Mike Rhodes 00:00
A mate of mine, Alex Hormozi, challenged his entire team to replace themselves using AI. It's a crazy full-on video, as Alex is a little crazy and full-on. But he just goes bang, bang, bang, bang, bang - through nine different tools of how his team are now using this, and he puts out 250 pieces of content a day using these tools.
David Bain 00:24
The Strategic Marketing Show is brought to you by Insights For Professionals: providing access to the latest industry insights from trusted brands, all on a customized, tailored experience. Find out more over at InsightsForProfessionals.com.
David Bain 00:40
Hey, it’s David. We're constantly hearing endless reports of how we should just trust paid ad platforms to deliver the right ads in the right places, and how you can now trust content to be written by AI. But, where should you draw the line? What should you automate? And what shouldn't you automate?
That's what we're gonna be discussing today, with a man who co-authored The Ultimate Guide to Google Ads with Perry Marshall, the world's best-selling book on Google Ads with over 140,000 sales.
He's spoken on four continents, at hundreds of conferences, clocked up over two decades in the digital marketing industry, and is CEO and founder of the award-winning Melbourne-based digital agency, Web Savvy. A warm welcome to the Strategic Marketing Show. Mike Rhodes.
Mike Rhodes 01:27
Thank you, David. Thanks for inviting me on. Pleasure to be here.
David Bain 01:30
Thanks so much for coming on, Mike. Well, you can find Mike over at websavvy.com.au.
So, Mike, where is the line between not enough automation and too much automation?
Mike Rhodes 01:42
It's funny, isn't it? How just the world has - I don't know about your feeds, but my feeds are just full of ChatGPT these past few weeks, and yet it isn't new technology. It's just a new moment where suddenly it became easy for everybody to play with AI; so easy that my nine-year-old the other night had it write her a bedtime story. And the technology has been around, as Google reminded us in their press release last week: “We invented this, everybody else is just using technology that we thought of!”
Mike Rhodes 02:12
Where's the line? I like to think of these tools as really good first-draft creators. You know, the hardest thing about writing is writing that first draft. It's that blank page, or that blank whiteboard behind me, and starting.
So, having an intern, an assistant, a co-pilot, be able to give you a bunch of ideas - things that maybe you wouldn't have thought of. Not relying on it to write the whole thing (whether we're talking about a blog post or a product description, or a job ad), but using an AI tool, for instance, to write that first draft for you is incredibly useful.
But, where the line is, is - well, the way I've been explaining this for a few years now is: look at the image creation things. You may have seen thispersondoesnotexist.com, which I think was 2018. That's how advanced “machines creating images” were. Now, we kind of take it for granted - after DALL-E, Midjourney, Stable, and all those others - that we can write a text prompt, and out will pop this incredible image. And they are incredible. But I think people that are trying to, with the perfect, amazing prompt, create that one image, I think that's missing the point.
Mike Rhodes 03:34
The idea is to play with it, to iterate with it. And if you find on YouTube, there are some wonderful time-lapse videos of a really good graphic designer going back and forth, and keeping part of an image and in-painting an area, and changing a piece, and maybe that takes 2, 3, 4 hours to do really, really well.
That's the same thing, the same mindset, we should be applying to the writing side. It's not that one perfect prompt is going to give you this instant, like, “Rewrite 10,000 product descriptions for my website. Here's my list of 10,000 products. Off you go.” That's a good first draft. But then, as humans, we need to bring what we bring to the table and not just take that, cut, paste, and go use it. Because, otherwise, the interwebs are just going to become a blurry facsimile of itself as this AI rewrites things created by AI and it all gets very, very messy. I'm not sure if I answered your question somewhere in there.
David Bain 04:33
But things are obviously changing very, very quickly. And, as you mentioned, the image search engine that you talked about first established itself in 2018, so that's five years ago from the point of recording (we're recording this in February 2023). And now we're at a point where the content - the written content, and the imagery - generated by an AI is quite incredible.
However, you're saying, “view it as a first draft”, how long do you think we'll be utilizing it like that? i.e., using it for a first draft and not just relying on it to create and publish content on your behalf without checking it.
Mike Rhodes 05:16
There is a great line, and I forget it in the moment, of how the future is best left to the crystal balls or something. There is a wonderful line (I should learn it) about that. I don't predict the future. I have no idea. The best AI experts, of which I am not one, cannot agree on where this technology is going and how fast it's getting there. And there are a lot of calls - I think, rightly so - for it to slow down (which it won't) and for us all to take a step back and sort of be a little bit concerned by just how fast this is wandering off in a bunch of different directions.
But it won't slow down, because the competitive advantages and the moats to be built are too big. But how long before it does too much? Well, people are inherently lazy. And I remember reading about Davos three years ago - all they were talking about, apparently, was RPA: Robotic Process Automation. Big, big business, just want efficiency, efficiency, efficiency. Push cost down. More, more, more. If we can squeeze a little bit more out, if those businesses can, then they're going to. And okay, if the quality is only 98% (or 95%? I don't know where that line is. If it's 80%, will they accept it? does it need to be 98% as good as the humans?) and, of course, the machine is getting better all the time. Compute is getting faster and better all the time, power is getting cheaper.
I was listening this morning to a piece on the way in about how, at the moment, the cost of, say, ChatGPT is about an order of magnitude above the cost for Google of running a search. That's going to come down. How quickly? I don't know, certainly within five years you would think - within two maybe. And that's a massive threat to Google, and marketing is going to change as a result of all of this. We've already seen, I mean, BuzzFeed - I believe their share price went up three times the other day when they said they were going to use GPT technology to write a lot of the content. And that's all they do, right? They're just this massive content farm.
And people coming out with these tools to see if content was created by an AI or not, is missing the point because all content will be partially created by AI and partially by humans. And some will be more so. Financial reports and sports reviews have been written purely by AI for many, many years. Now, because that's structured data that's easy for a machine to sort of wrap its head around.
But we think we're such special snowflakes. We think that “Oh, but the thing I'm writing is completely different. A machine wouldn't be able to do that.” and yet, we go to school and we learn writing frameworks. If you want to write a sales letter for a website, there are half a dozen or more different frameworks that you can apply, but they're all very similar. And a machine that has ingested hundreds of millions of pages of copy that used those frameworks can figure out what those frameworks are and get very, very good. And of course, we're all using - 100 million people are using ChatGPT now, so we're all training the machine to get better and better and better.
It's going to happen, right? I don't know when and I don't know what that curve looks like, but we're going to be using it more and more because we are inherently lazy. And everybody above us in every organization is going to be saying, “I want more for less.”
There was a great video, I’m not sure if you've seen it - I think it just came out. A mate of mine, Alex Hormozi, challenged his entire team to replace themselves using AI. It's a crazy full-on video, as Alex is a little crazy and full-on. But he just goes bang, bang, bang, bang, bang - through nine different tools of how his team are now using this, and he puts out 250 pieces of content a day using these tools.
David Bain 09:04
You're essentially saying that, to create the optimum quality content nowadays, you can still use AI as the initial version of the content. But you need that editorial view, that human to go in there and create an amended version to best compete with what else is out there and ensure that your content’s a little bit more differentiated.
Other tools out there that have used automation/AI for a number of years, include Google Ads - in that it used to be, five to ten years ago, that you would simply manually create your ad and manually select your keywords for targeting. But now, over the last few years, Google have encouraged you just to trust the machines.
Is trusting the machines for ad platforms like that the most efficient, optimized strategy nowadays?
Mike Rhodes 09:54
That's a very good question. It is the most efficient for them, right? I don't work for Google. We're an agency; we help businesses use those platforms. How Google have incentivized agencies like ours all over the world, is by - forcing is a hard word, but - kind of forcing us to play the game the way they want it played. So, they can see huge efficiencies in using all of these tools.
The way I used to describe it: I used to draw a pyramid, and let's put it into three layers. Across the bottom is bidding, in the middle we have targeting and, at the top, we have messaging. Those are the three main levels that you have - with any digital marketing, but certainly inside of Google Ads - and then the whole thing rests on data. And I used to say, “The robots are coming up from the bottom.”
Bidding? That's just a big math problem, right? machines can do that better and faster than us. Smart bidding has been the norm, I would say, for at least two years now. Give that one to the machines. Targeting used to be kind of, “Well, yeah, we can add things.” The machine knows a lot about everybody, but the machine has pretty much won that one. Five years ago, I stood on stage and said, “Messaging. That's the safe bit.” “Go learn persuasive copywriting”, I said - and got that one wrong, because we've just seen that the machine can write a pretty decent piece of persuasive copy.
It's rewriting ads, it's writing product descriptions, it’s writing entire essays, it's writing people's code, and Dr. Seuss poems, and song lyrics. It's the creative side - that's what surprised everybody - even though GPT-3 came out, three, two and a half, three years ago, and did much the same. It is better now, and the UI has made it really, really popular. But Google have seen the dollars, basically, in front of their eyes, I think, and gone, “Well, this is much better for us. How do we get everybody using this?” And so they've kind of forced the game.
I don't know how down in the weeds and tactical you want to go here, but they changed the ads from a block of ads (a couple of headlines, a block of text) to an ad that is made up from all of these little bits. The machine then decides which of those little bits to put together in the moment, when somebody is searching, based on everything that machine knows about that person. It's going to pull four or five of those little bits together and display that ad to the human. Why is it doing that? Because it can mix and match. There's no such thing as “the best ad” anymore. It's, “what's the best ad for that person, right now, in this particular moment in time based on everything we know about them?” So Google have forced us to use their AI in all of these different layers.
Now, right now, they still allow us to write those bits of ads, but for how much longer? They could easily incorporate their own version, which had a horrible demo last week (at the time of recording). But they could easily incorporate their version of a ChatGPT (Bard or whatever they're going to call it) and DeepMind’s Sparrow and put that in the interface where you write your ad - or just write the ad for us and say, “Click here to approve this ad. Let us get on with it.”
Google want to get to the point where: “Here's your credit card, and plug in your accounting system at the other end, and leave everything in the middle to us. We'll figure it out.” But, obviously, we're all a bit scared of doing that.
David Bain 13:24
So, just as you've been sharing that I've just asked ChatGPT to write a poem about marketing in the style of Dr. Zeus.
The first four lines:
Marketing, oh marketing, what a funny game,
With advertisements, promotions, it’s all the same,
From billboards to flyers to TV shows so bright,
Marketing is everywhere, day and night.
And it carries on. I mean, it's quite incredible.
Mike Rhodes 13:51
The thing that is so incredible to me, though, is you could put that exact same prompt in again - or I could put that exact same prompt in - and something different will come out. Because it's not following rules. It's a probabilistic, it's a statistical model. So it's running a whole bunch of scenarios in the background and trying to predict what the next word in that rhyme should be. And so, every time you run it (well, not every time, but most times you run it), you're gonna get something different to what you've gone before.
Yes, you talked about editorial before. The human still needs to step in and say “Well, actually, what's the type of thing that we do want to say? What's our point of view on this? What's our philosophy around this?” I read a fascinating piece - it was actually in the New Yorker, which is not a thing I normally read. It just makes me sound very posh. But I read this thing in the New Yorker over the weekend about a bug in a Xerox photocopying machine years and years ago that was photocopying things and changing things. And it was basically saying (I'll try and summarise it; I haven't said this out loud before) ChatGPT is kind of a blurry photocopy of the web. Yes, it can write a great first draft, but if you sit down to write not knowing what you're going to write, then the end piece probably isn't going to be very good. What a really good writer does is you sit down, and you go through those first drafts. You write a lot, and writing is editing, right? And you hone that piece, and you pull out - like David and the stone and the statue, you take away what's unnecessary and you leave what was there.
I'm not a great writer - I don't think like that in those writing terms - but it made sense to me that a great writer has this thing that they want to write and, in order to pull that out, they have to write all of this other stuff. And if we're teaching our kids: “Just stick a prompt in. There's my essay.” they're not going through the process of thinking about their writing. So, I'm conflicted now. I've been thinking of it as a great first draft assistant.
Businesses could use this for product descriptions at scale, writing bits of ad at scale, landing pages, etc. - I probably wouldn't use it for press releases yet, but blog posts. I use it for blog writing, I use it to get ideas around the topic I want to write about, and then to brainstorm titles, and then outlines and then maybe to write some paragraphs. I asked it the other day to give me a slide presentation for a 50-minute talk. I told it what the topic was, I told it who the audience was and sort of gave it some ideas. It maybe took me a minute or two to set that up, and out popped 50 lines: “Introduction, two slides, boom boom. Then talk about why, five slides, boom boom.” and it gave me the whole deck, basically, in a matter of seconds.
David Bain 17:30
Can you think of a brand or an industry sector that are actually using automation or AI effectively at the moment?
Mike Rhodes 17:40
I can't think of many that aren't, frankly. But we have a client that is in the apparel business, I will say. They specialize in apparel for plus-sized women. In Australia, at least, there are very, very few plus-size models, so it's very hard for them to book a shoot, it's expensive, there are long delays, etc. So they're experimenting with an AI tool where you load in a flat picture of the garment and then you can experiment. The machine will show that on various images of various people. Now, it’s too early to tell if that is going to work for them - they haven't put any of that live yet - but we're on this exponential curve, right? We, as linear beings, are not very good at thinking about exponentials, and there will come a point where we can't tell the difference.
I'm not sure if I mentioned it before, but you may have seen a website called thispersondoesnotexist. You hit refresh and you hit refresh and hit refresh, and that image looks exactly indistinguishable from a photograph of a person most of the time. Sometimes it goes a bit skewwhiff but, most of the time, it's amazing. So, to then do a full body shot of a person - once they get the fingers right, because most the time they get the fingers wrong in a lot of these images, but then there'll be an AI that specializes just in fixing the fingers on the AI created image. They will use that.
There's Robotic Process Automation that will watch what you do and how you do a task, and then learn just purely by watching. Businesses that collect data from the tasks the humans have done and said, “When presented with this, the human did that.” “When presented with this, the human did that.” If you collect enough of those, that's trivial to then feed that into a machine and say, “Okay, here's the new thing. What should you do?”
I don’t know if you ever saw a book called Principles by Ray Dalio. It came out a few years ago. First half was wonderful; second half, skip it. In the first half of the book, he talks about how, back in the 70s or 80s, what they did as a trading firm was to create a machine algorithm, a series of steps, for all of their trades. But the critical thing that they did differently from anybody else was to then run the two systems in parallel: the human and the machine. And every time those two spat out some kind of difference - whenever the human and the machine differed on what that trade should be - they compared them. Because maybe the human’s having a bad day, maybe they didn't follow their own rules - or maybe, actually, they were brilliant but there's something in that decision that they haven't yet built into the algorithm. So they would then either tweak the algorithm or tell the human to take the rest of the day off. They're having a bad day: “You go home, let the computer take this one.” They tweaked and tweaked and tweaked that algorithm for 40 years, and they became one of the largest hedge funds in the world.
If a business can start to think like that now: “How do we start to create an algorithm of the steps of something that we do, and then compare that to what the humans do? Then let's compare and tweak around the edges.” The algorithm may not be needed very soon. The machine will write its own algorithm, given enough data. That's what machine learning essentially is; the machine writes the code. Instead of us telling the machine what to do, you tell it how to recognize a cat, you give it 10,000 images of “cat” and 10,000 images of “not cat”, and it will figure out what a cat is and what a cat isn't most of the time - at least as well as a human can.
So if you can allow it to watch you work - give it access to your Gmail and it'll go look at all your past emails, it will learn the style that you tend to reply to emails in. Feed it a new email that you need to reply to and it will give you a pretty decent response. There are lots of AI tools like that popping up at the moment. Google have, for a while, been completing our sentences, these new tools will write the whole thing. Yeah, they won't be perfect, but they’ll be pretty good, and they'll keep getting better.
David Bain 21:59
Let’s move on, for the final question, from thinking about automation and AI just to thinking about marketing in general. So, what would you say is the number one thing that marketers need to incorporate into their strategy?
Mike Rhodes 22:13
I would love to see more – well, I think marketers get this. It's maybe the level above them that doesn't give this enough credence, and it's probably frustrating a lot of marketers everywhere right now. I don't think enough businesses really value data. One question I love asking C-suite is, “Where does data sit on your balance sheet? Do you really value data? Is it on the balance sheet?”
Because to be an AI company (this may have changed. I've been saying this line for five years, but maybe it changed a bit recently), you first need to be an analytics company. To be an analytics company, you have to value data - and most businesses don't. They maybe don't understand how to turn data into dollars so they just assume data is this geeky thing that lives off in the corner, and maybe somebody in the marketing department can just turn that Excel sheet into some insights for us. But that really should be at all levels of the business. I think marketing in particular is still one of those corners of the business that's mostly run on gut and not enough data.
David Bain 23:23
I've been your host ever been. You can find Mike Rhodes over at websavvy.com.au. Mike, thanks so much for being on The Strategic Marketing Show.
Mike Rhodes 23:31
Thank you, David.
David Bain 23:34
And thank you for listening. Here at IFP, our goal is simple: to connect you with the most relevant information, to help solve your business problems, all in one place. InsightsForProfessionals.com.
Access the latest business knowledge in Marketing
Join the conversation...