Return toArticles 9 mins watch

That movie where those two guys drink wine - How Google is understanding your searches

Posted July 23, 2013 by Will Critchlow

How does Google understand your searches? Well, in this video Tom Anthony and I talk about some of the ways that Google is ‘understanding’ the web. On a similar subject to my talk on the Future of Search, we talk through how Google is now interpreting our search results and their relevancy rather than just indexing them.

So, how can the search for ‘that film where those two guys drink wine’ really work? (It does!) And what does this mean for the state of search? Will and Tom talk through it using a few examples [here’s the Reddit one] in the video. You can find the full transcript below.

How do the changes we’re seeing in Google shape user expectations and what does this means for the future of search? We’d love to hear your thoughts in the comments below.

How Google is understanding your searches

Will: Hi everyone. I’m Will Critchlow. This is Tom Anthony, one of my colleagues here in London. Tom and I have been talking recently a lot about some new models for thinking about how Google is understanding the Web and how that is influencing the way people are searching. The first thing we talk about is a move from indexing to understanding. You want to tell us a little bit about that, Tom.

Tom: Yeah, basically the old model of just Google understanding a word and perhaps some synonyms has moved aside, and with the Knowledge Graph and everything Google are starting to understand the Web more than just index it. So a great example would be, if you search for Usain Bolt, you get the little fact box at the side, and it might tell you how tall Usain Bolt is.

Will: They know he’s a person, that’s it’s an entity.

Tom:  They know he’s a person. Yeah, so you get his name, nationality, the important stuff about a person. So it’s starting to really understand, and that’s the key I think.

Will: Yeah, and we see that as well with some kind of longer tail searches. There was a great Reddit thread, which we’ll link to . . .

Tom: Yeah.

Will:  . . .  where you can search for movies in really obscure ways. That one movie where those two guys drink wine is a search query that really shouldn’t work Yet on Google it does, and on pretty much any other search engine it doesn’t.

Tom: Doesn’t. Yeah.

Will: Precisely because there’s entity modeling.

Tom: Exactly. Another example would be, going back to the Usain Bolt thing, if you search for say Eileen Collins, who was an astronaut, it doesn’t tell you how tall Eileen Collins is. It tells you how long she spent in space. So not only are they . . .

Will:  Whereas you don’t get that for Usain Bolt.

Tom:  No, you get how tall he is because that’s what you’re probably interested in for an athlete. So not only is it understanding that he is a person, so understanding what properties about him are important, what properties about another person are important for a different reason.

Will: Which also relates to conversational searches, which we see in voice activated, where you can do a search for Usain Bolt and then follow up with how fast does he run the 100 meters.

Tom: Yeah.

Will: I don’t know if that exact query works, but those kind of queries.

Tom: Yeah. So the one I’ve seen is like, “How old is Justin Bieber,” and then, “how tall is he?” So you just use the pronoun, and it still understands. You’ve got this conversation where it’s understanding the entity we’re talking about.

Will: There was a keyword related example that I’ve been using recently, which is there’s been a growth, if look on Google Trends, there’s been a growth in searches including the word “best.” And as we were talking about, that doesn’t necessarily mean what it used to mean.

Tom: No. So before people might be searching for best because it was . . .

Will: There might not be a good reason.

Tom: No, no.

Will: But the model would have been pages that used the word “best” on them.

Tom: Exactly. Rather than expecting Google to understand the concept of best and somehow . . . also it shows the expectation of users of a search engines. They are now starting to believe. If you look at that trend, which we’ll flush out, you can see that there’s more and more trust in users that Google is going to understand what best means and help you.

Will: Right, people are assuming that that is a sensible thing to search for.

Tom:  Yes.

Will: Best restaurant. Best whatever.

Tom: Which it definitely didn’t use to be.

Will: Definitely didn’t use to be. So all of that tied together has led us to this new model for queries. We’re starting to break down the context of the query as well into . . . what was your word for it Tom?

Tom: Aspects.

Will: Aspects. Implicit and explicit aspects. So walk us through that.

Tom: Exactly. So basically, it used to be the old model was the whole of your query was the keywords you typed in, which was what you explicitly searched for.

Will: Yeah.

Tom: And over time, like a few years ago, SEOs were talking very heavily about personalization and all that stuff. But I think personalization is too narrow a concept. It doesn’t cover everything. So we’ve been using the term context to cover more and more of what we term the implicit aspect of a query.

Will: So where you are, what device you’re on.

Tom: Exactly.

Will:  What time of day is it. What searches you’ve done recently. All those things.

Tom: Yeah. Like a couple of years ago it came out that Google were using 57 signals when you weren’t logged in.

Will: Okay.

Tom: So these were 57 different signals. So we can’t even imagine what they all are. But that implicit part that Google’s trying to use these signals to give a better search result to you. Basically, these are part of the query, if you look at it like that. A great example is someone searching for “London tube stations.” If I just tell you someone searched London tube stations and tell you nothing else, it’s really difficult to understand what that user’s intent is. What are they looking for?

Will: Do they want a map? Do they want photos?

Tom: Is it a child looking for the history of the tube or something for some homework? But if I now tell you that London tube stations is the explicit part of the query, what they’ve typed in, but now to give you some of the implicit aspects. They’re an iPhone user. They’re on the street. They’re in London. And suddenly it becomes far easier to determine their intent.

Will: Right, that they want a tube station near where they are.

Tom: Nearby, exactly.

Will: That makes perfect sense. And so we have this model that we should be thinking of the query, not just as the words typed in, that’s the explicit aspect of the query.

Tom: Yeah.

Will: The implicit aspect of the query is everything else that goes along with that, which, as we said, is things like where you are, what time of day it is, what device you’re on, what searches you’ve done recently.

Tom: Yeah.

Will:  Your full search history.

Tom: Exactly.

Will: And even other information that they gather from other Google services.

Tom: Including the social aspects, which we’ve been deeming personalization. So personalization is part of this context.

Will: Sure. So who’s in your circles, who has you in their circles.

Tom: Exactly.

Will: Who’s plus’d one stuff and other social signals as well. So what does this mean over time? What we’re saying is that going back far enough the whole query was just the keyword.

Tom: Yeah.

Will: Then we started getting this additional implicit aspect on top of that.

Tom: Yeah.

Will: And what are we seeing now?

Tom: So now we’re seeing people are starting to understand that Google, as with the example with best, people are starting to understand and trust in Google, without knowing it, that’s the implicit part of the query . . .

Will: To understand them.

Tom: To understand them and to take in some implicit aspects. So it used to be, for example, that if I was searching for a restaurant and I was in Finchley in North London, I’d search “restaurant Finchley.” Nowadays I don’t do that. I just search the keyword “restaurant” because I understand and most users understand that Google is going to take into consideration where you are.

Will: Because they get that feedback. They do a search query like that. They start seeing it get personalized, all the implicit context. .

Tom:  Exactly. So over the last few years people have been doing more and more of those searches.

Will: Absolutely. Actually, I vividly remember giving a talk about keywords – it must have been four years ago – and mocking the kind of business owner who wanted to rank for “restaurant."

Tom: Yeah.

Will: What does that search even mean? But now, of course, we know what it means.

Tom: Yeah.

Will: And we’ve had personal experiences of this. I’ve been referring to a time I was in Boston and I did search for breakfast. Just typed the word “breakfast” into Google without even thinking  about it.

Tom: Yeah.

Will: This was not me testing out anything. And it worked. Instead of getting a definition of breakfast on Wikipedia, which is probably what we would have got a few years ago, I got great places to have breakfast near where I was at that moment in time. It knew where I was. It knew I was in a new city on the right device, etc. We’re getting these personal feedbacks, but we’re seeing this kind of growth of the total information being passed to the search engines coupled with potentially in some cases the decline of the explicit keyword part of the query.

Tom: Yeah.

Will: And we think that this probably has implications for all kinds of measurement and I guess the culture of Google as a whole.

Tom: Yeah, and the culture of SEOs.

Will: Right.

Tom: Because for the last decade or more, the number one word you associate with SEO would be keyword. People were very keyword focused, and we’re not at the point now where some people talk about the death of keywords and everything like that. We’re not at that point. But we’re certainly seeing a shift away from it, and Google are doing the not provided thing they say for privacy reasons, but I’m sure there’s other aspects of that. But also the whole keyword focus, we need to start moving away from that.

If you’re reporting on rankings for your clients, or you’re looking at their analytics and what keywords they’re ranking for, that’s becoming less and less useful. Now, if you’re a small Indian restaurant in Finchley and your SEO comes to you and says that you’re ranking really well for restaurants, you’re probably going to be like, “Whoa.” But that’s because you’re ranking for restaurants when someone within half a mile on a mobile device is searching for you.

Will: A subset of users. And right now that’s not an easy thing even to diagnose actually.

Tom: Yeah.

Will: So I guess probably where we’ll wrap this up is, looking to the future, we don’t really know what the measurements are going to look like, what the tools are going to look like, or actually in detail what some of the signals will be that will influence how you rank different kinds of implicit aspects to queries. But that’s what makes the whole game fun.

Tom: Yeah, I think the first step is for us as SEOs to start educating our clients and changing client expectations.

Will: Showing some of these queries. .

Will: Showing the change in user behavior and taking advantage of . . . there will be undoubtedly be tools and data that we can use to power this stuff as we go along. Well, thanks Tom. Thanks everyone. We’ll see you soon.

Sign up to receive the results of two of our most surprising SEO experiments every month