Latest / Semantic SEO Podcast / Query Augmentation & Query Semantics Explained | James Dooley Interviews Paul David Truscott
Transcript
- James Dooley: Query augmentation within Google Search. A lot of people talk about semantic SEO and specifically about query augmentation. Today I'm joined with Paul Truscott. So Paul, let's jump straight in. And what is query augmentation for anyone that wants to know with regards to how Google is passing queries? Paul Truscott: Okay, so query augmentation, it's not dissimilar to query fan-outs in a way that people are hearing an awful lot about now in, in the context of, uh, large language models and, and the way that they deal with, with when people are searching queries on those. Basically, what Google's doing is it's augmenting the query. So basically expanding the query into subqueries, which it... Google knows from... Google's in a privileged position, okay, James, because it has an awful lot of telemetric data, both from people's Google accounts and Google Chrome. So, it has access to all people's query pathways. So, it knows what people have searched before and after, uh, sequentially in queries. So, it understands what they're looking for, um, how they follow up. So it doesn't... It doesn't only involve what they're actually typing into Google, but where are they going on their search journey, right? And so as they go on their search journey and they're looking at different things, uh, within, let's say they land on a website or a Google Business Profile, they start clicking around on stuff. This feeds back data to Google. So Google understands, okay, what are the subtopics that people are interested in within their journey to arrive at a decision? It could be whether... Whatever the end search intent is. So whether that's to purchase something, to find out commercial investigation information, general information, or is it a navigational search? Are they just trying to find a company? So whatever the end result is that they're trying to achieve, where do they go on that journey, right? And then Google can augment a query based on all of this and it can also infer things too. So there's a... Obviously, you can't know. We can't know. Google has this aggregated information, but we don't know as, as SEOs what those journeys are in aggregate of people. So we have to take a more educated guess. But if you look at the SERP result, for example, and you look at the SERP page, you look at the autosuggest, that gives you ideas of how to augment a query, right? It might give... If it's local query, it might suggest some other places in there. So, you might want to augment the query to include those other locations. Um, then you got to look at, um... You found out what these are actually called. I call them SERP filter chips, but it was SERP something. Um, I can't remember the name of it. You may remember, but they're little chips that you sometimes get. You don't get them for all queries, but they sometimes appear at the top of the page. They're little bubbles and they've got things in that might be like, um, top rated with index distance if it's a local query. Or sometimes it's the colours of things. These are all formed by Google by people's search behaviour. So they're not that people are searching for these things. They're not putting that in the search bar, but Google infers it from all of the places they visited. And so, for example, if you were doing a... Um, if you, if you were doing a page for a product and you saw in these SERP filter chips that Google had put all these different colours in, let's say there were 10 colours, then you want to make damn sure that you augment your queries to include those 10 colours on your page because that's Google telling you these 10... These 10 colours are important because people are visiting things. They're visiting products or pages that include those colours. So, it's basically just taking the, your seed query and expanding it into other things. So, if it's like a product, you're expanding into sizes, colours. So, essentially attributes and values again, right? You're covering more attributes and values of those attributes. This is how you expand a query out. And the more expanded you do it, the more chances you are to be a match for whatever somebody is actually looking for. Um, because you're... You've, you've covered everything. Now, you have to balance that against making your page so long that the cost of retrieval is high. So, it is always a bit of a balancing act. Um, you don't want to go too far. Um, I have... I'm not going to delve into it here because we haven't got time, but I've got my own sort of ideas on how to structure this within a site so that you don't get in the way of the transaction. Um, but by having that information on your site is certainly a good thing to do. Uh, by just... Yeah, query, query augmentation. Just think, like, augmentation, it's an awkward word. It just really means expansion. Just expanding it. That's it. So query... You're saying there query augmentation is expanding upon the queries, almost like how Google query... So how query fan-out within AI, um, extrapolate synthetic queries. This is extrapolating subqueries to try to find out exactly what they are based on sequential searches and previous user experience. The biggest difference with Google is that they have access to that telemetric data that all of those LLMs do not have. So Google is at a distinct advantage there. It understands things in aggregate about search behaviour that all of the LLMs simply don't understand. They, they, they are guessing really, um, compared to Google who knows this stuff. James Dooley: Yeah. So someone, someone once explained to me query augmentation is being similar to PPC where you've got exact match, phrase match and broad match. And when you first have got a website that's not really trusted, unless you've got the exact query string match on that page, you might struggle to rank. However, almost like a BM25 query string matching, as you start to grow in brand and trust and you're getting clicks and you're building up topical authority, at that point, it's almost like you're moving through the tiers of exact match to phrase match, where you might not have exactly the exact phrase on that page, but you've got some words on there, they understand what you mean, and then you move to that. And then you move to the final bucket of broad match where you, you literally might not even have specific words that someone could be searching for, but they're saying, "We know there's a search intent match," and then we'll show you for it. How do you, right, move from a BM25 query string matching you where you're having to get the exact match to be able to rank a single page for thousands of different keywords that might not even be on the page? Is there a way of you being able to expand the ring for query augmentation? And if so, how? Paul Truscott: You've, you've hit on something that's really, really important that I think a lot of people, even SEOs, don't really understand this. You've got... I would say there are two camps of SEO. I mean, obviously there's lots of people going to be in the middle, but there's two thoughts in SEO. One is that nothing's really changed, right? And it's all still just keywords, right? And then other people at the other extreme are, no, no, it's all semantics. It's just Google doesn't use keywords anymore. It's all topics. It's all entities. It's all semantics. The truth is they use both, right? Google has a ranking pipeline and their first tier is to use lexicals to see if... So that's basically the words on the page to get matches, right? Because... And the reason they use that is because it's really cheap to do. And that might, that might give them in the initial stage of the pipeline, that might get the query down to, from say 10,000 candidates to a few hundred, right? Then they can use semantics. We've talked about this before where we're confident that there's certain low-volume queries where they're not using semantics because of the cost, right? So this is where the idea of coming in with, if you've got a... Um, if you're trying to compete with a, a more established brand by going more longtail. So these are... If we use the analogy again of, say, the product with different colours. If we try to rank, uh, for pink shoes, this pink shoe with, with certain colour laces, right? I mean, I'm being silly, but, like, if we're going to take it to that extreme, there's... Google are almost definitely not going to be using the big models that they have to process that query because it's such low volume. All right? So by being specific on your page and being such a good match for that query lexically, you can outrank more authoritative pages or more authoritative sites. Okay? And you do it by being... You, you drill down quite deep on that one specific area, right? So that's how you would do it if you're not an authority. So that's kind of the longtail, you know, what we used to refer to as longtail queries in SEO. And then over time, you can build those up. Um, you start getting traffic for them. You can, you can, um, have links to the page with the, the more broad query on and eventually maybe even just, um, redirect that page and take that information and put it on the main page. Now, if you're a big authority, if you've already got topical authority and, and really, you know, domain trust with Google, you can build out that page and have that longtail on the same page as the broad query and you're going to probably rank for it, right? But you can't do that out of the gate if you're a small business, small new website. So augmenting a query, if you're already an authoritative site, you may be just augmenting that query and all of it's on one page and you'll rank for everything by augmenting it because you've covered the whole gamut of this topic. If you're not an authority, what's probably going to happen is you've just made the cost of retrieval for your page too high for your authority level. Like everyone's got a crawl budget, and if you're a brand new site, that's going to be quite small. So if you've got a page with tons and tons of information on it, Google just aren't going to retrieve it. They're not... You just don't have the budget yet. So it's, it's a building block thing. Um, so that, that's, that's my experience of it anyway. Um, I know Koray Tuğberk has a different idea, which I get, which is the quality threshold. So increase the quality of that page. Um, and, but then you're still having other pages support it. So, it's kind of not a dissimilar idea, but it's, it's just a different approach to doing it. But I think either can work. I think either way can work. It's just, like, to do it Koray's way, I just think you have to, out of the gate, have a much deeper understanding. So, it's harder to do. Much harder to do. You, you need a much deeper knowledge of semantics to get away with doing that. Um, whereas to do it the other way, which is kind of like the Avalanche, um, theory of SEO, um, but also what you've got to bring into that now for the LLMs is making sure that that information is structured in a way that all of those LLMs can retrieve really easily. So, and the key with that is making sure that every... Where you're augmenting the query, make sure that every single section can stand alone. So if you just pluck that passage out, does it make sense on its own? If it doesn't, then an LLM is not going to use it. James Dooley: Yeah. I think an interesting part for me when I started to research a lot more with regards to query augmentation was how it is query-based and not term-based. And what I mean by that is there's certain topics where the word cheap, right, is a direct synonym with best. And they... To have cheap is, like, for flights and stuff like that. They seem to be cheap and best bring the same back. So, Google connected the word cheap with the word best. However, when you then go and do the word cheap and then you go and put it with, let's say, hotels, it's bringing back a completely different... It sees it as two completely different kind of attributes of what you're going after. Two completely different modifiers. And it says best is definitely not the cheapest and therefore you can have a premium hotel and you want to pay extra. You want a Michelin-star restaurant. However, with certain topics, they connect them. And I think that is related to what you're talking about there because a lot of people in the semantics, they talk about users, documents and queries. And they're the three ways of how they go and start with sequential searches and how they bind things together. And the user for certain topics could, could be appended together. Okay, that means the same thing. However, in other topics, it's not. It could mean something completely different. And I think people need to understand that with regards to query augmentation, that when they're looking at how they augment... Like Koray's got a great example with the, the voice-to-text, um, kind of audio-to-text kind of case study where he's ranking for thousands of words, right? Thousands of queries that are not even on the page. So it's got, like, best and top and all that. And it's like he doesn't even have that on the page anywhere, yet he's still ranking for it even though it's not on the page because the user experience... People are clicking through to it. The dwell time is great. They're having a great user experience and therefore they're saying we're now from a broad match going to open up for everything related to MP3 to text or voice to text, audio to text, everything, every possible variation because he's built that quality threshold up. He's been able to get that. But why? How do they determine whether cheap and best goes in the same bucket or not? Is it literally down to the users and how they interact? Paul Truscott: It's what Koray referred to this many times. Query semantics, right? So query semantics versus dictionary semantics. So what you conjure up in your head from a dictionary or encyclopaedic, uh, definition of something, um, isn't necessarily how Google sees it, right? So if, like, if we go back to your example, because it was a good example of saying cheap... Um, what, what did you, uh, couple that with? So cheap flights and best flights kind of came together, right? James Dooley: Yeah. Cheap hotels and, and, and best hotels came different, completely different sets. Paul Truscott: So let, let's use that analogy because it's a really good one. So if we look at flights, generally, it's commoditised, right? So, if I pay $200 for a Delta flight from Orlando to San Jose and someone else pays $500 for the same flight in the same class, right, then there's no difference between what we get. The only difference is what we've paid. So that's why that query, for example, can end up being synonymous because if you're, if you're getting the same thing but you're just getting it cheaper, then that becomes best because getting the same thing for a cheaper price is the best result, right? But if you're talking about hotels, cheap would infer in that instance, because there's such an array of hotels, cheap would infer a lower star rating. Like a seat on an aircraft, if it's an economy seat, doesn't have a star rating. It's the same. One economy seat is exactly the same as another no matter how much you've paid for it. But in a hotel, even within the hotel itself, they might have different types of standard rooms. Some with a view, some without a view. There's even, in bigger cities, rooms without windows, right, that you can book which are cheaper. So, it's much... It's a much more difficult and nuanced query. And you're right. In that instance, cheap would probably infer a lower star rating, a hotel in perhaps not such a desirable location, etc., right? You wouldn't be just looking for what's the cheapest room within a certain hotel. That's a different query. So, Google's looking at query semantics. They're looking at, um, what people are searching, not what does the dictionary say it means. So if enough people are searching for something in a certain way and there are enough credible sources that are responding to it in that way, then that determines the definition for Google of what that thing means, not the dictionary. And so it's very, very different. And that's how you arrive at these seemingly, um, sometimes strange results in Google or even outright, um, wrong results in Google. And it's to do with query semantics that there's just... And that's another thing. If we go back to when you're structuring pages, if people rely on ChatGPT or Claude or whatever to give them ideas for a web page instead of doing the research with Google, it's going to be a guess and it's going to be based on dictionary semantics. If you... One of the things that Koray talks about, the difference between topical map and a concept map. So, if you get an LLM to do a topical map, and there's loads of services out there, for example, that sell topical maps, and when you get them, what they'll be is a concept map because they've just got an LLM to do it. So, you have to be doing this with Google's help so that you build your topical map based on query semantics, not based on dictionary semantics. James Dooley: Yeah, that's the key. Yeah. Anyone who's watching this episode about query augmentation, let us know if we've missed anything. Let us know what you're trying to do to expand upon the queries across your whole website of what you're trying to rank within Google Search. Have me and Paul missed anything off? Leave a comment in the comment section and let me know what your thoughts are about query semantics, or sorry, about query augmentation. We hope you like this episode and Paul, thank you very much. Paul Truscott: Thank you, James, for having me.