SEO Experiments That Prove Everyone Wrong - Google Leaks, AI Myths & Ranking Signals (Mark Williams-Cook)

Mr. Mark Williams-Cook, you are a legend in search engine optimization. You are very well known and super interesting. And before we start, could you share your background?

Sure. So I've been working in SEO now for just over 20 years, I think 22 years, which could mean one of two things. It could mean either I'm really good at it or I'm really terrible at it, which is why I'm still doing it. But my kind of background getting into SEO was from like a technical approach really. So I didn't have any formal qualifications to begin with in marketing or anything. I was really into computers, doing programming, that kind of thing. And I think like a lot of people in that time, I just fell into SEO because there wasn't really any courses about it. Um, it wasn't that well known. It actually came about because I started building my own websites, putting them online when I was about 13 or 14. And I had a website for kind of local music in my area to put bands on and things like this. And as it happens, one of my friends had helped me install one of the first like Amazon shops you could have on the site. It was a site running on PHP Nuke for anyone that's that old. And one day I noticed I was making like £50, £60 a day in commission through this shop. And I was like, huh, how's that happening? And we looked at AWStats, which was the thing at the time. And I found most of that traffic was coming from Google. And that's actually the moment where I was like, huh, that's really interesting. How does that work? And then I started talking to people about that and ended up getting my first agency job. And those times, which was like early 2000s, I guess, was, um, an interesting time to work at agencies because most of what I was doing was like, here's a list with 2,000 directories on it and I need you to go through and submit the site to all of these. Um, but that was really good. Um, I kind of cut my teeth there and I was at agencies for about 2 or 3 years. And then I was also then in the background building my own sites. Um, and there came a point then where I was making more on my kind of affiliate stuff and my made-for-AdSense sites than I was actually at my full-time job. So I did that for a while, um, full-time. So like affiliate stuff and, um, kind of ad revenue sites, which was great. At that point I was like, oh, I've, you know, I've really got this nailed. I was like early 20s, um, you know, sitting in my underwear eating cereal, being like, this is the life. And then I think after—

Seems great.

Yeah. After I think about a year of doing that, Um, I think I just kind of got really low. It was a bit like, oh, I hope that pigeon comes back today. Um, and actually the best— it was really good for me. So I actually went, um, to another agency, and that agency was quite different. And I was surrounded by people that were like much better programs than me, people who had like years of knowledge in marketing. And then I got exposed to all different types of businesses through clients we worked at. And that was really then when I sort of learnt that kind of SEO. Because there's two— for me, there's always two kinds of SEO. The kind of SEO like I do now at an agency for clients, very kind of long-term, um, focused SEO built around brands. And there was more the SEO I was doing for myself, which was how can I rank as quickly as possible and make as much money as possible? Um, so yeah, so, um, I'm now a director at an agency called Kanda, which is based in Norwich in the UK. We've got 25 people and all we do from a marketing point of view is SEO, nothing else. And I've got a SaaS tool called Also Asked, which is for intent research. I've had about 4 or 5 years. And we've also— so Kanda owns a UK e-commerce business as well in the pet sector. So that's most of my time between those things. And we've got other little kind of side projects we run just to keep experimenting, really, because it's not always safe, obviously, to experiment on client sites. But I like seeing what happens when you pull Various levers.

I'm always sharing you because you do very interesting experiments and then you share the results on LinkedIn and oftentimes the results are not what people expect. Like you just, I just made a video. I don't know if you actually saw this video that I made. I put it out a couple of days ago about your like duck. Yeah. T-shirts.

Yes, I saw that. Yeah.

It was so good. It was so, it was so, and you've done so many. I've made, but it was like, I've made several videos about the experiments that you've done. Because you're always sharing things that are unexpected in the SEO world.

I think the reason I do that, um, is more, uh, out of frustration, I think, for me, which is that it's— there's so many resources to learn SEO online now. Like there's video tutorials, there's paid-for courses, there's dozens of books. You can get coaching, mentoring. Like there's, there's so many different ways to, to learn SEO and Especially now with large language models as a vector to learn SEO, you're really getting there just the thing that is said the most. And I've encountered SEOs who are like, oh, this works like this. And I'm like, that's interesting. How do you know that? And the answer is because I asked an LLM. And it's like, well, that's just going to tell you. And it creates like a feedback loop, right? So you read the thing on the LLM, so you think that's correct. So you go and write that everywhere and then the LLM hoovers it up and then you get this massive consensus. So yeah, the schema thing was interesting for me just because it's, as I said, when, you know, you explained in the video, the demonstrated proof was that, okay, it's picked up and it's not on the page. So it's reading schema. And I just wanted to prove it, it, you know, does it the other way round as well. So it's kind of like faulty thinking in a way. And I just think as an industry, it's good. It's healthy. to challenge each other because it makes us better at what we do because Google doesn't care if we get that wrong. They're just going to be like, huh, well, never mind. They're going to go and do that and think that. So I think it's always worth, and I do it to myself as well, and I invite people to do it to me, to be critical, not of me, but of my ideas. I'm very happy for you to be like, this is complete rubbish, as long as you can give me a fairly logical reason or some evidence as to why, and I'll change my mind if like, if you can show me a great reason, right?

This method of marketing is so effective, I had to make sure it wasn't against Google's rules before I kept doing it. It's a form of SEO I call compact keywords. Whereas most SEO focuses on putting up articles to answer questions— how, what, when— compact keywords focuses on putting up dozens of pages that sell to searchers who are actually looking to buy. These pages rank on Google and convert so much better than normal that when I discovered this years ago, I couldn't believe this was allowed. It's less work too. The average Compact Keywords page is only 415 words. Compact Keywords is a 13-hour deep course on getting sales with SEO. A customer recently said, each lesson is dense with information. You're giving years' worth of experience boiled down into 15 to 30-minute lessons with no filler or fluff. feel like I'm gaining a new superpower. Compact Keywords is about setting up an SEO funnel that brings you sales for years and years and years. It works with AI. It's less work than traditional SEO, and it makes way more money. You can get it now at compactkeywords.com. Back to the podcast. And for the experiment, for people who don't know, is you made up a bunch of like just completely invalid made-up schema. Like things like, like, I don't know, it was like waddle style aggressive. Like it was all just like complete nonsense. And the LLM treated, I guess it was ChatGPT, just treated it as if it was any other text on the page. And so the fact that it was schema didn't actually matter.

That's right. Yeah. So I think there's a lot of interesting things happening at the moment, especially around the LLM, geo, whatever you want to call it side with, you know, schema is an interesting one. Because we don't have in detail exactly what happens during pre-training for these models. You know, we know, um, a lot of these appear to use, um, like Common Crawl data, um, and there's been some really interesting studies to show that's a good theory in terms of how often sites are in the Common Crawl data and how often they're kind of cited, especially when the when the generated answer isn't prompted. And we know the Common Crawl data does things like it strips out the HTML because it makes sense, right? If your large language model isn't storing the training data, it's storing the probabilities of the chunks of tokens, which ones are coming next. So it doesn't necessarily make sense to include lots of noise if you're trying to model a language. It's a language model, right? So it makes sense to me that at that stage, at least, schema is, or most HTML is stripped out. Um, it doesn't appear to be the case when the actual kind of user-driven bot live goes to the site. And again, I wouldn't say even that test I did conclusively says they don't use schema in that ChatGPT, OpenAI could have a separate system that's going around and reading structured data. I'm not saying that's impossible. What I'm saying is there is zero evidence for it at the moment. So why would I believe something that could be true? Anything could be true, right? So that's just where I try and base my time and my thinking. And it's difficult as well. So I'm sure you just saw Cloudflare have announced that feature for the markdown files, right? LLMs. And that just seems bonkers to me. Um, I think it was Profound did a study, um, where they looked at how often the content was, uh, cited and they did a control test with Markdown, without Markdown, and there's basically no difference there. And it's such a simple thing to do, to take the content out of a page. You know, when you've got these tech companies with tens of millions or billions, sorry, of investment, hundreds of billions of investment, sort of thinking, oh, they need a Markdown file. There's no way they could get the content here. If I can do that in like 15 minutes, they definitely can. And I don't think the arguments around, oh, it takes less tokens, are particularly valid. So I'm interested what's happening there. I don't know if that's a move similar to how lots of kind of the SEO plugins now support llms.txt, basically because people want it. Yeah. Um, and there's no cost to doing it. So it's like, why wouldn't you do it? You know, so when you've got a feature list, but it irks me slightly when I log into, you know, enterprise-level platform and it flags up a caution for my website. You know, there's no llms.txt present. I'm a bit like, well.

The problem with all of that is that it distracts people from what actually matters. And then, you know, they go and spend, you know, like LLMs at TXT can be very easy to put up, but like lots of SEOs really hyperfocus. And so instead of spending like an hour on it, which you shouldn't spend any time on it, you shouldn't even use it. It's not necessary. But like lots of SEOs will go and spend like days on it instead of like doing keyword research and refining, or instead of like thinking through search intent, things that are like way more important, right? Like, and it's just, and then, you know, it sounds really minor, but it adds up to all these little things that other people will say matters that doesn't matter. And you've just spent months or years focusing on the wrong things that aren't actually going to drive results.

Yeah, especially with— this is especially the case with larger companies because again, I've heard that argument of, um, you might as well do llms.txt. It, you know, it takes half an hour to do. And again, if I was running my own affiliate site, I may well do it because yeah, it's a half an hour thing. If you've ever worked with like enterprise-level organizations, there's like whole machinery that needs to be kicked into action sometimes to get stuff done. It's like, okay, so we want to do llms.txt. We need to make a little business case. It needs a Jira ticket. The Jira ticket has to be prioritized.

Yeah.

3 people have to read it. It goes through into dev, into dev queue. Then that's got to be QA'd. It's like, it's, and like you said, it's all time. Which is why I think one of the other things I'd posted this or last year now, which was a little controversial, I don't even know why it was. I said for lots of clients now, we don't recommend by default even bothering to write meta descriptions. And the reason for that was, and there's data, there was 3 studies. Um, I call them studies. I say at least evidence. They weren't like set out studies in all cases. Um, but tests were done where we were demonstrating 2 things, which is, um, there's the data that's already out there, which shows like 60, 70% of the time Google's rewriting the meta description regardless of what you write anyway. So it's immediate. You know, there's 50, over 50% odds that the work you've done anyway isn't going to be seen. But what was really interesting we found was the meta descriptions that Google writes on average had a better click-through rate than if it showed the one that was written by, you know, the SEO or whatever. And I think some SEOs almost feel threatened by that. Like, oh no, Google, well, it's done a terrible job on this e-com page. Like, yeah, it doesn't always get it right. But the massive advantage Google has got is that it can write the meta description based on the query. So it can pull the bit of the information from the page to match the user's query. And Google can obviously be testing that itself and can, you know, that's what it does. It's its bread and butter, right? Whereas you as an SEO have to write one static meta description to try and serve every possible search. So of course you're at a massive disadvantage. And what we actually found was, If you don't write a meta description, it vastly increases the odds of Google being like, oh well, I need to write one here because there isn't one, and it picks something better than you did. Yes, there are cases where it gets confused, you know, on some pages, and then yeah, go in and fix it, especially if it's a page that's getting, you know, impressions and stuff. But that's again one of those activities where I've, you know, consulted at companies And there's like massive spreadsheets open with people, you know, not even using AI. They're like handcrafting meta descriptions. And I'm like, wow, this is like dozens and dozens of hours of time that you could use to actually do something useful, right?

Well, actually, this is a good question. Doing something useful. So what, what would you say one thing somebody can do this week would be to improve their rankings?

It's so many to pick from.

I mean, like things that maybe like common things that just people don't do that they should be doing.

So I still think in the areas of SEO, for me, link acquisition is still the biggest driver. If you want something that's basically guaranteed to move the needle, it's still, you know, Google still uses a version of PageRank. The interesting thing that I found, so I published last year some of the data from the Google exploit that we found. Um, and the really interesting thing there was because we were getting back actual, not just like, so we had the content warehouse leak, right? Which showed us a bunch of variable names with very little context as to how they were used.

No weightings.

Yeah, the data that we got back from the exploit had the actual scores of specific sites attached to them. It had, um, the classifications of queries. So we knew the query, we knew how it was classified, we knew how it was weighted. And the thing that I found really interesting there was it was demonstrable proof that there were layers that you had to get through before you kind of got to the next race. Like, I'm trying to think of a good way to explain. There was like a prequalification criteria basically. So the easiest example to give was Google has a score they just call Site Quality Score, right? And we found if your Site Quality Score was below 0.4, you were not eligible to get things like featured snippets or appear in People Also Ask Questions. So you could have a site where they're like, oh, we haven't got any featured snippets. Oh, we need to work on our content. Nothing to do with your content. Does not matter what you do to your content. Doesn't matter how fast your page loads. You're not in the race for appearing in featured snippets. And the more we dug into this, the more it looked like there was various layers to this system and forks where, okay, if we've decided this is a medical query, so we're going down this route and then there's these different score criteria you kind of need to be hit. To hit. So the thing I concluded from that was especially things like sites hit by the HCU, clients generally always turn to, oh, we've been hit by the HCU. Um, the problem is the, the content. And it's not necessarily the content, it's the vessel the content is delivered in, which is the, the site, right? That's the thing with the reputation. Um, and there's things you can do to affect that. So I did another experiment on LinkedIn. Because my hobby site, um, I'd made about D&D, um, got hit by the HCU like really badly. And it, it did quite surprise me. So I did have some AI content on there, um, but I had paid writers who played D&D to write that. Um, I'd had it edited. I'd written quite a bit myself. Um, the site had a purpose, which is that over lockdown, um, to avoid going completely mad, I'd, I'd started playing D&D for the first time, like online with some friends. And I'd found that I felt there was a little niche for like explaining the rules to me without adverts everywhere and without having to dig through 20 different answers on Reddit. So I was like doing these kind of beginner guides. And it was the first time actually I'd made a site where I had 3 people email me in 6 months, like thanking me for the content. They were like, this was so helpful. Like we've printed this out for our game. And I was like, wow, like that's so gratifying that someone's actually You know, I don't think I've ever emailed anyone to be like, thank you for the content. But then Google disagreed.

Uh-huh.

So Google was like, no. And it, it took like 99% of the traffic away within like a week. Um, so I was a bit like, hmm, I'll wait and see if anything changes, you know, because Google sometimes gets stuff wrong and it kind of stayed dead. So then I was like, oh, I wonder. So I just took some of the articles and posted them on LinkedIn Pulse. Immediately they ranked number 1. They're in AI overview, they're number 1 organic. Um, and then that made me look more at LinkedIn Pulse. And then at that time you could see there was an explosion of, um, like loans spam, like gambling spam on, on Pulse pages. And it was pure AI content and it was just getting a pass. Because Google was like, well, yeah, you're a really trustworthy website. So in you come. Um, and the other cool example was, uh, around that time the Vatican website got hacked. I don't know if you saw this. And so the Vatican website ended up ranking, I think it was the top 3 in Google for CBD gummies, just because there was a single page on there. And it just makes me think as well about, you know, when people talk about things like topical authority, like what topical authority does the Vatican website have about CBD gummies, right?

Well, they have so much authority in general that they can just do anything.

Exactly. So this is this is kind of again, why, you know, where's that authority come from? It, you know, it is a combination of things, but like links is still like a massive, massive—

Can you, can you explain site quality score?

Yeah, it's actually really straightforward. So I got, um, when I got site quality score, I was like, oh, it's going to be about links, right? So I did, I took, I think it was about 200,000 sites I had the quality score for. And I did a Spearman correlation against all of like Ahrefs and Majestic's different link metrics. And it was just spaghetti. There was like no correlation there whatsoever. And I started looking at the sites manually. So in the top 1%, you'd have like the Indonesian finance ministry or some random site like that. And somewhere in the middle, in the, in a, in the bell curve, you'd have like amazon.co.uk. And I was like, well, that's weird that Amazon has like a low site quality score. And then, um, later sort of the obvious thought of, oh, I should check if there's a patent for this. And there is literally a patent called Site Quality Score, um, which describes, um, how they calculate it, which fits perfectly with the data that I see, which is essentially, it's like the number of queries that appear to be referencing your site, um, as sort of like a percentage and how much you are selected specifically in a SERP. So for instance, um, so branded searches basically. Yeah. So the reason why, like, there was really high quality scores for things like university libraries and things like that, because they had decent search volume, but basically the only thing people were searching for was that site by its name. And the reason sites like amazon.co.uk were quite low was because Amazon ranks 95% for random products that aren't necessarily Amazon. Possibly a reason why as well Reddit does particularly well, because I think a lot of the things people search for on Reddit, they include Reddit in the search term.

Yes.

But also if you appear for a query and say you're third and it's obvious people are clicking on you because they trust you. So I think this was just a really nice method for Google to say like, yeah, does this, does, do people want this specific brand? Does it in itself have value over all of the other similar content on the web? Because that's, that's like such a major problem for them, right? Which is when all other things are roughly equal, which one do we show? And there's a whole— I think there's a whole interesting conversation there around— I've heard, you know, people say, oh, Google favours big brands, things like this. It's just a reflection of what the user data is showing, because people prefer big brands. They click on big brands. We've had those studies for over a decade.

Yeah, they trust them more with Reddit. They get their answers faster.

Exactly. So I, you know, I don't land that one at Google's feet of, oh, they're just propping up other big companies. It's It's a very, you know, like it's a very grounded psychological trait of people, you know, the thing you're familiar with, the thing you trust that hasn't hurt you. So I'm going to go with that one.

So you mentioned PageRank and backlinks. I want to ask your favorite link building techniques since, since that's the thing that everyone needs to focus on. But also, I also want to ask, since we're talking about branded searches, at what point does getting more backlinks not matter as much and you actually need to increase your branded searches and you need to do better marketing so people are searching your brand?

So I think nowadays for me, the type of kind of link acquisition I like to do, and it's not a very, it's a pretty vanilla answer, I'm afraid. is digital PR.

That's a great answer.

Yeah.

I mean, actually, I had, I had the, the owner of Help a Reporter Out on the show yesterday. He's the founder of featured.com, and he— crazy new feature in featured.com that uses Help a Reporter Out's database of 33,000 journalists. They're doubling that. So it's going to be 66,000 journalists. And it's just an AI agent. And you say, I have this news about this. Find me journalists who cover this beat. Who would be like, and it gives you the exact journalist, it will write the pitch for you. And then you, you know, you should change it because it reads like AI, but you change the pitch, but it gives you like the most targeted journalists. And like, it sounds like you've done PR. The more targeted your pitch is, like it just increases the likelihood of being accepted so much. I think even though vanilla answer for me, that's like, A great answer.

I mean, the thing, so we're really lucky now that we've got a dedicated digital PR team in-house because I think a lot of people steer away from digital PR because it is very hard, I think, to do, to come up with stories that people actually care about, to get it done quickly enough so it's still relevant. And like you say, even when you start out trying to find journalists, you have to pay for databases and stuff, which is expensive. Um, but the results we've had from that have been phenomenal. And what I like about it from a strategic point of view is, you know, I'm not saying things like buying links doesn't work in terms of ranking, you know, it demonstrably does. Um, but I can get links that people can't buy. Um, and the other thing I like about it is if you do land a good story, um, the story, the links grow themselves. If you buy a link, you get one link that you've placed.

Media.

If you land a good story in the papers, every other paper will write about it and cite you as well. Um, I've done, like, one of the first really big link building campaigns I did was for Expedia over, I think it was like 13 years ago. We did a really big like April Fool's thing that they were launching flights to Mars. It was still getting links like 10 years later in like roundups, right? Um, so for me that, and it ticks all the other boxes as well, especially now like LLM AI era, right? Which is that You're getting these branded mentions everywhere. Um, the, if you're doing a good job, you know, with the PR, the PR side should be representing the company in the way, um, that's aligning with, you know, your mission, your values. So AI is kind of inferring the correct things about who you are, what you're associated with. Um, so that's all good. Again, I think that's missing if you like, you buy link inserts on like blogs, you're not getting. Those other signals. And again, it depends what type of SEO you're doing. As I said right at the beginning, like if you're doing your own affiliate stuff, um, you know, yeah, I'd be buying links for, I need to do this long-term with brands that want to be around in 10 years' time. This for me is the better play because those links as well, in my opinion, aren't ever going to be, as long as Google and search engines are using links, they're not going to be devalued because they are a representation of the thing the search engines are trying to actually measure.

So I think some people listening might be like, well, I'm not a— I'm not Expedia. So how is this relevant to me? And so I— this is one of the questions that I had for later on, but I think it's good to ask now. It's a fun question. If you were starting SEO today with zero reputation, what would— and like new, a new site, what would you do in the first 90 days and the first 180 days?

So if we're going from site launch, we're taking the kind of premise here that the site's technically healthy. Um, so I mean, I'd have my obviously content plan publishing schedule in place. Generally, if you're starting from scratch and we've literally done this with, with clients, so I can give you kind of an example as we move here, if you like. So we worked with a company in the UK, uh, called Rehabs UK, which is, uh, like private drug rehab. Which is super competitive in SEO because you can't buy ads in Google for those kind of terms. Um, private rehabs are like really, really expensive. So they're high-ticket items. So SERPs are really competitive. They came to us, they were getting like a few hundred visitors a month. And that was just some brand searches. They wanted to rank for like big terms like alcohol addiction, things like that. And the first thing we did when we looked at the strategy was basically like, You can't have that in the first year. It's just not, it's not going to happen because you are— this is a David and Goliath scenario, right? Because there's some big players there. So we made a content strategy first off that was targeting kind of the longer tail stuff that they did. So there's all other kinds of addictions that they can help with that you can get kind of cognitive behavioral therapy for that don't have the search volume, but they're getting Google to see your site about addiction, the kind of media stuff around there is a bit smaller. So we started building out that content. We still had the kind of main pages that ideally we would want to rank for one day, but we were focusing on building out this other kind of more niche content. What we were then doing was trying to find how we can leapfrog ourselves into the media. And it's not actually as difficult as you think. So the first campaign we actually did for them, so this was in the first 90 days was, um, there was a news story going around at the time about how the parents of children that played Fortnite were suing the creators because it was addictive. And we have a whole system like set up that monitors RSS feeds for things that are relevant to clients. So this had like addiction in it, so it got flagged. And I was like, this is perfect, we can do reactive PR from this. So we essentially Then just use Google Trends data to look at what are the other video games people are searching for, for addiction help with. And then compared those search volumes to like different kinds of real-world, if you like, addiction, substance abuse, and then bounce that back within 24 hours to every journalist that covered that story, which was big. And I think we ended up getting like 70-something links. We got links from like The Guardian, which is one of the biggest sites in the UK. It was on, it got syndicated on MSN and Yahoo and all these big things. That site went from, as I said, a few hundred visitors a month to 60,000 a month. It was outranking the NHS.

So hold on, the content wasn't changing. You got this crazy influx of links and then the organic traffic takes off.

Yeah. So, I mean, the organic traffic took off over like, I'd say a 7-month period. So we were continually working on that. We managed to get other placements as well. Um, so the other thing that was, that we managed to do that was fairly straightforward was we found every kind of county council in the country has listings for places where you can get like help with things like addiction. So we emailed all of them and we're like, This is UK-based, it does this, these qualified people. So we got a link on half or a few dozen different .gov sites, which certainly helped as well. But we were still kind of drip-feeding content onto the site. But the thing that made the needle move there was this influx of links suddenly from sites that we knew search engines really, really trusted. And it was evident because as I say, we ended up just over 12 months from the start, with about 60K a month traffic, mainly based around addiction terms. We had some terms like ketamine addiction where we were number 1, like above the NHS. So the National Health Service in the UK, um, that still fluctuates. We're still fighting them. But, um, you know, they should be first really, but it really showed that again, it wasn't much to do with the content quality, the missing ingredient there. was nobody knew who they were, nobody cared. And as far as search engines were sort of concerned, nobody had referenced them, so they weren't trustworthy. So by getting these links that you can't pay for, that really kickstarted them.

These stories, they just inspire me so much. It's like why I love doing this, because you think of some creative marketing campaign And it's not just that it brings you backlinks, it also brings you referral traffic. It brings you branded searches. It increases your top-of-mind awareness. It does everything. And you have the keywords in place, you know, so, so like when you are submitting your brand to journalists, you have a relevant blurb that, that is informed by keywords and everything plays together and everything grows together. Yeah.

I mean, overall, the reason that strategy worked was there were several things we were, you know, we'd done in the lead. Like when we inherited that site, it was small things like, okay, if I went here and I had an issue with alcohol addiction, there was like a get help button. There was a page about alcohol addiction. There was a page about rehab. And I was kind of like, well, hang on a minute, which one of these would I go to? So we you know, merged all of those pages into one place to make it very obvious. If you want help for this thing, this is where you go. Um, so there were things we were doing to make the site, I think from a technical user experience point of view, um, work well. But the thing that was the, the, for me, the, the real, uh, missing ingredient, and you're absolutely right. And I think there are things that Google looks at when it comes to how many links you're getting. As to what's reasonable. So I don't think they use this system as like quite how they used to, but there used to be a lot more conversations around like link velocity, the rate at which you gain links, right? And one thing that I always used to notice was if you landed some, what you'd define as big links, like in the newspaper, it would give you some bandwidth to get like a few hundred you know, fairly small links. What Google didn't seem to like was if you just suddenly acquired hundreds of tiny links. Because if you think about it logically, how would you be so popular? All of these disparate sites that aren't necessarily connected themselves suddenly find out about you, but you're not mentioned anywhere big. Whereas if you are mentioned on a big high-traffic profile website with a big brand, there is like a waterfall effect of then other smaller sites will hear about you. And it makes sense why you suddenly acquired 200 links in those 2 weeks. So I think there are other signals around, you mentioned things like you get more branded searches, things like that, that Google can look at over time. That really makes sense. There's actually a patent as well that I've written about before, which is their rank transition function. I don't know if you've seen that. Um, and it was specific, essentially specifically designed to try and catch people manipulating search results. So if you do some things that Google's like, oh, these look like very positive signals. Um, you're currently 10th. We're going to rank you 3rd. What they might do sometimes is actually rank you lower for a period first to see if you then, if the things that were good are suddenly undone. Because that's again, not a natural thing to happen. If I built a bunch of links naturally or through digital PR, I can't just go really to all these places and be like, can you please take them down? My rankings are worse. Like they're not going to do it. They're not going to care. But then if you leave it, the idea is it will then actually go up to where it's meant to be. And the patent talks about finding patterns of this behavior where it seems like you're poking at the metrics to try and work out what to do, which again, I think super interesting and maybe why sometimes it's possible to overthink what you're doing with SEO and play the longer game.

A lot of SEOs don't let the dust settle. They freak out at the most like minute changes in rankings. It can be very harmful. What signals do you think we're talking about signals? So, so what SEO signals do you think are massively underdiscussed?

I mean, I think it really comes down now to links and UX signals. UX signals is interesting. And I think it's an area of contention because—

Huge area of contention.

Yeah.

But I love it, by the way. I've like, I've made episodes on why UX is super important. So, but, but it is a big debate.

Yeah. So Google, because Google has told us, you know, flat out for many years, things like we do not use click-through rate in our ranking algorithm.

Right.

Which I believe is true to an extent, which is that there is a core ranking algorithm and there is a post-ranking adjustment, which is documented on Google. We had the fairly detailed link about the Twiddler framework, which are the unique stand— the standalone programs that run after ranking that readjust results based on how the query is classified. Now, I think to us as SEOs, it changes the rank. Therefore, it is ranking. I think to Google, those things are not site dependent. They're not ranking the site. They're re-ranking the SERP based on the user query. Therefore, it's not a ranking signal for a website. And I think it's a very important distinction because, um, the click-through rate isn't necessarily affecting or You know, it's not in that like core ranking algorithm where Google's working out, is this content, does it match the query? Is it good? Um, or, or, or more site-centric things. So there's definitely, um, I think a difference in terminology that's used between SEOs and Google. Um, because they very, very specifically word things. The other thing is like the, in our exploit, we saw there are absolutely Uh, site-wide scores, 100%. And people say, oh, Google lied to us about this. They said there was never domain-wide scores. And already there's a difference there because when I look back, the only references I can find is Google saying we don't use domain-wide scores. They never say site-wide because they're diff— they're completely different concepts. And that was the other thing that Like there's different site quality scores for every subdomain of a site. A help.domain.com has a different site quality score to www. And the concept of sites, I think, is far more useful from a ranking perspective than domains. So again, I think it's very possible that Google within their ranking algorithm do not have a domain-wide score. But they recognize sites which may well be by coincidence a whole domain. And that's what they assign a score for. But obviously we've got very upset about that.

If, um, I like, I really like focusing on the Pareto of SEO. Um, so the 20% of things that get 80% of the results when it comes to like a site-wide quality score, what do you think the 20% of things that move the needle 80% are?

Over a long enough time period, that's definitely the user signals. And this is again, so the examples that I gave in talks last year, was it the year before, was when sites were launching with loads of AI content, right? So they generate like 100,000 pages and generally you see the same trend, which is that rankings go and skyrocket up. And the— it, I think it comes back to site quality because my question when I was looking into the site quality stuff was, well, how does Google do this before they get the user signals? Right. So before the brand is known. So your question, the site's new. And funnily enough, there is a patent for that. Called Estimating or Predicting Site Quality. I can't remember which, estimating or predicting, which is all about how do you predict the quality of a site before you have user data. And essentially what they were talking about was they take all the different scores of sites that they have user data for and they kind of vectorize that content, make it a mathematical shape if you like. And then when you publish a new website, From a math point of view, they can be like, okay, your website looks like this shape and it looks closest to these sites, which are all around 0.6. So we're going to give you a preliminary 0.6 rating. I think what broke with Google, and because we had this whole renaissance of like niche blogging, right, was when we had all this AI content suddenly pour out, the AI content is literally modeled. Primarily on some of the best content on the web, because those sites are biggest, they're in Common Crawl the most. So the text generated suddenly went from, you know, instead of Markov chaining and, you know, mad-libbing, swapping words out that you could read, you know, they're spammy, they didn't make sense. Suddenly we could generate very nicely written things. So the initial scoring for Google got really screwed up. It was like, oh wow, yeah, this content is good. You're, you're like a preliminary 0.8, 0.9. And then we see all these rankings go, you know, skyrocket as, as these, um, as these sites publish all of this content. But then what happens is the kind of slow-moving train of user signals of there's no real brand searches happening for this because people find it and it's very meh. Um, a lot of the content's rushed. It's just not actually that good. People bounce off it. And Google's always talked about, you know, there's this rolling average, if you like, of user data that they've got. And I think that's what catches up with it. Because it's normally, they normally last between 6 and 12 months. And then I think they suddenly hit that inflection point where the score's suddenly too low. And Google's like, nah, people aren't digging it. I'm sorry, you know, we got it wrong. And then, you know, you never see the end of those case studies, do you, where they just kind of go like this.

So I think I share a lot of them.

Yeah. But I think for me, that the more experienced SEOs that I meet, and again, talking from an agency perspective, the longer term they think in, in my experience, because they've seen these things play out time and time and time and time and time again. So for me, the most important underrated thing about SEO is that you are, you have got a strategy. that firstly is Google-friendly. And by that, I mean, you are not getting in the way of Google making money, because they will steamroll you and they will not look in the rearview mirror if you get in their way. And that you've got something that is going to light up these user engagement scores, because it will kill you in the long run if you don't do that. And you get clients come because they've seen this great video. And this is the the other side of the sword, right, of being able to learn SEO everywhere is there's lots of TikTok videos saying, oh, you can do this brilliant hack. And yeah, some of the stuff works. And, you know, as Lily Rae says, until it doesn't, which I think is very true. So I don't think that gets talked about enough because it's not popular. I had a couple of people I was speaking to last year who were very big on like programmatic content, because again, that had a little revival, right? that's been going on over a decade, you know, it's just now got a new name. And I knew exactly what was going to happen with that. And we've seen it now, especially with the last update, right around all the best-of type content and the competitor versus content. A lot of it's tanked.

Yeah.

Um, because it was made because it was very attractive, because it looked like it would work. But I just think that's where if you focused all of that time and money Into a harder thing that's slower, but will last, you're going to, it really is like a turtle and hare race for me.

I think SEO is extremely easy if you, if you have a long-term focus. It's, uh, and it's, it's so, it's so fun and gratifying too, because as, as your domain authority grows, as your authority grows, you can also get away with more stuff too.

Absolutely.

And you, you can target keywords that you never thought in a million years that you would be able to rank well for.

Yes, absolutely.

It's incredible.

CBD gummies. Yeah, that's gonna—

here we go.

Yeah.

So programmatic SEO, do you have a difference between programmatic SEO that is Google-friendly, that is good, that could stick around, that's like, hey, this isn't bad at all, versus like what distinguishes programmatic SEO as Pretty bad.

So, I mean, it used to be done that we would just have kind of template type pages and we would scrape database. We would make our own database off the internet by scraping a load of information and then just kind of fill in the blanks and do programmatic that way. Um, some of it's lasted way longer than I ever hoped it would, especially around like local pages. Like people have been getting away with that for just years and years now, and I think it's Fairly safe doorway pages. Yeah.

Yeah.

Um, so it comes down to the, and I kind of sigh saying this because it's almost Google's definition of if you are making the pages for search or because there is a genuine reason to, to do that. Um, the way I think I would go about it. Is again, using your own data rather than relying on stuff that exists elsewhere already. Um, when, so Gary Iesh last beginning of last year was in London at an SEO FOMO meetup and super nice guy, very smart, really interesting to speak to because he does not mince his words at all, positive or negative about the work he's involved in and seeing. And one of the things he said on stage really interested me, which was people asking him around what Google is going to be focused on over the next few years. And his answer was basically originality. And I remember that. It seems like a very obvious answer to me because it's so easy now to just generate, you know, I can generate local pages, I can generate this, that if I've got like a good example, I say is, um, like how Amazon are, um, summarizing all of their reviews. I think it's a really great use case for generative AI, right? Which is, okay, a product has 3,000 reviews. I'm not going to sit there and read 3,000 reviews. It tells me as a user, you know, that's some information that so many people have bought it, but having Amazon say, people particularly like this feature, there is some people upset with the battery life or something, is just amazingly helpful. So I think there's ways that you can extract additional value from the things you have programmatically. I'm not anti-AI whatsoever. I've always been a really big proponent of AI. I think generative AI is now overhyped and used for too many things. It's not meant, it's not good at at doing. So it would, yeah, it comes down to, are we actually achieving something with the user? Because the thing, the search engine, whether it's a platform like ChatGPT, if I can call it a search engine for now, or Google, what they're trying to work out how to measure is how helpful is this to the user? So if you're focused on that, even if they're currently getting it wrong, you will get closer and closer to alignment. I think too much of the problem is focusing on what is the robot measuring, and you're the human, right? So you already know the answers, and you can already think in a way that the robot can't and is desperately— they've got really smart people desperately trying to make a robot think like you. So I almost feel sometimes we're diminishing our own knowledge to try and create something for a robot when actually they're trying to figure us out. And there is, and again, I don't want to sound naive here because there is absolutely a middle ground between these things. I'm not saying, you know, if a specific technique works, you shouldn't lean into it. Um, you know, and that's the reason why even we've got technical tools like Canonicals, like hreflangs, because there is a gap between what they want the machine to be able to achieve. and things that are very obvious to us as humans. So there's always play there. And it's a very grey area that is situation by situation on what should you lean into and what, what's going to cause you to fall over if something changes. So again, it's about risk mitigation, understanding what the search engine is trying to achieve, figuring out if you leaning into that is kind of getting in the way. What's the cost if all of this did fall down? Have you got a plan B? Or have you built 80% of your business on sand, essentially, that Google's going to come and kick over one day.

Yeah, I wanted to ask about your other, like, other marketing channels that you like. But since we're talking about AI, it's just like a common question, comment that I get, which is like, you know, how do you— how do I use AI to make my job as an SEO easier? And what do you say when you get asked that?

But if we're talking, so again, I've used AI. I did my first qualification in machine learning like 9 years ago now. And I've used AI in some aspect within SEO pretty much all that time. I know now when people say AI, we're generally talking about generative AI, LLM type stuff.

Yeah.

So if we talk about that specifically, my quick answer is I never use it for something that I need to be sure is correct, um, because it just can't guarantee that. So, you know, anything that needs to be correct, if it's content, if it's code, like, I'm not gonna put it on the front line without—

Does content need to be correct?

Let's, let's come back to that. Let's come back to that in one sec. So the way I currently use it is essentially to speed up a whole bunch of things I'm doing so I can reel off some examples here for you. So things like site migrations, we've got AI models that we do. So we vectorize all the pages. We can, can use an AI model to then do vector comparison for how similar are these pages when I'm doing 301 mapping. So I no longer need to try and do it on page titles. I can take the whole content or just the title or various elements, mix it together, and then say, how similar are these pages? So it doesn't matter if the title's changed, doesn't matter if the URL's changed. Um, you get a brilliant redirect map now in minutes. That would take hours previously. Um, it's generally better than a human. Um, the thing I mentioned earlier with digital PR, so we have, uh, an instance of Ollama running locally and we have built, um, just a very simple Python script that accesses a Google Sheet which lists about 100 different news RSS feeds. There's another Google Sheet which lists our clients and a description of the type of news they're interested in. RSS feeds gets pulled. Every single news story goes to the LLM with the description of what the client's interested in. And then there's like a true or a couple of true or false. Is this relevant to this? Is this not relevant to this? If it's a relevant thing we could do reactive PR for, the script then pings it into our Slack. So we're monitoring like thousands of news stories every day across multiple clients.

How do you monitor that many?

Sorry?

How do you monitor? How many stories are going into the Slack? How do you see all of those stories?

So the only stories that go into the Slack are the ones the LLM has determined are relevant to that client.

It's not that many.

Which is a small number. Yeah, it's a small percentage.

I thought you were saying thousands were going into the Slack. I'm like, what?

No, no, no, no, no, no, no. So this is the idea, right? So it's such a perfect job for an LLM to do because Classifying content. Yeah. Traditionally writing a classical programming rule to be like, is this content about this? It's very challenging. It's like, do you list keywords? There's all kinds of matching possibilities, false positives. LLM is so good at it. This is really cool.

That's a really cool idea to do reactive PR. So you have all these RSS feeds and you have a script and anyone can probably just use ChatGPT or Claude to write that script. And then anything that is relevant goes into, you get notified by email or into Slack or whatever it is. And then you go, oh my gosh, like we could, like, this is trending, this is hot. We can relate it to our business and get a lot of links that way. Yeah.

And being first is so important with things like that. So that's one way. Another thing we're doing is, so a lot of the content now that we do, so historically, We would do things like we would write a content brief. We might send that to one of the specialists at the client because we want a true expert to write it. They're not maybe a great writer, so it's like getting blood out of a stone and that comes back and then we edit it and they've just answered the questions we've given them verbatim. Now we can have a conversation like we're doing here, just record it on something like Fathom. Then we've got the transcript. We're still running off a brief to talk to them, but—

Yeah.

Because most people are very, a lot more comfortable just talking than they are trying to write an article. They tend to go off on tangents. They tell us other things that we never would have even thought about asking that we, because our brief is limited to the keyword research we've done, whereas they have subject expertise. We can get that back immediately, get AI to create if we want an article draft from that. And then that goes to an editor and, you know, that can be manually then edited. Massively sped up what we do with content. And then we've got kind of like you have, right? We've got video, we've got audio we can use, we've got text, we've got clips we can use, we can disseminate. So that's 3 areas we use AI for tech content for digital PR. There's lots of other ways, but all those ways are assisting what we do. We're not reliant. It doesn't matter if it gets something slightly wrong because it will get spotted. If it flags the wrong news story, we can see that all the 301s are checked anyway. You know, transcript errors are easy, fairly easy to spot.

And so like when you're publishing, when you're publishing content and it's AI written, how much are you editing that for the brand voice, for accuracy? How important do you believe accuracy to be? Like, are you, are you removing the em dashes? Are you making it obviously not look like it's AI?

So firstly, I think in that case where you are, so from a technical viewpoint, if you have done an interview and you transcribe it using AI, I don't think that matters. Google has said they don't care how the content is produced. The difficulty comes is if the content is made from the language graph, i.e., we want something about you know, different flooring substrates or something, you know, and I'm just asking ChatGPT.

Yeah, right.

And then it writes it. That's a problem.

Whereas if I speak to someone—

Exactly. Um, they are heavily edited anyway, um, because part of the editing process is this is the tone of voice of this client, you know, this is what they want to say. It will get signed off, um, because the cost— I feel the cost of getting something wrong, like we work, for instance, with people that make like industrial windows and they sell them to like architects and stuff. They're very quickly going to see if some important terminology or detail is wrong. The same as if you write an article about SEO, you know, from ChatGPT pretending to be an expert, and then I read it and I can see it's done some terrible error. It's probably very obvious to me as an SEO, but maybe not someone reviewing it. So no, everything does go through that because the cost of being wrong I feel from a brand perspective is, is, is high for those things. You think it's still way faster than writing it.

Do you think searchers can spot AI writing or is it really dependent on niche?

Probably dependent on the person, I guess. Person and niche. That's what—

So do you think when they— sorry, do you think when they spot the AI writing, do you think that causes them to pogo stick and go back to the search results?

So my, I don't think there's enough data out yet to answer. Can people in general spot AI writing? I've seen some, I've seen some studies talk about that where people couldn't tell the difference. But when they were told, they, they disliked it. When they were told it's AI, they disliked it. I think as marketers working very closely to this, you know, it's almost a meme on LinkedIn, right? You know, em dashes is the big one and delve deeper. And there are lots of common sayings now where, you know, I get the AI ick where I'm reading something and I'm like, yeah, now this is AI. And for me, it immediately switches me off. And from the kind of surveys I've read, I think that is fairly prevalent in terms of if someone believes they are reading something written by a human, and it turns out it's an AI, they are going to be very disappointed in that and have a negative reaction. I think if you tell me it's AI, there are specific instances where I will like and accept that, which the example I gave you earlier was, if you tell me it's an AI summarizing all the Amazon reviews, perfect. That's exactly what I want to be AI. I don't need a human to write that. If I'm reading an article, if I'm reading something that's meant to be expertise, then I do not want to read AI because I know it doesn't have those attributes. However, that's not publicly— I don't think that's widely understood publicly because AI is artificial intelligence. And so people infer a lot when it comes to intelligence. And when you type something into a lot of these LLMs, LLMs, it says something like thinking or reasoning, which of course they're not doing that in the way that a human does, but it's very much in their advantage to get people to think that it is thinking that way. Um, you know, and I've seen people from outside the industry use Google AIOs and just accept, you know, whatever it is that's there. It's just, it's just that. I've had conversations even with my, you know, my parents where they've said something outlandish to me and I'm like, is that right? And it's like, yeah, no, it says it right here on Google. And I'm like, is it? one of those AIOs. Okay, click the website. Oh no, that's completely wrong. Yeah.

It's dangerous.

Yeah. So I think there's a knowledge thing there. I think we probably care more than most people. So by I mean, we people working with that kind of content. So we're a lot more alert to something being AI. I think if you pick random people off the street and show them AI, non-AI, at the moment at least, I don't think they're going to notice a difference. Probably right.

Go on.

I was gonna say you develop ad blindness as well, don't you? I feel like I'm developing AI blindness in a way.

Really? Oh, I don't think I am at all. I think it depends on the niche. Is this niche likely to work a lot with like ChatGPT? And if the answer is yes, then they're more easily able to spot AI writing.

AI blindness. I mean, sorry, when I When I scan read something and I start to think it's AI, I immediately skip over it now most of the time.

Ah, yeah, yeah, yeah. Oh no, I'm the same.

So like how when I see ads, I don't particularly read them, you know, like display ads to me, like I couldn't tell you a display ad I've seen.

It's harder when it's like a small amount. It's harder to identify when it's a small amount. But no, I mean, I'm the same. If I get to a page and even if that page is about exactly what I'm looking for, But I feel that it's written by AI. I don't trust it because I've seen too many times personally AI making grievous— generative AI making grievous mistakes. So I want to, I want to ask, do you think it's accurate to say that you've done SEO for basically like dozens of niches? Yes.

Yeah.

Okay. So this is a fun question because I'm working on like a new thing and something that we're doing is we're We are deliberately going into a niche. We're picking and we're starting with a niche that has bad SEO so we can just like conquer the entire thing. And so I want to ask you, what niches have the most competitive SEO and the least competitive SEO?

I think the obvious answer for most competitive is where the kind of black hat stuff is rife. So it's, it's like iGaming, things like that. Drugs rehab was fairly competitive, to be honest. Um, but it's always just going to be where, where the money is, right? Um, so again, like, big ones used to be like payday loans, that kind of stuff. Um, and you see some— it's a great place to learn as well because that's where some of the kind of extreme, you know, the, the, the gambling SERPs at the minute are wild, and there's so much interesting, uh, stuff going on there that I know you've had some sort of quite well-known black hats on here, but I see some very interesting stuff happening with, you know, domains and site-wide canonicals after domains have been hit and fresh domains suddenly ranking number 1. And an example in my talk last year of a school domain that got purchased and they put a gambling affiliate page on the homepage, snipped all the other links off, but kept the rest of the site live. And it ranked for 16,000 terms in the UK, gambling terms. It was like in top 3 for— The school site did?

Yeah.

Yeah, it literally had a school title in the URL, but they just changed the homepage out for a long one gambling affiliate page and it absolutely dominated.

So it was just the homepage that was ranking for 16,000 terms?

Yes.

Wow. Yes. So they left on all the other school content on like all the, they didn't take down that content.

Well, no, I think why is because, um, if you buy a domain and then obviously flip it completely in terms of topic, right. I think Google is going to have like a bit of a reset there. Whereas I think it was about a 500-page website. So it basically meant 95% of the website was still a school.

Same.

So it kind of looked, as Google's categorization goes, the categorization hasn't changed. So it's kind of kept that authority. Um, you know, same example as CBD gummies for the Vatican. It, and there could be some reasons why a school would have a page about gambling, you know, like, hey kids, don't gamble kind of thing. That obviously they don't—

Not as likely, but okay.

But, but it really worked, right? Yeah, yeah, absolutely dominated. So I always, I find it fascinating looking, um, in those areas. The easiest one, I mean, that's a really good question, right? So I always find, um, a lot of older, old-school type businesses, especially, um, ones that are local just have the most awful SEO. Like, um, I've worked over the years with a few different types of solicitors and lawyers. I think that's changing now, especially in the States, but in the UK, um, you know, for many years it was just the worst, worst websites, um, because they, you know, it was a very paper-based job, I think. And, and it was how they got business historically is kind of through referrals. So they weren't that interested in the web. Um, any new, any kind of new areas products as well. I think there's always that time delay on who's, who's getting in there first. Like I had a good friend of mine, um, make a lot of money selling an e-scooter website when they first became a thing. Um, I unfortunately bet on the wrong horse. I made some of the first when, um, when before vaping we had, uh, e-cigarettes. So I had a really nice e-cigarette domain because I was like, I reckon these will stop, you know, people smoking, but the terminology didn't stick. It, you know, became vaping. Yeah. Um, yeah, so I'd say probably some kind of local services like gardening, like, and in the UK you can't get them for love nor money at the moment because they're so busy.

What do you mean?

So it's really like I tried to, I tried to get someone to do something to—

Oh, you can't get them.

Our garden, and like nobody responds because they're all so busy. But then I'm like, how are they getting all this business? Because the sites are just awful, but there's quite a lot of money in it, right? Because they're expensive things to do.

Yeah, I was thinking, I was thinking those like services, like, because what we're going to vibe code something, basically what we're going to do is we start with, we start with a niche where people don't want to vibe code their own solutions. And then we find a crossover for a niche that doesn't have great SEO. And then we take the top players in that niche and we look at the core services and then we vibe code those core services into one product. Won't take long. Maybe we just start with one core service. We do bottom of funnel SEO for that core service. People start, people come to the product through SEO and then we actually talk to the users and we say, why are you using this? Because like we might go into a niche where we know zero about the niche beforehand. We have zero experience with it. So we talk to them like, what are you hoping to achieve by using this? Like, why are you doing this? What do you wish we had? And then we go and build that. We vibe code it. It's very easy.

Nice.

And yeah, and then we, now we have more features to do SEO for. And then we grow and then we reinvest into other marketing channels and that. Yeah. So, so that's, that's why I wanted your opinion on that. And we're looking at like a lot of right now. We're also, we're looking at makeup as, as a niche. Oh, I'll tell you what's crazy. Augmented reality. We were looking at augmented reality yesterday. So much money in augmented reality and the SEO that we were seeing in it is not very good because, you know, because up until now, in order to do like augmented reality software, There was such a crazy barrier to entry. And so there just wasn't that many companies. And now you can vibe code a lot of solutions using AR. And I think that's a really amazing niche to be in with SEO because the SEO that we were seeing was not very good. Yeah.

It's, uh, so the makeup niche thing is interesting.

Oh yeah. We're looking, That we're looking at that actively right now too. We might even— yeah, go on.

I was just going to say, so in terms of the makeup niche stuff I did, it was a couple of years ago now. I did, you know, used to get those competitions right on Twitter. There'd be like, like and retweet to enter. So I saw these and I was like, well, that's something you can quite easily automate, right? So I wrote a script that looked for tweets that were saying like and retweet for enter, and it categorized if it needs a retweet, if it liked. And I left it running for like 3 months and I entered something like 50,000 different competitions. And it turns out the win rate, if you enter random competitions on Twitter, was about 1 in 3,000 at the time. So, I kept this sheet of all of these prizes that I won. But one thing I noticed was probably of all the competitions out there, like half were makeup giveaways. It was like, I'd never realized how big—

So, you got a lot of makeup.

I got a lot of makeup and I hadn't actually told my wife at the time.

No.

I kept getting these bits of makeup and I was like, oh, look, I got you this lipstick and I got this eyeliner. And then I probably shouldn't share this. So, one of the things that I won, so I won everything from like straws to like a Call of Duty Xbox game, like books, like all these random things. And most of it I gave to charity. Um, and I had like a fake name on the Twitter account. And, uh, one of the things I won, and when I had to tell my wife that this is where I was acquiring all this random makeup from, was we had a parcel arrive and it was addressed to like someone else. And my wife texted me and be like, oh, what's this for? Because I was at work. And I was like, oh, you can open it. And it was actually some crotchless leather pants. And she was like, why have you ordered like this? What is happening? Oh.

I was like, okay.

I need to own up about this, but yeah, I won that in a random Twitter giveaway.

I saw that you also won $13,300 from Google for discovering a bug, and then you gave that to charity as well.

Yeah. Yeah. Well, I figure if you can't give money you're not expecting to get to charity, what money can you give?

What was the bug?

So this was the Google exploit that we found, which was Basically, whenever you're doing a Google search, Google's a bunch of— has a bunch of microservices, if you want to call them that, that talk to each other. And because we are doing a lot of work with Also Asked and scraping bits, we're looking at all that stuff very closely. And it was literally in the kind of network tab. This is— you could tell how I spend my evenings now, looking through kind of where the data was being sent to and from. And there were some interesting sort of requests, and I took one of them out and was running it locally just through curl. And I got back like some partial HTML when I did it, and I was— there was a lot of parameters on there which were determining kind of the— what I thought the format of the response would be. Um, and I've just— out of curiosity, I was playing around with that and I deleted the format parameter, and it basically gave me back then structured data So for every query I was doing, it was doing things like it was giving me estimated click-through rates for all the different titles. It was giving me site quality scores. Um, it was giving me like, um, TDIF scores for, um, how close things were matching. It was giving me, uh, categories for search queries. So it was telling me this is an informational search. This is a comparison search. It was giving me commercial scores for queries, medical classifiers. So all the stuff that Google would need to pass to another system to make a decision. So I think that was a lot of the data that's used post-ranking. So metrics that are generated during ranking are kind of pushed forward. So when the post- ranking, re-ranking happens, they can say, well, we scored this site this, okay. Because they don't know the query until it happens, right? So you can only act on the re-ranking once you have that. And some of the stuff that you previously calculated is important. So it was that, um, we, you know, I, we sat on that for quite a while, just kind of looking at it.

Um, could we share this? This is so valuable.

Yeah. And then eventually I was like, okay, I think I've learned everything. We're going to learn on this.

What did you learn? What did you learn?

Um, well, a lot. So this is what I was saying in terms of, it was, and this is why as well, I didn't, I didn't release this, like all this, I didn't dump all this stuff publicly. One for, um, just fear that, um, I'd hear from Google's legal team. Um, cause they have a legal team that's much larger than any business I own, right? And I don't particularly want to get in a protracted legal battle with Alphabet. But also, I don't think there was anything in there that was really like super exploitable, like, oh, you can just— this is one metric. Actually, a lot of the stuff we saw, in my opinion, backed up what Googlers had been telling us for a long time. So things like the click-through rate were very similar to what you see in Google Ads in that I think they predict it based on the query and the kind of mathematical shape of the content. And they're like, I think from previous data, our model says this, users are most likely to click on this, which makes sense. There are of course post-ranking things that will adjust ranking, which is why click-through rate manipulation works while it's happening. And then when you stop it, it kind of goes back because it's a post-ranking thing, not, not a ranking thing. Um, so I learned a lot about, I think, how Google actually functions, um, in terms of those different stages of ranking and how they link into each other. Um, the reason, the other reason I didn't kind of publish everything at the time was at that time, the content warehouse leak came out. And to be honest, I don't feel as an industry as a whole, we reacted very well or fairly to that. Because I think quite a few Googlers got a lot of direct abuse from that, of people just calling them liars because, oh, there's this random variable name included in this thing, which You know, you don't even know if that's used in ranking and re-ranking, in indexing, in storage. Like, you don't know where that's used. Like, in the exploit that we found, there was a thing called, uh, social signals. And I was like, wow, like, that sounds like Google uses social signals for ranking. And then when I, I actually dug into that and was looking at where it was being used, because I said we had the sites, all it was was, um, basically figuring out whether to display like the little social icons, um, as a feature.

That's fine.

Nothing to do with how sites rank, right? But I was like, oh, if I dump this information out there, it's going to create, I think, 10 times more misinformation than it will helpful information, um, because people are going to just take that and run with it, right? There's going to be TikToks about, oh, Google uses social signals. And I was like, I don't want to be at the center of all of that. So I will, I think, and that's the year I started doing more conference talks because I don't really, you know, I very rarely actually speak at conferences. It was mainly last year. Um, apart from that, I've done like half a dozen in 15 or so years because I was like, I will package up this information with the lens of the experience I have, and I will genuinely tell you the things that I've learned that I think are helpful. And like, they're all on YouTube now of This is what I discovered. This is how I think it fits into Google. There's a patent that backs this up. There's also some Googler telling us this is roughly how this works. So all of these things align, you know, the evidence that we've got, the firsthand account, what we actually see happening, all of these things align. So this makes sense. Um, and also this is where I really got into the whole kind of, okay, this is what Google appears to be classing as ranking and post-ranking, and you need to listen very carefully.

Yeah.

to the precise language they're using when they're speaking to you and what they're telling you and what they're not telling you. Because I don't think they're going out there and just intentionally like lying. There's a game being played, right? They're not going to tell you how to game the algorithm. But they're not just out there, in my opinion, spreading misinformation. There's no reason for them to do that. Gary will just tell you no if he doesn't want to tell you something. But I mean, he's given me really useful bits of information as well. Like, There was a lot of SEOs at one point talking about updating the last mod attribute of documents to make them look fresher. And Gary was sort of saying, oh, that's only meant to be used when there's significant changes to the document. And he was like, if you keep changing it and we see you're not changing document much, we'll stop trusting you. And so that led me to ask him, well, How does that scoring work with Google? Do you have like a score for how much you trust the signals? And he came back and was basically like, no, it's a binary thing. So basically, if we, you know, if we think you're not being accurate with this, we just stop using it as a signal from your site. And it makes so much sense.

How do they determine, how do they determine if there's no updates happening?

Well, they're fetching the content, I guess, aren't they? So if they can tell how much the content's changed, And if you're saying the last mod is changing every day, but the content looks the same.

They have several cached versions, right?

Yeah, yeah, absolutely.

Yeah.

Yeah. Um, so, you know, that, and for me, that's all really helpful information because there's all kinds of signals that we can give Google, like canonicals is another one. You know, it makes a lot of sense to me that if a lot of your canonicals are bad, that Google is just going to be like, I'm just going to make my own mind up. And a lot of these things—

So you mean you have the incorrect canonicals?

Yeah. If you're canonicalizing to loads of 404s or noindex pages, I think a lot of these systems are simpler than we might imagine because I think the complexity when you start moving away from binary decision tree stuff, starts to just explode. And it was something, you know, again, Google's been telling us that for a long time. One of my favorite talks was Paul Ha back in 2016 at SMX, and he did a talk called like Google from an Engineer's Perspective, and he was talking about how they use user signals. And he was like, look, um, user signals are actually really noisy, and despite what you think, it's actually very hard to Um, generate scalable insights from them. Um, which is why when people talk about user signals, I think we're talking mainly about things like interactions with the SERP, because that's— Google knows exactly what SERP layout they had. There's parity there. They can measure all of those things. They can measure hover clicks, you know, all that stuff. Some people think that Google's measuring more user interaction kind of on the site, which I just think is so incredibly difficult to do that because you need a way that scales and you can compare all sites. You know, there might be a couple of metrics they can use, but I just don't think they're doing that kind of in-depth stuff because that trying to then put that into an algorithm is, I think, you know, for the reasons Paul was explaining, is incredibly difficult to do.

You had to guess some UX signals within a site that Google might use. Just guesses. What would you guess?

Oh, I hate wildly speculating.

It's so much fun.

It's going to just get clipped.

No, no, no, no, no, it's not going to.

It's not.

Yeah, I, I won't take, I won't take anything out of context.

So honestly, I don't think there's outside of the actual like site performance stuff, which again isn't even that big a ranking signal. I think that's more of a proxy of People like that, so you get better signals. When the conversations came up about things like time on site, my argument was always, if you had the perfect content that answered the user's query and you'd highlighted it well and they landed on the site and they're like, cool, that answers my question, and then they left, that's like 100% bounce with 10 seconds on site. But that's probably the best user experience they could have had, right? Whereas if they're coming in for the Guardian political long read, you know, you want scroll actions, you want hovers, those kind of things. So I just don't think there's a generic— I don't think there's a good answer to that question because I don't think it's something they do use. That's what's so special with, for me, about the Core Web Vitals, because without any context you can say this number higher or lower is better or worse. And there's actually very few metrics, like bounce rate is a great example, where you can definitely say higher or lower is better. Because there are cases where having a high bounce rate indicates better user experience, and there is cases where low rates indicate better user experience. But there's no user experience that's better with a, you know, a site taking longer to load or, you know, being jumpy around as it loads, right? So they're, they're interesting metrics for me.

Is Core Web Vitals a ranking signal?

Like, I mean, again, Gary was pretty straight on this. He said, he said, uh, it was like an on-the-fence tiebreaker type thing. And I think it's always been that way. I think the more important signal from Core Web Vitals is if it's annoying users. If you land on the site and it takes 8 seconds to load, I don't think Google minds that much as long as users stay. and do what they want to do. If it's taking 8 seconds to load and everyone's like, nah, screw this, and they're off, that's what I think damages you. And there's a bunch of stuff in SEO where I'm like, this isn't an SEO problem. It just so happens to be within the halo of things that might impact SEO. If your site is slow, you should fix it because slow sites suck and people hate them. Don't even think about SEO.

You'll make less money, you'll get less conversions.

Exactly. And there's tons of data that shows how it negatively impacts like conversion rate and revenue and Uh, stuff like that. So, and there's a bunch of things, um, that I think SEOs try and hoover up as like SEO things when they're more, no, this is a separate, um, area, but it impacts that channel. Um, you know, security is another one that I tend to stay away from, although I'm interested in security. You know, there's a whole bunch of, um, auditing tools that try and look at security stuff. And I genuinely think SEOs, you know, you need to flag that, but you need to probably stay in your lane unless you are like a cybersecurity person.

Yeah. Yeah. Um, you, so you've done like tons of SEO experiments. What are some like, aside from the schema one, which we talked about, what are some other ones that you've done recently that, uh, where you're, you've done just any SEO experiments, honestly, because pretty much every one that you do is interesting. And then like, are there any upcoming SEO experiments that you've that you've been thinking about?

So yes, I've got a few that I'm actually going to be doing kind of for my talks this year. So I'm at several conferences this year and the kind of premise of my talks is going to be around SEO decision-making. So kind of what we talked about with the schema, and I've set up a few different tests to demonstrate that, okay, maybe this obvious truth isn't quite so obvious. I do publish most of the things that have any kind of meaning. I mean, I've got my own personal blog as well at i83.uk, which is for kind of the more unhinged tests. So I did one, um, where Cindy Crumb did a really brilliant talk, um, around a year ago. We're looking at Chrome telemetry data and the hypothesis was that Chrome was essentially, or Google was using Chrome renders from users and basically nicking that render for their index to save a lot of their, um, kind of computational power. So I set up an experiment there that was doing kind of different couple of levels of cloaking of showing one thing to a user, showing one thing to Googlebot. For users, it was also generating like a unique hash that Google should never see. And then I got a few thousand people to visit all of those URLs in Chrome over a few weeks. And the idea was to keep monitoring the index, um, because we could use the site, um, thing at the time to see, and the cache still existed, to see ever if that, um, any of those hashes got into the index. Um, and they never did. And we test— I tested that for a year solid in the, you know, hope of that would be massive if, if we caught Google doing that. I didn't think they were doing it because I can just see again, so many privacy issues of if you're grabbing accidentally like a logged in area or like, it's just, you know, I think there's a massive lawsuit waiting to happen if they were doing that. Um, well, I didn't shout about that one because nobody's particularly interested in tests that don't show something interesting. But again, it's about, it looked like Google could be doing that because there was so much telemetry in Chrome and it made sense from a, you know, there's like a potentially a financial motive to do it, performance motive, um, ignoring the massive legal issues. Um, and again, that was like a thing that I think this is testable if we design an experiment. Um, and just if that's not the case, then you can rule that out of your thinking. Um, so I've done stuff like that before. Um, I've tested, uh, so there's a chap called Oliver who runs, uh, OGHM. Um, he's very quiet technical SEO. He's just discovered loads of little very neat niche things. Like, um, one of the favorite things he discovered I saw was it's way quicker to get your pages indexed through multiple sitemaps than a single one. So he was showing like if you submit 5,000 URLs in a sitemap, it takes ages for them to get indexed. And he was like, whereas if I put a page on its own in a sitemap, a separate one, it's indexed almost straight away. And I was like, it's all these weird little things of, so actually it would be, um, you know, it'd be more. So one of the things we actually do now with clients is if they've got like you know, pages that update quickly or are newly published, we have a separate kind of new sitemap for them to go into. Um, cause it just gets them indexed quicker. Um, and again, from maybe a more black hat point of view, you could technically have a whole bunch of sitemaps with only a few pages in. Um, so I always find those things really interesting, not because they always give you a specific like tactic you can use, but it gives you insight into how things are actually working. You can demonstrate, say, oh, that's interesting, that works like this, and then apply that thinking to what you're doing more generally.

Yeah. I like the experiments that gives tactics, that just gives you a best practice. I don't know. Are there any of those experiments that you're doing or other ones that you've done recently that has—

They're mostly about geo stuff.

Okay.

I'm doing now.

Yeah.

But I don't want to talk about that because I want to, I want to have them ready to go for these conferences. But I'd say the last, the last one I did that was like, I think truly actionable was probably the meta description stuff, which is, you know, that to me was, you know, we worked, we got data as well from the SEO testing guys there. Ryan at SEO Testing shared from— they have the kind of metadata of tests people are running and the meta description tests basically did nothing. He found when people were A/B testing meta descriptions that had no impact. And I was just like, the amount of time. And the other thing was as well, then we immediately got people saying, oh, well, the LLMs use meta descriptions, so you have to do them. It's actually part of geo now. And I'm like, nobody bothered testing it. So again, blank meta description, ask ChatGPT, and it just gets the meta description in that case that Google had generated because it wasn't reading it itself. It was getting it from the search engine. So I was like, nope, doesn't matter.

Funny. Do you think your method of SEO differs from like standard best practices in any ways?

That's a really good question.

Mine does.

Sorry?

I said mine does.

Yeah, I think it's more what we don't do in some cases. I think It's very easy to waste time in SEO with various checkbox things that aren't that important. That's sometimes hard to communicate because you get that with experience. Um, I've never been a fan of kind of, you know, the gold, silver, bronze type approach to SEO because, you know, you can get a site and when even we're at the proposal stage, I'll go through that site and essentially be like, In some cases, it's barely worth doing a technical audit because I can see things are performing well. I can see from Search Console, things are in pretty good shape. And again, when it comes to— I feel there's a lot of things like content creation which are just done for the sake of it. So a lot of our approach, because of the team we've got, We, while we've got kind of SOPs for some of the tasks we do, it's more the senior SEO gets it. And I'm like, tell me what you think we need to do on this site to, you know, hit this target or whatever. So yeah, I wouldn't say on a strategic level. And again, it depends who you're kind of comparing us to. We're doing anything wildly different. But I'm always happy, you know, when we when we're doing pitches and stuff, especially when you have discussions around things like, oh, someone says they'll do this many days for this much money. I'm normally quite happy to be like, cool, well, I'm quite happy to do half that amount of days and I think we'll absolutely flatten them with what we're going to do.

You mentioned you have like an also ask tool.

Yep.

What is that?

So it's alsoasked.com if people want to check it out. And essentially what it does is you can put in any kind of question query and it mines Google's People Also Ask queries.

Ooh.

So it will go a couple of levels down and you can specify your language and location. So it will actually set those for you. And it's interesting because especially in places like the US, You will get very different People Also Ask results depending on the query.

Oh my gosh, dude, I might make a TikTok about this. This is great.

Yeah. So, um, it's a little bit slow at the moment because Google kindly—

What's the URL? Also Asked? Is it—

alsoasked.com. Also Asked.

I'm getting, I'm getting, I got redirected to vpntoggle.com. Maybe I put in the wrong URL. alsoasked.com?

Asked, E-D.

Oh, E-D. Okay. Alsoasked.com. Find the questions people also asked. Oh man.

If you just Google people also asking Google for their feature name, we're number one, of course. I think it's the most overlooked keyword data. We've been using it for years because it's the only keyword data that firstly gets updated within hours. Like if something's on the news, if there's like an election on, the PAA results change within hours. So you get the first signals of how the intents shifted. And, you know, Google has talked before about things like time to result. And I think if you're writing any— so once you've decided what you're going to write about, you put it into Also Asked, and you're essentially getting a list of what Google is expecting you to answer on that page in terms of the completeness of the topic. If you're a brand, so if you type in something like Revolut, like the online bank, It gives you a really good brand footprint as well. So for Revolut, it will say things like, is Revolut legit? Is it UK? You know, what UK institutions back it? Is my money safe in Revolut? They're not even necessarily phrases I would target, but I would tell the team our brand kind of impression that we need to make on the homepage is people don't know us, they don't trust us. So we need to establish that on the homepage because that's the emotional trigger behind these searches. So there's so many different uses for it. We actually just published in the— we just launched an insights section. There's a study there by Chris Green, really brilliant, experienced technical SEO, did an independent study that showed the number of PAA questions answered in your content correlates to how well you rank for the core term. So Google is looking for that completeness. Um, we've got a whole process set up for, uh, like AI keyword research where it helps you map out the conversations. Um, I've done talks on that and they're online. Um, and we've actually got an integration with Screaming Frog as well. So if you have Screaming Frog, one of their custom JS, uh, plugins that comes with it is our one. And what it does is it uses ChatGPT and also ours that you can crawl all of the content on your site. The page headers or titles get passed to Also Asked to break out what questions might be asked about that topic. ChatGPT then automatically determines if you have that answer in your content, and then what you get is a Screaming Frog crawl of your 1,000 content URLs, and it gives you lists of questions that you haven't answered in that content. So it's one of the things we used to do manually for content auditing that now takes Like you can just press run and then leave it for an hour, come back and it's like, here's where your gaps are, or here's where the intent has shifted. Because sometimes what people want from a query changes over time and that's reflected really well in this data. Yeah, I've, I, for me it's clutch. It's like once I've decided the topics and what I'm going to write about, that's the main tool I use to determine, you know, what I need. And again, it's not about answering them verbatim in an FAQ format. It's about, you just need to include this information in it.

I searched makeup software and it's, I didn't change the country or city, so I don't know, it was just set to London and it's still getting the results. I'm really curious to see. The people also asked for makeup software. I searched it a few minutes ago and it just keeps refreshing and it says fetching live data, ETA 1 minute. Yeah.

So in— so the average response time was about 6 or 7 seconds until December. And basically in December, Google completely changed all of their kind of anti-scraping mechanisms. So now a lot of the time we're solving, uh, live reCAPTCHAs in the background, um, which is something we're currently working on. If you're— if you make an account and log in, basically when you do the searches, they'll run in the background in your dashboard anyway. So you can do like a dozen searches And then just come back. Yeah, they normally take a couple of minutes sometimes.

Now, what's the pricing? I'm looking right now.

Starts from, depending where you are, $12 a month.

Nice. The reason that this is so— another reason this is so cool, you're friends with David Quaid. David Quaid is who connected us. And David loves People Also Ask. It's one of like his techniques. Then he came on the show and he shared it. And what he does is he finds out people also ask questions. He gives them to Perplexity. He says like, give me plain text of just like one paragraph answers to all of these. And then he takes those and, you know, changes it for his tone and like makes them more readable. Then just puts them onto his site. And he takes the ones that start ranking. Then he builds out those pages. He uses them to interlink to other pages to pass on PageRank. And, uh, and he loves people also, and, and increases his, like, I mean, I think we were talking about topical authority, but he's like, yeah, increases topical authority and it improves the site as a whole. And he, uh, so it's, it's very cool that you have this tool. Mark Williams-Cook, you are a legend. This was so much fun. And you, man, I still have like so many questions for you. I hope we can, we, we can do this again. Thank you again.

For sure.

For coming on the show. For everybody, for anyone who's not already following you, where should they find you?

Best place is LinkedIn. Yeah, just search Mark Williams-Cook. I think I'm the only Mark Williams-Cook online. So for better or for worse, it's very easy to find me.

I really appreciate it. This was— you're like, I've made so many episodes about your stuff. So yeah, thank you again. For everybody who watched on YouTube, thank you so much for watching. If you listen on Spotify or Apple Podcasts, thank you so much for listening. This is episode 961 of The Edward Show, 961 days in a row doing this podcast. And I will talk—

Oh yeah.

And I will talk to you again tomorrow. Bye now.