Feb 24, 2023
Unless you've been on the moon the last few months, you'll have heard lots about GPT and Chat GPT - the chatbot built upon it. But as a marketer, what does this hyper-acceleration in predictive AI actually mean for your job, the world of search, advertising and content, and how can it actually open new opportunities?
This episode features digital transformation expert Clark Boyd and SEO expert Danny Richman chatting during the Q&A of our recent webinar on GPT, as well as tips and futher thoughts recorded specially for the podcast. All brought to you by our podcast (and webinar) host Will Francis.
If you're not penalised by Google for badly-written content, you'll be penalised by your readers and then Google will penalize you after because they'll see that your readers didn't like it... Clark Boyd
Listen to the end to hear Danny's and Clark's top 3 tips on what to do next with AI to help you in your digital marketing role.
Some of the resources mentioned for following AI developments mentioned are Google Alerts:
If you enjoyed this episode please leave a review so others can find us! If you'd like to be a guest on the show, drop us a line at email@example.com.
Danny Richman is the founder of Richman SEO Training. Danny has helped organisations grow their online traffic since the birth of the internet. He has been actively working with GPT-3 to explore and create tools for marketers (click here for examples). Clients include BBC, Vodafone & Salesforce. Danny also volunteers for the Prince’s Trust supporting disadvantaged young people to start their own business.
Welcome to Ahead of the Game, a podcast brought to you by the Digital Marketing Institute. I'm your host Will Francis and in this episode, you'll be hearing handpicked highlights from our recent webinar about ChatGPT and GPT3, in which digital transformation expert Clark Boyd, and SEO expert, Danny Richman, both gave us some great insight along with some tips, tricks and tools to try. And throughout, you will hear reference to some of these tools, particularly ones that Danny built himself that very easy to use, and very effective SEO. And we've linked to those in the show notes, I would encourage you to go and check them out. Even if you're completely non technical, you should be able to use them in your work. Anyway, just as a bit of background ChatGPT is a computer program developed by open AI that can understand and generate human-like responses to natural language input is based on a type of deep learning model called GPT. And has been trained on very large amounts of text data to generate responses to a wide range of topics. It can be used for natural language processing, chatbots, and text based AI assistants. And yes, I did ask charge GPT to write that definition of charging. Anyway, let's hear all about what GPT is, how it works and how we can use it in our digital marketing. Quite a few people are asking about problems with accuracy. What would you say about that?
Danny: Yeah, so it has been a much bigger issue up until quite recently, because you had this problem where GPT three had this kind of corpus of training data, which you know, gathered from the web from all these different sources. And so it was generating responses based on that information. The thing is, once you start playing around with it, you soon you're able to see that GPC three, even though it does sometimes give inaccurate information. It knows when it's giving inaccurate information, it does have an understanding of where it's not certain about something. And so the latest developments, what we're now seeing is where it's coming back with a response where it isn't quite certain, or it realizes that that information may not be current, it's then pulling that information from the web, and then generating a response based on that live information from the web. So I think that whole issue around accuracy, I mean, you know, will it ever be 100%? Perfect? I don't know. But then, so yeah, I think you know, there always may be some accuracy issue to some degree, but I think we're going to see in a very short space of time, that almost becoming a non issue.
Okay, so what about copyright? Like if the training data for these tools is previous content? That was someone's content? Right? Where do we stand when we create sort of new content with something like ChatGPT?
Yeah, that's an intriguing one. Yeah. Cuz I saw that was added mentioned as well. The recent win for Hermes as a legal victory over a digital artists that use one of their products to create an NFT that then generated a large income. So there is that question of whether the legal industry is able to keep pace with this level of changes going around, there are actually quite a few lawsuits going in against open AI at the minute and escapes me at the moment, I think it might be stapled to fusion, there is no an opt out option for artists. So you can say I do not permit for my artworks to be used within your algorithms. So there might be a bit more of that sense of control in future opting out of being part of the training corpus in the first place. But that doesn't seem all that useful in the long term. Because if everybody did that, there'd be there'd be nothing left for them to use. So there were questions as well I saw about, well, could you just create a product using this and start selling it? And yes, I mean, that is what a lot of these people are already doing. Because there is no sense of any new legal framework in place. There's nothing new to me, that's, that's really what open AI is trying to ask people to do. Stable diffusion is saying, Yep, make a copy of our code and go and go and amend it to whatever you want to do. So there isn't really a sense at the moment that these things are, say a big danger, if you wanted to make a service and start selling it, I think that's always going to be fine. The other one is more about the underlying data ownership or the product ownership. If you have say, submitted artworks, or it's taken, it's taken your style of writing, and someone then says, Write me a novel in the style of whoever and then they go and publish the novel, and they make a million pounds off it because it's actually done better than the original one. That's another case. So I think we can separate. There are tools using the API's using the frameworks. Those are pretty much what are already making 10s of millions of dollars a month for a lot of companies. And then there's the other copyright issue around IP essentially or the intellectual property you would have as a content creator. That That's the murkier bit. And there are already some pretty big cases going into the courts as we start. So I don't know what will happen with those. But it looks like open AI might actually have to pay out on some of those and stay with the fusion is allowing people to opt out of its algorithms.
And what about search engine rankings? Do you think, Danny, that Google and its counterparts will detect and punish generated text articles?
Yeah, I mean, so I mean, as I said, at the kind of the beginning of my slides, I've not personally found this technology to be quite good enough yet for producing long form content that for me is not its core strength. In terms of SEO, it's been a real mixed bag from Google, because on the one hand, they're saying don't produce whole articles generated by AI. On the other hand, they're saying we we have no problem with people producing content produced with AI, we don't care how its produced, all we care about is the quality of the content. And I think that that's probably actually the in the middle of all of that there's kind of the best response, which is absolutely feel free. And I do use this as at all, as a first draft or, and to generate kind of an outline and a structure for content really, really useful for doing that kind of thing. But then it absolutely needs a human touch to then go over that, refine it and prove it, make sure it's completely accurate in all these areas. But as a writing tool, it's absolutely fine just to leave it churning out automatically 1000s of pages, I would not advise doing that. I would add
that as well dynamic, completely agree with that. And I've written a bit about that. Because it gets it gets a bit murky sometimes, because if I use it, sometimes I'll write a whole article. And then I'll say, Actually, could there be a different structure to this, and TVT will say, you can add these headings that might be helpful. And the authorship then is a little bit murkier, you know that it's getting involved. And I can even say, rephrase these three sentences, and it will do it and it'll do it pretty well. I might just use those sentences. So what it's 70% my work, 30% ChatGPT. But the big part of this, of course, is we don't think that often about the audience. So if you could churn out this content, and Google wouldn't penalize you, well, the ad reader will, they're going to look at it and know that it's not good enough. If you wouldn't read this yourself. You shouldn't be pumping it out there for supposedly customers to read at the end of it. And that's the bit where I think Google is right not to get to wait, wait, Dawn in isn't AI? Is it human? Whatever a lot of the human content out there is is pretty awful. The AI can do that pretty well.
Yeah, Google's just ranking on quality and relevance, right. And it's refining based on clicks and dwell time on those results. And so useful content wins. And that's it.
Yeah. So if you think if ChatGPT was creating content that you thought, Oh, this is fantastic. But I better edit it a bit, just so I don't get penalized by Google. That's not really the way to go about it. If you really think it's almost like a constant sort of smell test, if it is gets to the stage where it's good enough. And that will be a while away, I don't see how it would be able to do that straightaway, then then fine. You could pump out that content. If you're not penalized by Google, you'll be penalized by your readers. And then Google will penalize you after because it will see that the people didn't like it.
Yeah, I really liked your point, Danny, about how this is not just a lazy copywriting hack is currently not the best copywriter on the planet by far. And he's actually really good at lots of other types of jobs, which can be very hard or laborious for us. Yeah,
the only thing I would say to that well, is that you know, I mean, I've been involved in tech now for 45 years, and I've kind of, I've gone through every stage of tech from the first ever personal computers, the internet, the mobile phone, I have never seen anything move as quickly as this. I mean, I am giving a lot of talks and presentations at the moment. And people are asking me to submit my slide deck like a week or two before the event. It's completely pointless, because it's out of date. I can guarantee it's gonna be out of date. This is changing by the day. And so when Clark says it may be a little while before we're going to get this tech producing high quality content. When he says a little while, you might mean a month.
I know. I know. Because I noticed, Danny, I saw your iterations of our slides and saw the first version of yours two weeks ago, and you had mocked up, you know, the Google homepage 2027 best laptops? It could be the end of the week.
Yeah, a week later. And that's the reality. Does
this, pull the rug from under things like Jasper and copy.ai
Interesting with that, because we see Microsoft saying that they're going to release tools on Google as well, to allow developers to get a bit more hands on with the models and train them to their own purposes. And that might allow those companies then to create that that moat that we'd like to talk about. So at the minute, if you if you really wanted to create a copy of copy.ai You'd be able to do it. And that's what venture capital money is going into at the minute. So start-ups are incentivized to try and create something similar problem is, in six months’ time, six weeks, even six days at the minute, the rug can be pulled from under them, what they'll be looking at is, are we going to be able to adapt this so that it becomes something unique to us even that they could register as their own IP. And then it becomes the Jasper framework built on top of open AI built on top of Google's Lambda, that will be more interesting for them. So there was the word the word she'd say, a few, a few crumbs from the table yesterday from Microsoft and saying they will open this up to developers, and you'll be able to train this for yourself. No, I still think other companies, individual companies could just train their own version that would be adapted to their very own purposes, you know, it's going to be built into Microsoft as your, so why wouldn't you if you're even a medium sized company, and you know, I'm no great programmer, or coder, Danny was talking about how you can get ChatGPT to or GPT. Three to write the queries for you. I've been doing that myself, for a lot of programs, it doesn't seem that the the barrier to entry is going to be so high, that companies couldn't just generate their own versions of these tools, as well. So there's a real rush, I think there are a lot of companies that are very highly valued at the minute because they're in the right place at the right time, that will disappear in six months, they just won't be valid anymore. They need to build that that we like to call it a moat in our business. Good talk.
Okay, so in terms of adding this to our marketing toolkit, is getting good at prompting these kinds of tools, writing effective prompts, something every marketer needs to learn to be good at,
I think it's quite likely. Yeah. Yeah, I mean, I have to say, that's one of the areas that I find most really enjoyable about using this technology. That process of constructing a prompt is, is a kind of a mix of logic and creativity, I think kind of Clark touched on that it is an art and a science. And just noticing how the results you get can vary so much, depending on a very careful selection of words that you put into these prompts, is a really, I find a very creative process. I've really enjoyed that. And it's something that you get better and better at over time. And I think yes, it is going to very likely be a skill that many people will start having to get to grips with. And there'll be people that are good at that, and people that are less good at that less good at that. And I think
a way just to get into it. And Danny touched on this, as well as understanding those fundamental mechanics that are going on here. What is it that the tool is doing in the background? And how could you adapt that to your own means so I put together this little chatbot yesterday just to demonstrate what this can do for someone. And I put it on just staging website, I have something I'm gonna put out soon. And within every page, it's making calls to the open AI API and encourages you to ask questions. And all I've done within the API call is append some different things. So the person could come in and say, of how do I explain broad match versus exact match. But I've just put a little bit at the end of it that says, explain this using an analogy. You are a seasoned digital marketing expert explaining this to a novice. So the person doesn't see that they're just saying, What's keyword research. But the prompt that's going is saying, make sure you use an analogy, keep it simple. And just understanding those little building blocks doesn't take a huge amount of time. But it puts you way ahead because it looks like magic when people go to use it. And that's what this has put into our hands. But you do just need to get a little bit hands on with some of that there's some good resources you can look at, they'll explain to you how it all works. Build your first thing, and I'm sure Danny would agree with this, getting hands on realizing what's possible for yourself, will just open lots more avenues to explore.
Now, a few people have asked about the risks, letting this kind of tech loose on your website. Have you noticed any worrying responses in your website? But
oh, yeah, the main thing that I want to do before because it's really just so that'll help assist them don't want to have within the website is figure out the best ways to train it on my own examples and ways of thinking I'm putting restrictions on it. So there are things that it won't talk about, because the use case that I wanted to do is pretty clear trying to teach people about digital marketing here. So you could at the minute ask it about, you know, the moon landings happen or whatever. And it will chat to you about that using an analogy as if it's a digital marketing expert. That's not really what I wanted to do. But you can confine those things quite easily. And there are a lot more capabilities coming to us that will allow us to train it on our own corpus of materials. So I have loads of digital marketing examples, loads of ways of explaining things and everything. So just teach it. These are the kinds of things to do. It'll create something new but similar, and it won't talk about those other things because, like people are mentioning if I were running, say a legal website or something really important, like that's not just my own little chatbot for Digital Marketing, I would want to be very, very, very careful with that first, and I would want it to do a lot of testing before it went out in the wild.
Yeah, Danny, you showed us a really nice use of this technology in a form on your WordPress website, which offers dinner recipe suggestions based on what's in the fridge. So tell us more about that. And again, we'll link to that in the podcast episode show notes. For everyone.
So many people have a WordPress website. And if you do have a WordPress website, there's a very good chance you've got a gravity forms plugin, because it's one of the most popular forms plugins for WordPress. A little while ago, they released an ad on four Gravity Forms, which allows you to make an open AI or a GPT3 Request API request from the form. But they've really kind of made it so easy for people because you don't really need to know anything about the technicalities behind it. All it does is takes a field from the form. So you can design any form you want. So in my form, in the example I showed in the deck before, it was a field with only one a form with only one field on it, which was my wife enters the ingredients that we've got in the fridge, you know, that then gets submitted over to GPT, three in an API call, together with a prompt that says did generate a low carb recipe with the following ingredients. And then you take the value of that field that she entered, and it sends it over and then sends a response. And then when the response comes back, you can then choose what you want to do with that you can then have it displayed underneath the form on a different page used in a post or another page, you can send it wherever you want. But it's really, really easy to use. And that add on the open AI add on is completely free if you've already got Gravity Forms.
Well, that's the real big leap that's been made recently. You know, the tech itself has been around for a while. But it's the ease with which we can implement it through open API's integrations or API. And of course, most famously, ChatGPT like anyone can leverage this without touching a line of code.
But you know, like Clark was mentioning before he's trying to create this chatbot, this digital marketing expert Chatbot. You know, you could have done that with that gravity form and just had a form as the Chatbot, you would never have had to use any code at all to create that. I don't know whether you did need any code. But you could have done that just using a gravity form, get the response back, that you asked a question, that would be one application. But just think of all the different tools and fun apps that you can build on a website that would then drive traffic into that website. So many applications, it's only limited by your imagination.
Hello, a quick reminder from me that if you're enjoying our podcast series, why not become a member of the DMI so that you can enjoy loads more content from webinars and case studies to toolkits and more real life insights from the world of digital marketing, head to digital marketing institute.com forward slash ahead of the game, sign up for free. Now back to the podcast. People are asking how this will affect advertising? Like? How will it affect ad copywriting? And what happens if everyone's copy is written using the same kind of tools?
Well, one of the big bets this, this actually reminds me even though what's happening is a bit different. The underlying pattern is similar to when Google move to these p max campaigns actually or when it moved to Universal App Campaigns a few years ago, and we were asking the same questions. So we're, we're all using the exact same technology in the middle bit. What's going to be the difference? Well, the difference is going to be things like your brand, it's going to be the inputs that you put into the machine to allow it to do something, it's never going to be completely automated, it isn't going to just mark it on behalf of every brand. As one simple machine, it's going to need inputs, it's going to have to be trained on things, it's going to be using me as a language model large language model, it's going to go out there and see what's written about you and adapt itself to that as well as what you're feeding into it. So just like it had happened with universal app campaigns, and then p max campaigns, a lot of people, thereby, by extension companies got a bit lazy, because they thought, well, the machines doing it all, it's all automated, we just need to hit the big button and hope for the best. But the ones who got the best results really started thinking about the creative assets that they were putting into the machine at the start. And they found that the outputs were much more effective. So again, we need to break it down a bit and see where are we having our own input, there will be a lot of parts there. And that's where we have leverage over the results. So the way that it affects online advertising could be the fundamental mechanics of it from the customer's point of view, and that they can interact more directly. They can have an adaptive experience, they can converse a bit more, there'll be new content fed back to them. That means that the monetization model is different. So if you're a Google search, you haven't just got static search queries that you can serve up in and sell to advertisers all the time. If that's going to change as well, and the way that brands go about it is going to be different. So it really reorders. Absolutely everything that we're thinking about here.
Yeah, that makes today's setup sound quite quaint. Yeah, it's,
I think it will become much more dynamic, it will require new ways of creating and monetizing value. And that's why Google has been reluctant to get involved in this, it has a pretty amazing profit making machine at its disposal, and it does not want to give it up.
That's very true. I mean, I'm, I found myself feeling a little bit reluctant to make any kind of predictions around what's gonna happen in the future around this technology. Because it's just changing too fast, really. And I'm just, I'd make myself look like a fool. The one thing I would say is, I don't think anyone gets into digital marketing or SEO, because they don't like change. If you did, you made a big mistake. You're in the wrong industry, if you don't like change. And as someone that likes change, and embraces it, and loves the fact that there's just always something new to learn, that you've got to stay on top of, I'm really enjoying the ride. And that's, that's what I would recommend to people is, you know, sit tight, hold on and enjoy the ride and learn as much as you can. And be prepared to change everything about the way you do stuff on a day to day basis. Don't get too stuck in doing things the way you've always done them, because that's very likely to change. So that's what I would say to people in the industry.
Well, I won't ask for any predictions, because as you say, that does seem a bit foolish at this point. But where would you suggest that people go to keep up to speed with these topics?
For me, it's been, Twitter has been my main source of learning following the right kind of people on Twitter. So if you're not sure who to follow, go and have a look at who I'm following on Twitter. And anyone I've followed within the last year is going to be something to do with AI. But there is also an excellent newsletter that goes out called bends bites, produced by a guy called Ben tousle, and it's a daily newsletter. So, you know, hats off to him for putting something out every day. But it's just a really easy read. It's just kind of links to all the cool stuff and latest stuff that's happening on a day to day basis and AI. So check out Ben's bytes and just sign up for his free newsletters a great way to keep abreast of what's going on.
Yeah, I have a few free newsletters. So follow as well. Even just simple things like getting TLDR in every morning at 11am will give me the top five things that have happened in the last day. That's a good one to start with. I definitely curate who you're following on social networks, Twitter's definitely the best one for that even. And it's current guys with things breaking every day, it's still where the interesting people are sharing interesting thoughts on, I just have basic Google alerts set up for things as well just try and read what news is coming in. I'm gonna get that then integrated with GPT. Three is going to tell me a little bit more about those stories and summarize them for me and try to build that into a workflow. But yeah, just think about those ways that you can get the news as it happens, because there's so much happens every day, if you try and catch up with the end of the week. What happened on Monday won't even be relevant anymore. I keep getting webinars pushed back, because people are like, have you seen what just happened this morning? That's gonna take let's think about it. And then by the next day, then I'll go Microsoft's got a new event come in, and it's one to keep up with on a daily basis.
Indeed, that constant change is absolutely what makes it an opportunity for absolutely anyone who's keen enough to keep learning. Well, that was taken from the Q&A in our recent webinar. But as promised, I had a few other questions to put to Clark and Danny. First up, tell us how can people get started with ChatGPT? What's a good starting point? What are some good prompts to try, or my
advice would be not to play around with ChatGPT. But to go and register an account in the open AI website, and go and play around with the underlying API. Once you log into the open AI website, what you're going to see is an area called the playground. And in the playground, you're going to see many, many different examples that open API provide, you can then tweak those examples and play around with them, adapt them. And you're also going to see a whole load of settings on the right hand side, that you can then adjust and get all kinds of different results coming out. And that's the best way really, to learn how this technology works. Understand how the underlying API operates. And you'll learn much more through using the playground than you will through playing around with ChatGPT.
I would say the first thing we need to do as marketers is try and understand that distinction between ChatGPT as this API wrapper, it's a user interface that sits on top of some underlying technology. If we are simply using ChatGPT as that interface, we're going to limit what we're able to do because we'll really just have to copy and paste what we get from it into somewhere else. It won't be embedded in the way that we do things, we can access a lot of the underlying technology there. And I know Danny has some great ideas on how you can do that already. But if you realize that you realize that you could essentially build your own version of ChatGPT, you can have it built into your website straightaway, it really wouldn't take too much, no good prompts to try, I still think it's worthwhile playing around with it, because it's very friendly user interface that has taken off very quickly, because it does show you some of the inner workings is one way of understanding it. In terms of prompts, I would say, start small and have something in mind that you're trying to get to. So let's say you've been challenged to come up with 10 ideas for YouTube videos for the next quarter. Go to ChatGPT, and simply start with suggest 10 ideas for YouTube videos for and then say a little bit about your company, and then start adding modifiers. And that will change the question that you're putting in, but you can also change how it responds. So you could say, go beyond obvious ideas, or my target audience is x. And play around with that and see how that changes the output that you get what gets you closer to what you're looking for. Now, that's great, because it'll get you hopefully 10 ideas that you can use, but it will also help you understand how the technology functions. So when you do come to using the API, for example, you'll have a better chance of getting the results that you're looking for.
Do you guys have any thoughts on what will happen between Bing and Google as they start to integrate this kind of ChatGPT or something similar into their search engines?
If you go to the Bing subreddit, you can see just how many mistakes their new AI is making already. If Google did that, it would be in big trouble. It made a couple of mistakes in its new release, and it wiped $100 billion of the company's market value is completely dependent on search. Another thing to say here is not just about the market position of the two companies or even their existing business models, which are radically different. I think first we need to say that Microsoft, sure it will see search as being important. But it's it's a means to an end. For them. It's not an end in itself. It will want to integrate this technology into its cloud computing services so that you can have your own ChatGPT that goes with you everywhere.
Well, this is going to be a really interesting time for the world of search. Bing and Google are approaching this from very different places. Google depends on advertising for 85% of its revenue. And that's going to mean they're going to take a much more cautious approach about how they integrate AI into the search results, they're going to be very, very wary about doing anything that might impact people clicking on their ads, they're going to be more wary about making sure that the AI is giving accurate results. Bing will be approaching this from a very different place, because Microsoft only depends on advertising for 5% of its revenue. And so they can take a much braver approach to this. They can afford more risk. And they can really just focus entirely on providing the best user experience. So it's very hard to see how this is going to pan out at the moment. But what we do know is that the world of search has changed forever.
Do you see any specific industries or verticals using ChatGPT or similar tech particularly effectively?
Or customer support is a really obvious area where AI is going to have a big impact. You know, at the moment, if you phone up a company and you go through to customer support or technical support, you're kind of disappointed if you get through to a computer rather than a human agent. Well, I could really see a time where that is reversed, where you're going to be disappointed to speak to the human agent, because the computer can be more knowledgeable, more helpful and be more specific to your needs. Again, in sales, I've already been working on a project where I've worked with a company where we fed it a whole load of information about that company's products and services and the benefits of their products. And so when a prospective customer comes along, and they describe their business, it can give really helpful detailed advice to the sales agent to assist that customer in choosing the right product for their needs. In the world of travel and food and beverage, I mean, some of the more obvious applications would be in terms of product recommendations. At the moment, if you go to a website like booking.com It's a very simple interface where you kind of choose your dates and a destination that will list a whole load of hotels. We're going to be able to really give some detailed information about the kind of holiday that we're looking for the kind of restaurant or bar that we want to visit or maybe the eye AI will already know enough about you to make some really strong and relevant recommendations. Another area where I think we're going to see a lot of development is in the world of professional advice. So law and medicine and financial advice, we're going to have bots that are trained, and have far more knowledge than their human equivalent.