’t much of an individual or personal thing any longer, except for the incredibly tiny number of people who completely disable any type of search history at all, including that associated with their browsers.
In a future article, I’ll go into more depth about how to know exactly when personalization is happening (Google isn’t making this clear, as it once did), as well as how to turn it off if you like. For now, you can view Google’s instructions here.
Let me also add that the personalization I’ve discussed: previous query, social influence, social sharing, geography? All of that also happens with Bing.
- Google’s Personalized Results: The “New Normal” That Deserves Extraordinary Attention
- Google’s Results Get More Personal With “Search Plus Your World”
- DuckDuckGo’s New Video Targets Google’s “Filter Bubble” Of Personalized Results
- Bing Results Get Localized & Personalized
- Bing Gets More Personal With Adaptive Search
- Why Google “Personalizes” Results Based On Obama Searches But Not Romney