Google’s latest podcast episode — Search Off the Record — analyzes how the company tackles specific things in search results.
The latest episode looks at spam and the way it ranks for certain things in search results. Participating in the podcast were John Mueller, Gary Illyes, Martin Splitt, as well as Duy Nguyen, a guest from the Google search quality team.
Ranking always has been Google's secret, but Illyes, Google webmaster trends analyst, in the podcast sheds some light on how the search engine approaches it.
“Ranking is one of the topics where we don’t want to say too much,” Illyes said. “If you went to an information-retrieval class, you heard about ranking and the public version. Ranking results is just math and figuring out basic relevancy, and some magic we’re not talking about.”
Ranking results in an index generated from a search query is done using hundreds of signals, he said. One is referred to as relevancy, which is based on query and page content, because there might be billions of results for one specific topic such as cookies — how to make them, what type, and more.
“There’s no reason to return all the results,” Illyes said.
Google then ranks the pages to reduce the number of results retuned to the user — perhaps one or two thousand, something more manageable in query results. This would be an amount the search company believes people are willing to go through.
Its algorithms rank the pages and then applies RankBrain, HTTPS boost, relevancy, and other quality signals from the query to create a list in the reverse order based on those numbers assigned.
“These magic algorithms can still make massive changes in the results, but they are only working with about 1,000 presented to them,” he said.
Ranking is number-based. Google assigns a number for each result, and calculate the number using the signal collected during indexing. The results is a reverse order based on the numbers assigned. The magic algorithms like RankBrain multiply the numbers assigned to each result.
“For example, if someone wanted to promote a result because it was determined that it would be a better result for lemon coconut cookies, then you would multiple the results score by two,” he said. “If we wanted to remove a result from the set for whatever reason, we could multiple the score by zero.”