How Google search personalization works: Professors and politicians just don’t understand

Ramesh Srinivasan, a UCLA professor and author of books on the impact of technology on our lives, joined the “Forward with Andrew Yang” podcast for an hour of conversation that included, among many other topics, Google search.

What Srinivasan got wrong. Pretty much everything he said about Google’s personalization “stuff.” Here’s what he said:

“…when they [Google] started doing this personalization stuff, what happened is we became googled and we became googled not based on … some sort of neutral notion of relevance but based on what would grab our attention. And the way that works, which I think is really interesting, is it’s all based on correlation. So you know if you Andrew Yang have looked at you know a million web pages and I have all this data about your engagement on those different web pages, which we call documents in the information sciences, and then I have very similar profile to you it will recommend content to me based on correlation mapping.”

– Ramesh Srinivasan

Now, I know they don’t teach a course on how search engines work at most universities. But I’m pretty amazed to discover that this professor has such a poor grasp on the topic of search. There are so many available resources – from Google and others who study search engines and information retrieval – that are easily findable via “Googling.”

Sullivan corrects Srinivasan. In a long series of tweets spanning multiple discussion points, one of the core ones was how Sullivan explains personalization, and where Srinivasan is getting it wrong.

The key point Sullivan makes is this: “Personalization means we’re showing content uniquely to you based on information uniquely about you. Location, language are not personalization factors because they are common to many people & generate common results.”

Srinivasan’s response? “Yes again that is how Google defines personalization but the term has a range of other meanings for the rest of us.”

Basically, Srinivasan has his own personal definition of “personalization.” But how Google defines personalization (and how personalization in Google search works) and Srinivasan’s perception of what “personalization” means don’t match.

Even though a lot of people spout the cliche “perception is reality,” it actually isn’t. Sure, perception can become your personal reality. But really, it’s a sign of mental laziness. It’s the equivalent of saying “I’m right and you’re wrong.” Assumptions are not facts. Words and definitions matter.

How Google search personalization works. Google has told us how it works. We’ll start back in 2009 when Google introduced the idea of “personalized search for everyone.” (If you want to go further back, see “Google Search History Expands, Becomes Web History” from 2007.)

Up until this point, Google had only offered personalized search to signed-in Google users who had enabled web history. The “for everyone” was new in that Google could now provide personalization based on 180 days of a searcher’s past activity. However, personalized search was on by default, which meant searchers had to opt out.

Then, in 2011, Google’s Amit Singhal shared some thoughts on this very topic:

“Personalization is a narrow class of context. It’s the context of you, the searcher, including your interests and your network of contacts. This special kind of context has a subtle, but important, impact on search results. Personalization is understanding who you are to give you the best answers, and is definitely not about making search results look like your reflection in a mirror.”

– Amit Singhal

Google went on to explain that “personalization” as it applied to Google search at the time, meant that Google may use up to 180 days from your web search history to “personalize” your results. In the most simple terms, Google knew the following about searchers:

• Their search queries.
• What they clicked on.
• The types of sites they visited.
• Which topics they are generally interested in.

But by 2018, Google said there was “very little search personalization” going on beyond using a “user’s location or immediate context from a prior search.” Google did a lot of testing, but ultimately didn’t see enough improvement in search quality.

To be a bit clearer here: A user’s location does not mean “localization.” Google is very specific in how they define terms. As Sullivan said, a location itself is not personalization, because many people are in the same location and can see common results. The differences are typically minor and unique to you.

In many ways, this is a debate about semantics. Many SEO practitioners have argued that localization is personalization. But it’s really an unwinnable argument because we know how narrowly Google defines the term “personalization.”

Also in 2018, DuckDuckDuckGo released a study that attacked Google (and not for the first time) for creating “filter bubbles.” They had 87 people search for “gun control,” “immigration,” and “vaccinations” at the same time on June 24, 2018. In response, Sullivan tweeted: “Over the years, a myth has developed that Google Search personalizes so much that for the same query, different people might get significantly different results from each other. This isn’t the case. Results can differ, but usually for non-personalized reasons.”

Why we care. This probably looks like yet another argument about search. Just another day on Twitter, nothing new here, right? Well, yes and no. The reason we care goes back to a message in one of our recent newsletters. “The general public knows so little about how they get the information they’re looking for, which isn’t acceptable since it’s their source of truth,” wrote our own Editor George Nguyen. That is clearly on display here.

Google has published help documents and guides that explain how Google works. Google is telling us how search works. And as SEOs, we’ve certainly learned plenty about how search works. So the point of this article is to educate Srinivasan and Yang, and anyone else who may be confused about how Google uses personalization in search results. But we also realize we can’t force anyone who wants to hold onto the belief that we’re all trapped in some sort of Google correlation map.

We here at Search Engine Land will continue our mission of educating people about how search works. And all we can do is continue to play this endless game of whack-a-mole, helping spread some truth to those outside our search bubble. One person at a time.