 Hello. My talk is about keyword research. Trash in garbage out. However, so I've noticed that there's quite a lot of my fellow Brits at MozCon this year. You've seen, thank you, you've seen a few of them on stage already yesterday. And Rob, of course, although he's kind of half and half these days. But fortunately, because I'm a, as Rob mentioned, I'm a Brit, but I've been working for over a year now for an American company. I'm able to provide translations throughout. If something confuses you, you can come find me later. I'll explain, not just in my talk. For example, this is a dust, oh, this, that wasn't, this is a dustbin lorry. This has nothing to do with trash or garbage or trucks. None of those are English words. Dustbin lorry, a bit of a weird one, like there's a bin you put dust in and then a lorry. Anyway, indeed, this was my original title slide. Rubbish in, rubbish out. Now, I was cautioned against this by my well-meaning colleagues. They're backstage now, head in hands. Fortunately, though, this isn't a talk about English variants and dialects. That is an interesting topic, and I would love to talk to you about it. But there's, it is also a keyword research problem. But there's a bigger problem with keyword research that I'd like to discuss. I think it's worth talking about the stakes here. Keyword research is, although we consider it a fundamental task, it's incredibly high impact. So not just marketing tactics in SEO, but in other disciplines as well. So I'm thinking about things like title tags, headings, briefs you send to copywriters. But also, if you've ever done that sort of back-of-the-envelope spreadsheet forecasting and say, oh, what if we ranked for this keyword or that keyword? If we ranked in this position, how much more traffic, how many more conversions would we get? Often, the strategy itself will be chosen or honed based on keyword research. We might decide which site section is important, which tactic to pursue. And I've often even seen businesses decide which product or service or even locale to invest in partly based on keyword research. So this is important. And yet, if there was one SEO activity that you could imagine being delegated to the most junior member of a team, yep, this is a problem. Now, I've made this mistake myself. I've been both the intern and the person delegating it multiple times. But it's probably a mistake. Not because interns do bad work or anything like that, but just because it kind of betrays that we perceive this as grunt work, which it should not be. Keyword research and particularly senior bind for keyword research is really important. This is just an illustrative chart. So I used to work, when I was a consultant, I used to work for a long time with a client in the UK in the online flower delivery space. These are just some equivalent US businesses, so that we're on the same page you know the kind of company I'm talking about. But you can see on this chart, which is of the US market, how many keywords these sites tend to rank for. So flower delivery is not that complex proposition compared to some of the businesses that I've talked to people from while I've been here in the last couple of days. But you can see tens of thousands of keywords being ranked for by some of these businesses. And it was similar in the UK example. And yet, my client only cared about these two keywords, which they ranked pretty well for. This story sounds familiar to you, right? You've all had clients or bosses or perhaps suppressed parts of your own psyche that are obsessed with a head term or two. In this particular case, these keywords represented 2% of organic traffic, which is not untypical. But this is my failure, right? So much money spent, so many consulting hours, engineering tickets, C-level meetings, hundreds of reports, all of this and all they care about is 2% of traffic. And the root of this evil is keyword research, or lack of buying keyword research perhaps. This is what I'm talking about today, or maybe a little bit ranting about today. I think we, and in we I include myself, I think we have failed to give keyword research enough thought or seriousness. And I'm going to try and break that down into three main critical failings and then talk about what we might do instead. So the first thing, oh, I just want to make sure on the same page. While I was preparing this talk, I Googled a bunch of keyword research guides in the industry. I found all the sites you'd expect, and I just had a look at their process, and it's all fairly similar. So this is what a typical process apparently looks like. So you start by getting an absolutely, I shouldn't swear, a large number of keywords. Maybe you use a keyword multiplication tool like this, or some sort of vast ideation project. Then you might throw in, if you're smart, some like Google Analytics site search data and some Google Search Console impression keywords. And when you've got, oh, of course, you use your favorite keyword suggestion tool, MozPro, then you put some numbers next to them. Now, in my experience, sadly, it doesn't particularly matter what numbers you put next to them. It doesn't seem to matter people what numbers are used. I have a weird one. It's almost just for the sake of it. So you might start with something like volume, which is okay, but then it goes a bit left field from there and quickly we're in another industry, and it just gets more and more obscure, and then you just don't know what's going on or how it applies here. Anyway, once you've got this fast out of the way, you then get on to grouping the keywords, which might imply some sort of prioritization, sometimes more or less structured, kind of according to the author's whim, what I actually see people doing is just chaos. Every system at once, group names have no relation to what's inside, all of them are tiny, nobody's cleaned up since 2015. Oh, well. Okay. So that's apparently the consensus process, and yes, I am sharing a SERP, where we're outranked by HubSpot. Oh, well. More failure. So that's roughly the process we're using right now. So like I said, I'm going to break this down because it's going to be critical failings, and the first is what I perceive to be, you all including me, wasting our time. So that's this bit, right? Where we just produce this enormous list of keywords so much that we complain that Excel doesn't have enough rows. No. This isn't sensible. Let's use the Moz blog as an example. So on the Moz blog, which gets a decent amount of organic traffic, as you can imagine, 48% of our daily traffic comes from keywords that are sending us one click. And what's in that list of keywords that is only sending us one click a day varies a lot day to day. It's not like it's always the same one-click keywords. And that's not even counting the various low-traffic keywords that Search Console, this is Search Console data. It's not even counting the various keywords that it samples out for privacy reasons. Indeed, Google themselves, they say that 15% of the searches they see every day are totally new to them. So does it really make sense for SEOs to be compiling these exhaustive lists? Not exhaustive at all. They can't be. That's impossible. Now, those of you who know me well will understand how difficult it is for me to deliver you this message. But realistically, you are not going to capture every possible keyword. You're just not. It's not realistic. I'm so sorry. So what are we actually trying to achieve? So I'm going to look at some potential flavors of keyword research and how we might think that this method works. So first I'm going to look at keywords that we already ranked for. So this is the most simple case, maybe a site migration or something like that, or a site section migration. We're just interested in the keywords that currently drive us traffic and at which pages should target those keywords. So we're asking this question, how can I best assign pages to the keywords that currently drive traffic? So what's the value of the massive list here? You only need to capture the intense so you can map pages to them. You do not need to know that in 2016 12 people searched for Moz with 3 Zs sorry, 3 Zs and an umlaut. That would tell you nothing about customer journeys. If you just want to know the keywords that people currently use to find your site, that's not difficult. Okay, so not that. We're asking keywords that I currently rank for, but also keywords I should rank for. Perhaps because I'm setting up rank tracking, for example. Well, in this case, you know, I might be interested in the keywords my competitors rank for as a way of sort of approximating that. But again, this doesn't have to be some massive job. Keyword overlapped tools are a thing. Obviously, I'm showing off Moz's new one, but others do exist. It's slightly embarrassing, by the way, while I have that screen shot up. This was a virtual Mozcon. My colleague, Dr. Pete, talked about how a post of mine on Moz.com had vastly improved our rankings for various disavow-related terms. One year later, it's now on our list of keywords to improve. Sorry about that, Pete. So that didn't justify all that effort. But surely it's this last example. It's keywords we could rank for. Maybe we've got a new page, a new site section, new product, new website. The open ocean of keywords. Maybe in this case, we're interested in that expansive method. Well, these days, a lot of people rightly talk about scraping SERP features and this kind of thing using specialist tools to get better ideas for keywords to further expand that list. But you realize you can just do this in your favorite keyword research tool that I mentioned. Not only this one, by the way, I should say, but by default, these kinds of tool include autocomplete, related search, people also ask, topically relevant pages. If you just used search console on one other tool, you'd have skipped all those arcane steps and you'd have all the intents. Now, this is the key, right? Are you trying to get all the intents or are you trying to get all the keywords? I would say you're trying to get all the intents. You cannot get all the keywords because it's a fool's game. We already talked about this, right? Many of them are totally unique, new. So you're not going to get them. Now, you can show that an intent exists in 500 keywords or the two main ones. After it's not 2008, right? You're not going to produce a page for every single keyword for one intent. Or at least you shouldn't. Please don't. How many hours have we wasted of our mortal lives analyzing enormous lists like this, which ultimately contain about four intents and a bunch of relevancies? I don't know why I went for a badger theme through all my examples in this deck. Doesn't really matter. Anyway. Hang on a minute, though. Hang on. Wasn't I just saying there's no way you'll capture every keyword variant? So you should use these easy methods to capture the intents. Didn't I say not to get an intent to do this? Well, I did. And that's because what I'm saying here is not that this is easy or trivial or you can take shortcuts. What I'm saying is that people over-emphasize the grunt work aspect of keyword research, which is not the difficult bit. Prioritization is the difficult bit. Do not tweet this. New last-minute addition to MozCon, TNCs, you may not tweet this. That is not what I said. I said this. Do not gather thousands of low-volume keywords for one intent. I'm sorry. It's a waste of your time. So the trouble is whatever you did there, sad reality is I don't think it'll matter. And the reason I don't matter as I just mentioned is because we'll probably prioritize wrongly. This is about this bit, right? This step, the one with the random metrics. And, of course, I am being facetious here with miles per gallon. Although I realized recently that when I was looking at higher cars, I'm going hiking. This is a bit of a tangent. Never mind. I'm going hiking after MozCon and that means I need to drive upstate in Washington and that means I need a higher car and I notice that some of the miles per gallon numbers for the higher cars are really weird and neither miles nor gallons are standardized metrics between the UK and US. What? Anyway, people actually do use cost per click. This is much less ridiculous than miles per gallon but I'm not sure it's that much less ridiculous. Actually, there is a time and a place for this metric, obviously. Now, most obviously paid search is a good time and place for this metric but I also don't hate it in rank tracking and reporting. You might want to subtly hint to your boss about saving them but in keyword research, not a fan. At Moz, we have a metric called difficulty so it looks at the strength of the competition in terms of page authority and domain authority on a SERP as well as the dominant Google features, this kind of thing. Basically, it's answering this question. How realistic is it for me to rank for this? This is the metric, this is the question, sorry, that people are trying to answer when they use cost per click in keyword research. If you're struggling to read this chart, that's because it's an absolute shit show. It's not you, it's the chart. PPC and SEO are not the same question. Please stop. Someone at the back is even allergic to this. But maybe a less controversial choice is volume. And again, of course, I'm not suggesting you should never use keyword volume keyword research, that would be absurd but this is useful in the right place but really, we're interested in volume because we're interested in clicks. It's a means to an end, right? So consider that on average, the word average is doing a lot of work, but consider that on average, 30% of searches will click on a position one result. 10%, these aren't super precise, but roughly 10% will click on a position three result. And yet on the Moz blog, on a typical day, huge chunk of our traffic comes from keywords where we have under 3% click-through rate. Now, you might be sitting at the back saying, Tom Kappa, I thought you understood SEO. You probably rank in 15th for some high volume term. Well, there is a bit of that. I'm aware that position 15 is quite low click-through rate. Some of it is that. But a lot of it is not. And you've heard about this from several speakers yesterday and you will continue to hear about this. It's becoming a bit of a theme for Mozcon. It's shit like this. We rank first for this keyword. Do you think we get 30% click-through rate? Hmm, probably not. Total click-through rate, by which I mean the percentage of people who click on something on the page, on the SERP, varies massively from keyword to keyword. Which means that volume is a bit of a suspect metric if you are not using it to get to clicks in some way, right? Because the actual volume may not translate very well to the number of clicks you can get. So this is just a random sample data set from from Mozcast, actually. Fairly representative. I threw 750 keywords from this list into MozPro. It's got an idea of how many keywords in this set had various click-through rates. I'm just showing you this so you get an idea of how varied this is. So 85 keywords in this list had a total click-through rate between 0 and 20%. So that means the vast majority of people clicked on nothing for those 85 keywords. And then, you know, as we look at various other click-through rate ranges, bit of a spread, and about a third of these keywords had the click-through rate that as SEOs we tend to assume they all would. Where the volume is actually the representative number. You can get this data for your own site from Google Search Console or for individual results or if you're interested in that, like, total for the SERP from tools like Moz. It's only natural, right? If we're talking about clicks there's something more nuanced than volume, though, that we might want to go a bit further. And I see sometimes people, if they're doing forecasting, like we mentioned earlier, they might start to look at using a similar landing page or something like this estimating not just how many clicks and sessions they get, but, like, how many good sessions they get. And then they start to look at metrics like this one. Dana touched on this a little bit as well. So some of you will realize where this is going because anyway, so if we have volume, then we're interested in clicks as being a bit better than that. Then we're interested in non-bounce clicks. It doesn't seem absurd, but it is. Some of you know I have a long-standing history of not liking bounce rate. And this has been in 2015. This has been in 2018. The short version is a bounce is a single-page visit. There's nothing necessarily wrong with that. The user may stay for 30 minutes, read your article five times, love it, mention it at their wedding, have it engraved on their tombstone. It's still a bounce. I wouldn't recommend you optimize away from fulfilling user intent. So why not bounce rate? Well, partly, it's generally a misunderstood metric, and this is sadly true for a lot of Google Analytics engagement metrics. But secondly, and this is sort of what Dana touched on, even if you choose a better metric than bounce rate, I think it's a mistake to discount impressions. I think people who just view your web page, actually, that's pretty good. Again, this has been a sticking point for me in the past, but it's also related to people more recently saying this. And I'm not saying, like, don't use top of funnel metrics. I'm kind of saying the opposite. I'm saying be careful about dismissing sessions which don't result in a conversion or many pages or something like that. There are whole industries around just spending vast amounts of money to provide an impression. Think about billboard advertising, TV advertising. As digital marketers, we shouldn't think we're above that. The trouble is, though, picking the right metrics, avoiding the wrong ones, it's not always enough. So far, these are sort of two of the main points along this road that I've made to you. One is that I don't think you should waste your time on massive keyword lists, and the other is that I think you should compromise for clicks rather than volume, for example, or bounces. Trouble is you can't really escape volume. Whatever tool you're using to get click estimates from, it's based on volume. And that brings you to some old problems. Seven years ago now, the late great Russ Jones observed that Google keyword planner volumes had become bad. There's two ways this happens just quickly. One is this one, you've got four synonyms or, in this case, misspellings that have obviously very different volumes. Google keyword planner will report them all as having the volume of the total of all the variants combined. Okay. The other way that this happens, if you go down a list of keywords sorted by volume, you'll notice there's these suspiciously round and distant buckets that sort of go down in big jumps. And if you put these on a chart, again using that 750 random keywords from the past, you can sort of see how these fall out, and at first it doesn't look too bad, but at higher volumes it gets a bit gappy. Like, do we seriously believe there's dozens of keywords with exactly 270 searches a month, but nothing on either side? No, we don't believe that. And of course the higher you go, the more ridiculous it is. So how bad is this in practice? Now, I'm telling you this because a lot of you probably do end up using straight keyword planner data in a lot of tools and a lot of context. So what does it mean to keyword research? So at Moz, obviously, a big part of my job, we have a vested interest in metric accuracy, search volume accuracy. It's one of the things that I do is benchmark it. So I'll compare it in this case, among other things with Google search console data. So in this case, if we have access to search console for some sites, we can say, well, if they're consistently ranking position one, then the number of impressions they get is roughly the search volume of that query, right? Now, there's a bunch of data cleaning that has to happen here. Like, what if they only rank first in one local or for a certain kind of user on some days, blah, blah, blah. But like, we do that data cleaning. Long story short, this is a rough thing I'm comparing with here. And before you tweet at me, yes, I have seen this article. So this has been true for a while now, but it isn't a problem here. So this is various keywords are redacted from search consoles, but this is keywords being removed and not under-reporting volumes for keywords that are reported. So it's fine for this method. So this is the state of AdWords data, basically. But within a... So I think this was on all and there's a slightly larger benchmark set than just the 750 I talked about earlier. So the source of truth data suggested that they had on average about 100 searches a month. And it could have been anything, right? In this sample, it happened to be 100 searches a month. And the MOS data, which I'm pretty pleased with in this case, average that about 100 also. And then a couple of our competitors who also don't use straight AdWords data were also in the right order of magnitude. And then this, the AdWords space program. This is because of the issues I just mentioned. Now, to be clear, I'm not criticizing the tools, many of which I love, that use straight AdWords data. There's good reason for that. Very few platforms have the scale required to do something meaningfully better and still give real-time answers. What I would say is if you're making decisions that really matter, based on volume data or click data, pick carefully. This goes back to one of these, right? Imagine you're doing any of these things and you overestimate by a factor of 18, your boss is going to be happy. This is an 18 times overestimate, right? If your forecast is out by a factor of 18, I don't think my boss wouldn't be amused and she's very forgiving. I want to share a couple of the other metrics we look at as well. So one is correlation. So this shows the metrics might be in the wrong proportion, but are they roughly in the right order or right sides relative to each other, even if the magnitude is off? So for example, on this chart, these metrics are well correlated even though one axis goes up to 1,000, one goes up to only 10. Whereas these wouldn't be. So again, the three SEO suites I'm comparing here, they all roughly manage this. Oops. What happened there? And lastly, the last one I want to share with you that looks at the sort of other side of this is, okay, so the keywords are in the right order and they're at proportion to one another, but how consistently accurate are they? Like, how bad is a typical mistake? So, again, I think Moz does okay here. This is what a bad error looks like in Moz's search volumes. 123% out. Obviously a number I'd love to bring down. But you can see at best this is not a precise science. Nobody does amazingly. But obviously you knew this was coming. There's the space program again. So if we look at Google AdWords data for these kinds of big decisions, you're going to get an 18 times total overestimate. Keywords aren't even in the right order relative to each other. You don't rank them correctly. And often when they're wrong, they're really wrong. Now, as you can probably tell, I'm very deep down this rabbit hole. I love to talk about these kinds of metrics, so I won't go any further in this presentation, but I'd love to talk to you about it afterwards if you catch me. However, I did say I would do this. We'd talk about where we should actually go with all of this. Good keyword research. What am I actually asking you to do? So firstly, quality not quantity. Remember this chart? It's pointless trying to be exhaustive with keywords. Focus on intents. Do not miss the word for the trees. Secondly, clicks. Don't assume that because you know the volumes and rankings, you have a rough idea of clicks. You certainly do not. Thirdly, don't take accuracy for granted. Just because a tool can show you a number doesn't mean it means anything. Remember this again. So the overall message here is this is important. You shouldn't take a chance on lazy metrics or methods, and you shouldn't comfort yourself by just adding more and more rows, but there is one more, possibly most important take away. In Britain, we call it a dustbin lorry. Thank you.