Analysis of online search habits reveals that the first result - often a Wikipedia page - is by far the most popular. But, asks Charles Arthur, does that mean it's the right answer?
One of the most unusual features that the Google search engine boasted when it first went online was that little button called "I'm Feeling Lucky" - which, when pressed after entering a search phrase, would take you straight to the site that Google judged to be the top result for that search. What boosted searchers' confidence in Google, at a time when rival search engines such as AltaVista were stuffed by spammed results with no bearing on any phrase you might have wanted to find, was that if you felt lucky, you almost always were lucky.
The first result was, indeed, very often just what you were looking for, far more often than on the spam-bombed competitors. Google rose to the top.
But how powerful is Google's grasp over our searches? And what effect will that have on the web? While many have focused on the identities of the people revealed by the logs of Google searches accidentally made public by AOL earlier this month, other groups online have begun to use that research just as was intended: anonymously, to understand peoples' web-searching habits.
It turns out that in a world where we hear so much about the "long tail" of a million niches, there is a very short tail indeed of "clickthroughs" - where someone clicks on a search result to go to the linked site (which happened, in the AOL data, 52% of the time).
According to the logs, the top result on any search engine gets 42.1% of the clickthroughs; the second, 11.2%. That's more than half of all clickthroughs in just two results. It's rapidly downhill from there, apart from the 10th result (before you click onto the next page), which sees a slight uptick in popularity over the ninth .
The latter seems to stem from a psychological effect: on reaching the bottom of the results page, people decide to start looking at something external rather than another page of potential answers.
Even so, the 10th result only garners 3% of clickthroughs. The next 990 results (since Google only ever offers links to nine other pages) share just 11% of clickthroughs, an average of 0.01% each, though almost every click beyond the first page of results occurs on the second. The furthest that any searcher went, according to an analysis at tinyurl.com/hnh8a, was the 449th result - in a search for "beastiality" [sic]. "Guess you've got to go a long way for your kicks," observed the person who dug up that datum.
Our tendency not to search very far matters to two disparate groups: those concerned about where we find sources of authoritative knowledge, and those keen to tweak their web pages - and at the extreme to spam the web - in order to move up the search rankings for all sorts of commercial activity.
This second group, of self-described "search engine optimisers" (SEOs) - whose mission is not actually to tweak Google or Yahoo, but to optimise the web pages of their clients - has fallen upon the AOL data with glee. Analysing the dataset is not trivial; it contains 20 million entries, and occupies 439MB when compressed, and about 2GB when decompressed. But for SEOs, it is literally a gold mine: there is money in understanding how many people will click on a particular item in the rankings, because SEO clients pay by results. If you can work out how much more valuable it is to be the No.2 link than the No.3, you will know how to allocate your resources better.
There are some caveats: the database of results came from AOL users, who tend to be less web-savvy than people who have chosen a particular search engine (Google is the default search engine for people using AOL's service or web pages). Some SEOs noted that the results seem skewed towards the top result compared to their own experience - that the first three results might see a variation of 10% in popularity - while others said the results matched theirs almost exactly.
Do you feel lucky?
The data also reveal the problems some users still have with search engines - such as the words "search terms" which appear in the search box in light text. Some people, apparently, click the search button at once. "That's where we got all the useless clicks for the term 'search terms' when we ran Adwords. I'd love to see how many of the 20 million searches were for 'Type AOL Keywords Or Search Terms Here'," said Dan Thies of seoresearchlabs.com.
But while the data files are pleasing SEOs, they will probably distress those who feel that the bias towards feeling lucky - and particularly that too many people rely unquestioningly on the top few links on any search engine - is not positive. Just because something gets a lot of links doesn't mean it's true. The practice of "Googlebombing" - creating links to a phrase aimed at a particular page to move it up the rankings - probably saw its clearest example when the page pointing to George W Bush's profile on the whitehouse.gov site was linked to by pages using the phrase "miserable failure". Type that phrase in, and you'll still get pointed there.
Wikipedia doesn't rely on Googlebombing, but the frequency with which its pages are linked to means its pages rank very highly in search results. But can those results be trusted as sources of factual information?
Recently The Onion, the satirical newspaper, ran an article entitled Wikipedia celebrates 750 years of American independence, and quoted the site's co-founder Jimmy Wales, as saying: "At 750 years, the US is by far the world's oldest surviving democracy, and is certainly deserving of our recognition. According to our database, that's 212 years older than the Eiffel Tower, 347 years older than the earliest-known woolly-mammoth fossil, and a full 493 years older than the microwave oven."
Of course, none of those "facts" is correct, and Wales never said nor believes any such thing. But the satire bites, because Wikipedia is simultaneously widely relied on, yet also often blatantly or subtly wrong. It is one of the Internet's great unspoken truths.
And the reliance is both wide and deep. "I'm still trying to find a common keyword that doesn't have Wikipedia in the top 10 results," noted Nick Carr, the author of Does IT Matter, on his blog (at roughtype.com) earlier this month. Gunpowder, Mona Lisa, trans fat, Holy Roman Empire, sauerkraut: for all, Wikipedia was top. If you feel lucky, Google will take you to them. And brothel, lawn mower, Alfred Hitchcock, even Nicholas Carr: according to Google, Wikipedia had the second or third most authoritative web page about them, meaning they will garner around 11% of all clickthroughs on those topics.
But it's not only common words that Wikipedia dominates. Where would you expect to find the best - as in most accurate - source of online information about John Harrison, the British inventor of the maritime clock that would keep time accurately on a ship? His efforts were the subject of Dava Sobel's painstakingly researched, and bestselling book Longitude; but the top result on Google, and so the one which will see 40% of clickthroughs, is Wikipedia's.
Patrick Ross, of the Center for the Study of Digital Property, a US thinktank, decided to follow this thread. He used his interest in cartography to see how far Wikipedia's grip on Google results - and hence on people's expected reliance on it as a topic - extended into the realm he knew about. Vinland, the Vikings' apparent landing in North America in the 11th century? The top result links to the Wikipedia page. And so on. (Often, results further down the listings - such as those at answers.com - are themselves taken directly from Wikipedia, where the open licence allows direct copying of information; so the appearance of multiple reinforcing answers is false, and the Wikipedia entered data will get much more than the expected ratio of clickthroughs.)
Ross then pointed out, on the IPcentral blog, that there was at the time no Wikipedia entry for Battista Agnese, a 16th century Italian cartographer. One soon appeared - plagiarised, said Ross, from a Library of Congress page, and with errors added. Then that entry disappeared - deleted by a Wikipedia "administrator" and replaced by a different one. And within days of appearing, the Wikipedia version was the 10th-ranked result for "Battista Agnese" on Google, although the Library of Congress page remains top at present. But as the entry spreads to other online encyclopedias that reuse Wikipedia's content and acknowledge it with a link, how long will that remain true?
For SEOs, though, accuracy of content doesn't matter - as long it's their content someone is looking at. To that end, they are all mining the data as vigorously as they can; there is now even a site where you can query the database by user ID, search keywords, and what website people went to. Ask it what search terms people put in that saw them go to wikipedia.org, and the 2,853 hits indicate that they were most commonly looking for ... Wikipedia.
On that, at least, one can be certain that Google got it right.
No responsibility can be taken for the content of external Internet sites.