You are here
What do we know about how Google works?
ONLINE search and Google are synonymous. About 90 per cent of all searches on the Web run through Google, giving the company an enormous role in directing the worldwide flow of information on the Internet.
Last week, US President Donald Trump accused Google of abusing that power by intentionally suppressing positive stories about his administration. His assertions were discredited by search experts and even some vocal opponents of Google, but they spoke to a growing unease about the influence that tech companies have over what we see online.
As the Web has grown in size and complexity, the importance of Google Search has grown. A minor tweak to its algorithm can redirect huge amounts of Web traffic. Still, not many people have a good understanding of how Google delivers search results. And for good reason - Google tries to keep it a secret.
Here, we'll explain how it works, and why search experts say there is little evidence behind the president's claims.
So, just how does Google Search work?
Let's say you want to do a search for "health benefits of kale". When you start typing, Google's computer systems begin sifting through its index of hundreds of billions of Web pages that use that exact wording or a related phrase. (Google estimates that there are more than 53 million pages that match "health benefits of kale".) Then, Google puts those pages in order, using a secret algorithm.
There are hundreds of factors weighed by Google's algorithm. While Google keeps most details of its search formula under wraps, it has revealed some specifics.
One of the major breakthroughs with Google's search engine was a formula called PageRank, named after Larry Page, one of Google's founders and now the chief executive of its parent company Alphabet. PageRank works on the basic premise that a page's value can be determined by how many sites link to it. In the early days of Web search, this was a novel concept and it helped to propel Google past competitors like Yahoo and AltaVista.
The search engine has gotten more sophisticated over the years. (It was founded 20 years ago on Tuesday.) In addition to PageRank, the company has also said that the software looks at how often and where the keywords being searched for show up on a specific page, how recently the page was created (a sign of the freshness of the information) and the location of the person making the search.
Google said there were no major differences between how it retrieved news stories and other search results, although some factors, such as when the page was created, take on greater importance in news searches.
Why won't Google reveal the algorithm?
Google says that revealing its formula would make it easier for people who try to game search results. There is already a cottage industry of people who specializs in search engine optimization, or SEO, and help companies gain greater visibility for their pages.
With greater insight, the thinking goes, spam sites and advertisers could displace the most relevant pages from the top of search results.
But of course, Google has another reason for keeping its search engine formula secret: it's proprietary. For the most part, Google has established its dominance in search because it did a better job of revealing the best answers to a given query. Google wants to keep its search algorithm from competitors, just as Coca-Cola doesn't reveal its recipe.
Trump accused Google search of being biased against right-leaning media. True?
Google said political ideology was not a factor in any aspect of its search results. It said whether a user is conservative or liberal is not part of the information collected by the company, and it didn't categorise Web pages by political leanings.
However, the scrutiny over misinformation after the 2016 presidential election pushed Google to make a change to its search algorithm. At the time, Google found that 0.25 per cent of its daily traffic linked to intentionally misleading, false or offensive information. It wanted to surface what it called more "authoritative" content in search results. The change drew complaints that it prompted a steep drop-off in traffic. But the organisation that complained, and all the sites whose traffic it cited, leaned to the left.
How does Google determine the authoritativeness of search results?
It relies on an army of human "raters" to inform the quality of search results. Google has 10,000 raters spread around the world. They rate the quality of search results to determine whether the pages ranked first deliver expertise, authoritativeness and trustworthiness.
While the raters cannot directly change how Google's search algorithm functions, their opinions can flag issues with specific Web pages or blind spots in the search formula. Google publishes the guidelines used by the raters to determine search quality.
So Google search is completely neutral?
Not quite. When Google search was created, the search results were bare-bones - a list of 10 blue links. The main objective at the time was to deliver relevant results to get people out of Google as quickly as possible. That has changed significantly over the years.
Instead of simply providing links, Google is packing more information onto its Web pages - deepening the tension between publishers and other services like the review site Yelp, that rely on Google to direct traffic to their sites. Google has argued that users go to Google for more than just links - they go for information.
But as Google has incorporated travel information, shopping services or reviews about local restaurants and businesses into its search results, competitors argue that Google is self-dealing and favouring its own services over that of its competitors. The European Union has said such preferential treatment violates its antitrust laws.
There is also the problem of hidden bias. These are issues not about deliberate bias against one political ideology like Mr Trump raised, but about how algorithms or artificial intelligence will unintentionally amplify societal biases about groups like women and racial minorities.
The concern is that since many Google engineers are male and white or Asian, they are less likely to spot subtle problems that arise for underrepresented groups. For example, Google has a feature called Autocomplete, which offers suggestions for searches once a user begins typing. In the past, some of those suggestions came back with racist or sexist stereotypes.
Google has said it is aware that these types of biases may creep into search results, but that it remains vigilant in trying to stamp out those problems. Now, when starting a search with "Donald Trump is", the top three Autocomplete suggestions are "a democrat", "a great president", and "my president". NYTIMES