Andy Chadwick says: “With all the chatter of AI and automatically generated content, you can definitely use it to speed up what you're doing and optimize what you're writing, but don't rely on it. There're certain software out there, not naming any names, but that basically claim to, and they do automate the whole writing process for you. They'll basically write the whole paragraph, write the whole blog post or piece for you. So my tip is not to rely on that. There are tools out there that aren't developed by companies that aren’t Google, that can pick up on the likelihood of something being AI generated. If they're being made by hobbyists and not someone with the resources of Google, and they're picking up AI generated content, you can only imagine what Google and Bing and these other ones can do. Google have also announced a new ‘E’ in the Quality Raters guidelines, which stands for experience. I think the jury's out for how it's going to work out what ‘experience’ means or how that looks, but certainly, if you have first-party experience in talking about a thing, it's going to come through a lot more than AI generated content. Being able to overlay your actual, original thoughts and put that original spin on things will definitely come through, and I can see it myself when you look at a piece on even something as inane as "how to tie a shoelace", you can read through what is generally spat out and something that's been written by someone with more experience. I don't know if there is someone with more experience of tying a shoelace, but that was an example.”

So, when it comes to using AI, it feels like we're on the precipice of AI being good enough to be able to write significant volume of content and get away with it. But as you say, it's easy for search engines to detect if something has been written by AI. So how do you actually leverage the power of AI without actually being tripped up by search engines, and perhaps even not ranked at all?

“Firstly, there is a case for helping to outline an article. Without plugging anything too much, we own keyword insights, which helps to outline blog articles, and you can use it to really quickly pull headings together, and you can use AI to help come up with some ideas you haven't thought about. Let's say you're talking about writing an article on everything you need to know about mortgages, and one of the headings is "what is a mortgage?". Now, if I was to write that, it might take me five or six minutes to put that together. But you could use AI to generate a piece of text that answers that quicker because that's a pretty standard question with a standard answer. I would then go over it and rewrite it to put my human element into it. But then there are questions that will be in that piece, like “how to find a good mortgage?”. Now that question, I would say, should lean more on personal experience, someone who's actually gone through looking through a mortgage. You could generate that piece with AI, but I imagine it would say the same thing, because it's obviously based on learnt data, it would say the same thing every time. A lot of these tools, you can refresh it, and it will rewrite it, and they'll say the same thing in about five different ways. So definitely generate it to see what it's saying, but then I would put my own personal experience into that as well. What have I found looking for mortgage rates? Are there any additional insights that I found that aren't there in all the standard articles out there? And that's really how you're going to stand out. When I said at the beginning of this, don't use AI to generate content or don't rely on it. I don't mean necessarily that you'll get penalized for it. A lot of people are saying you'll get done for it and Google will penalize your website, but I don't think you'll necessarily get penalized for it. Basically, don't waste your time quickly generating a load of AI paragraphs and then publishing it, because it won't do as well as if someone then puts that human element on it.”

Is it a good practice, then to combine them both? Then you perhaps take advantage from the elements that AI comes up with that you haven't actually thought of beforehand? And then perhaps you're interviewing someone on a particular topic just to hone that personal experience from them. And then you incorporate both elements on the same page.

“Yeah, that's exactly what I'm saying. There are tests going on, for example, there are quite a few people I follow on Twitter doing this, where they're generating the whole thing using AI. I think that letting it get you 50% there, and then going over it with a human element is the obvious thing. You need to be careful though, because some of the stuff it's generating isn't factually accurate, which goes without saying, and if people aren't rewording that, then I don't know what you're doing in this industry, because that should be the first thing you change. What I'm saying is even the things that are factually accurate, and they make sense, and they read like a human, if there's a capacity to bring a bit of your own experience to it, that E for that new EAT, then definitely do that. You should also play around with the wording of it to make it not seem like everyone else's. The problem with everyone using these AI tools is that everyone's going to start coming up with the same things. I know it can rewrite it in a million different ways, but they all pretty much read the same, and that's going to become very obvious.”

So that's using AI to augment longtail keyword phrase-based blog posts. Is there any type of content on a website, that you absolutely would not use AI for?

“Definitely anything to do with the medical stuff because It changes all the time. I don't want to keep bringing it back to EAT, but for medical things in general, you do need an expert and someone who knows what they're talking about. I've seen it, particularly with CBD, because CBD websites are popping up everywhere, and all the content across all the websites is exactly the same. We're fortunate enough to be working with a CBD site that isn't the same and they’ve have come up with a different type of CBD. They're actually making it themselves and they've developed it in such a way that your blood absorbs it four times better. So we originally started to butt heads because part of what we're doing with them in our engagement is writing their content. We were open with them and said we can get you 60-70% of there, but then you'll need your in-house experts to go over this and tweak it, and there's been loads of interesting things we found. Every CBD apart from this website tells you to take it under your tongue because apparently it's better for infusing into your bloodstream and all that. But this website has actually gone: no, there's no scientific proof. The company is actually run by two scientists and they've got loads of investment from VC backing because their product is genuinely different, and they've just come away and said: no. We've written this whole article on them and how to increase CBD absorption. A big section of it was about putting it on your tongue because it's where the blood vessels are, and instead the clients said, "there's actually no scientific evidence in that at all. Actually, this is how we would increase it", and they've put their own spin on it. And it's already ranking in position one or two. They have got basically no authority in the US; the site's brand new, but because they've actually put that slanted experience on it, the ranking increased. It's gone against the grain if you like, but they have backed it up. They've linked to papers and things. I didn't know it was an experiment at the time. It was a really quick example for us showing that maybe Google is looking at experience somehow. So, for anything medically related, the AI just churns out the same thing, because we did generate that using AI. Every version of an AI tool produces that same paragraph about putting CBD under your tongue. And then we went over it with a human. That's the best way to do it.”

That's great piece of advice there because many SEO say things like, if you want to rank for something, look at the SERP, see what the SERP is spitting out and follow the structure of already successful pages. I think maybe some SEOs take that advice and think, okay, I need to produce very similar type of content, but you can actually combine both pieces of advice there and actually follow the structure of successful pages but add your own original views to give you that increased opportunity to rank well.

“Exactly - we tell our content writers to look at the SERPs and get all the best headings from all the top 10, because what you find in position one isn't always covering the same topics in position two, and in position two it isn't saying the same things in position three, and so on. Some of them have maybe 80% of the same headings, but if you take a Venn diagram, they have got headings where they sit outside of each other, and then way in position four or five, you'll have this website that has no authority at all but is ranking really well. That’ll be because it has got other headings they haven’t thought about. So, what we do is try and combine the best bits of all the headings, and then we look for an alternative. We go to places like Reddit and Quora, and the idea behind that is if people are turning to forums to ask questions, it's because they haven't found a suitable answer in an article online. So we'll scrape them for any additional questions that may not have been answered in the other 10 or 20, and the top 10 will include them. Then if it's something medically related, we will ask someone who maybe has a different opinion to offer a more balanced idea about it and include that as well. But yeah, it’s important to not follow the same headings in position one.”

Moving on from headings to titles and meta descriptions. If an SEO is in charge of a big website with millions of pages, is it good practice to use AI to generate titles and meta descriptions? Or are they going to be shooting themselves in the foot?

“I use AI for meta descriptions purely because I hate writing them, and according to Google themselves, meta descriptions don't have any bearing on rankings. However, I will often tweak them. I’ll use it to generate the main part of the description, mainly because I just want to summarize my article quickly and get it done. When it comes to page titles, I nearly always do those myself, because, again, I do like to stand out. AI would generate one that looks like the top 10 or the top 20. I want to stand out from them, so I'll do the standard stuff, include a key word, use title case, all that stuff, but I'll always try and put a different slant on it just to make it stand out a little bit. So that's personally how I would use them. Again, I don't think you're going to get penalized for it. I just think if you're not doing something to stand out, especially in a world where everyone's going to start using AI, you just won’t have the edge over people who are doing something to stand out.”

So, you've shared what SEOs should be doing in 2023. Now let's talk about what SEOs shouldn't be doing. So, what's something that's seductive in terms of time, but ultimately counterproductive? What's something that SEOs shouldn't be doing in 2023?

“They shouldn't still neglect things like link building. We're getting better at Google, we're getting better at evaluating expertise and experience, we're getting better at understanding things, AI is becoming amazing, and you don't need links anymore. But I hugely disagree with that. At the moment, we've been running tests ourselves, and good links still work. Don't go after rubbish links, and don't buy them off dodgy websites, which a good link building strategy wouldn't have done anyway. The main thing to take away is links are still working. It's a good double negative. Don't be ignoring your link building still, because of all this fancy AI stuff.”

You said good links still work, so what is a good link in 2023?

“We've got guidelines that all our staff follow when they're evaluating links. We have link building strategies, and we get links and we make them stick by quite rigid guidelines on what we will and won't accept and what we will and won't disavow and that kind of thing. I’ll run through top things. DR, DA, TF whatever you want to call, it doesn't really matter. We have clients who sometimes go “we want links, and they've got to be a minimum DR of whatever”. For me, I look at two things. I look at the domain first of all, and if the domain has got a steady traffic curve, and a third party tool is fine if you look at Ahrefs or Semrush or anything like that, a steady traffic curve is fine. It doesn't even matter if it's going down slightly or up slightly. If it’s steady, it is good. If it's going rapidly up like it's jumped from 1 to 100,000 in two weeks, then there's something dodgy about that site, and I wouldn't want to link from there. Conversely, if it's gone from 100,000 to 1 in a few weeks, then this site's been penalized, or it is doing something wrong, so I wouldn't want a link from that. The DA, DR, TF irrespective if it's going up or is going down slightly, I don't care. So first of all, that's what I look at in a domain level.

Then I look at a page level, and I'm looking at first of all, is the page related, and actually is the domain related to whatever industry I am linking to. I don't want, newspapers, the Dudley government or Birmingham government sites because they don't do anything. If we can get a link from a page in a niche which is aligned, then that is a million times better than one that isn't. We have done tests where we've seen an interesting case study where a site, I can't remember the site, let's call it a traffic cone website, bought a sort of similar but not quite other website, it acquired one to do with bags or something. But the bags were for builders, so sort of related traffic. The idea was they were going to fold the two websites together, link them all up, redirect and get all the link equity, because they saw that the bags website had a lot of good links.

What ended up happening was the pages which aligned with the bags with whatever pages on the traffic cones site did see a big jump, but the pages that didn't, so for example, the homepage, which had a load of links pointing at it, they redirected that to the homepage of the traffic cone, to get all the link equity push there, and we actually saw the homepage drop it for all its rankings. We believe that the minute we lifted that redirect, they went back up again, and we believe it's because the links just weren't relevant. So the pages that were relevant went up, but the pages that weren't relevant actually went down.

So just a link for link sake isn't good enough, the link needs to be contextually relevant, otherwise, you actually may see the reverse of what you want. Even if the good links from good websites, you might be confusing the intent a lot.

The other thing is we look at the other blogs, or whatever is in that website, we put the images of the writers through TinEye, which is an image reverse lookup software. If we see that that image is a stock image, we won't have a link from them, because we know that they're just producing stock content with stock writers which aren't real or genuine. We want links from real sites, and we make sure that the anchor text is contextually relevant. I've seen it loads, where we've audited links and we've seen someone talking about houses in general, and then they've just rammed in the word “wallpaper” really randomly just to get that anchor link in there, and it doesn't work. So, we look at all these things.

To summarize your point, it's going to be from a contextually relevant website, that has steady traffic in one way or the other, the anchor text has got to be quite natural, not just rammed in there, and that the actual rest of the quality of the content on that website is also good. It's not from stock image writers it's not from guest blog posts, it's from genuine people.”

That was one of the most valuable additional questions that I've asked actually on this podcast series, sometimes I just select that little nugget from someone's answer, “a good link” and that that was a great summary of that. I loved you mentioned the fact that you looked for consistent site traffic and I think that's a metric that I haven't considered looking at before to establish whether or not a site is valuable to get a link from. I think that's a wonderful starting place.

“Majestic is obviously a great place to start working that out as well, so we've been using Majestic to find those things. We've got a checklist if anyone's ever interested, but I can share that another time.”

Andy Chadwick, Co-founder and SEO strategy lead at Snippet Digital. You can find them over at snippet.digital.