If you’ve been anywhere near the internet for the last half year or so, it’s a safe bet you’ve probably heard about how generative AI – visual image generators like Midjourney or language generators like ChatGPT, just to name two of the more prominent examples – is going to revolutionize everything.
The internet is going to dramatically change, how we search for things is going to dramatically change, how we find data is going to dramatically change, and so on – those are the promises being made about these technologies.
But how true are those claims? For marketers, specifically, what will be different? Should we throw out 20 years (and then some) of what we know about SEO on the spot, and tailor all of our content to AIO (that is, AI Optimization)?
Well, no. Kind of yes, but also no.
ChatGPT and its AI kin, like Google’s Bard, are promising new technologies that will certainly have a place in the toolbox of marketers of tomorrow. But thus far, the idea that marketers have to forget everything we know to placate these AI tools isn’t quite passing the test.
So, let’s talk about ChatGPT and marketing – and in a broader sense, talk about AI and marketing.
In this blog, we’ll discuss these topics:
- What are tools like ChatGPT?
- What are their strengths? What are their limitations?
- How should marketers plan to change how they work?
- Seriously, is AIO going to be a thing?
What Is ChatGPT, and How Does It Relate to Marketing?
ChatGPT – short for “chat generative pre-trained transformer” is the consumer-focused version of OpenAI’s AI chat bot. ChatGPT, like Google Bard, is an LLM, or “Large Language Model,” which is literally trained with billions of parameters to sift through. The fact that these models have almost the entirety of humanity’s written language as a base is one of the things that makes them so realistic when we talk to them.
But what does this actually mean? When we talk about these LLMs, what are they actually doing behind the scenes? Are they thinking for themselves, like R2-D2 or Skynet?
The answer is “no, not at all,” but they’re very good at making humans think they are.
It may help to think of LLMs like ChatGPT and Bard as essentially the world’s most advanced text autocomplete. You know how your phone’s predictive text will learn that you usually follow “those goddamn noisy” with the word “neighbors”? That’s what’s going on here, just on the scale of “all of humanity’s written work.”
Naturally, people have suggested that tools like ChatGPT and Bard can start phasing out human writers almost overnight. It’s no wonder that in the (current, as of this writing) Writer’s Guild of America strike, people have suggested that studios turn to ChatGPT to pump out scripts instead of human writers. Do you need a blog written? Use ChatGPT. Research text? ChatGPT. Optimize the SEO on your website? Yep, ChatGPT.
Can ChatGPT do these things, technically? Sure. Can ChatGPT do these things well? Ah, there’s the pertinent question, isn’t it – and one we’ll visit in the next section.
However, there’s a much more immediate use of LLM tools like ChatGPT for marketers, and it’s in the question-and-answer stage. After all, ChatGPT doesn’t need to be creative to answer questions, right?
This, in particular, is how Microsoft has been pushing AI – it spent $10 billion to invest in ChatGPT’s creator for a reason. Microsoft is working to integrate OpenAI technology into its Bing search engine, and while the results can be a little horrifying at times (Bing being a little too realistic in its emotional responses), it’s very clear that Redmond thinks that AI will be the future of how people find information in the future.
Are they right?
What Can AI for Marketers Do? (And What Can’t It)
Let me start off with a caveat: Generative AI is still very early on, relatively, in its life span. It is entirely possible that everything I am about to say about the relative limitations of things like ChatGPT will change in the next decade or so, and I will look like those very silly people who thought the internet was going to be a passing fad. Oops. Certainly, Microsoft isn’t wasting $10 billion for no good reason.
I’ve been playing around with ChatGPT now for work and for play, and while it’s certainly a powerful tool, I’ve often been struck by what it can’t do as much as I am what it can.
ChatGPT, in many ways, feels like a product that’s almost there – close enough to be usable for some things, but dramatically far away in others. Let’s call this the 90/10 principle: The amount of effort to get something to 90% usability is dramatically less than the amount of effort to get something the remaining 10%.
I was recently in the market for an electric lawn mower. I asked ChatGPT for some recommendations, and it helpfully provided some brands I should consider. Fair enough, that’s five brands to look into.
But when you read what it’s actually saying, have I learned anything useful or actionable? No, I haven’t. I have these five brands, which ChatGPT helpfully recommends I “read reviews and compare features,” but for all of them, its lets me know that they have “different cutting widths” and other features.
ChatGPT is very reluctant to make value judgments or recommendations beyond a certain point – and it should be. Again, it isn’t actually thinking. It is scraping data off the internet and synthesizing it, and then presenting it in a manner that makes it sound like a real person offering their recommendations.
But even when you try to tweak it into giving something resembling a recommendation, it might meet you halfway, but it insists that you do the rest of the work.
Now, that’s not a bad thing. The last thing we want is a humanity that becomes accustomed to just doing as the AI tells us to do (that’s how you get robot revolutions). But it is striking to see the limitations here – because these are the exact same five brands I got when I simply used Google to search “best electric lawn mowers.”
(Incidentally, I got the Ryobi, due to the fact that the battery is shared between all of the brand’s power tools, which is not a feature or benefit that came up in ChatGPT’s analysis.)
However, buying power tools like a lawn mower is an expensive venture. You should expect to do your own research before you pull the trigger on this. How about something a little more commonplace, like going out to eat?
I live near the Mississippi neighborhood in Portland, Oregon, known for its great food, drink, and nightlife. I asked ChatGPT where my partner and I should go for dinner, and – to be fair – its first five choices were excellent.
All five of these restaurants do exist, and they are all in the Mississippi neighborhood. There’s a very slight miss here in that Gravy is strictly a breakfast and brunch spot and closes at 2 in the afternoon, but otherwise, this is quite solid.
However, when I expand the list, things start to break down a little.
This is a much less successful round of suggestions. Stormbreaker Brewing and Prost are spot-on – if you like beer in Portland, they’re two fantastic places to grab a pint and dinner – but the others are more questionable. Tamale Boy is not far from Mississippi, but it isn’t on it either (about a 15 minute walk). Miss Zumstein is a good 15 minute drive away from us. And the Wayback is permanently closed, as Google Maps tells me.
I know for a fact that the Mississippi neighborhood has many, many more great restaurants that aren’t on this list. So why would it include ones that A) aren’t on the list and B) aren’t even open anymore?
In the end, if I didn’t already know about the restaurants it was recommending me, I would have had to Google them all anyway to make sure that they were near me and, you know, open as an option for some dinner. Is ChatGPT simplifying things, or adding an extra step?
Reason and Rhyme
Other than doing research for electric mowers or finding where to eat, I’ve been using ChatGPT quite a bit for inspiration in my nerdiest of hobbies, playing Dungeons & Dragons. I’ve spent the last two years (and change) building my own fantasy setting to play D&D with my friends, and ChatGPT has been a marvel at saving me time in creating towns, characters, and the like. “Come up with names of rulers of the central empire over the last 300 years” is something that these programs can do in a snap.
But even then, it always requires a human touch. The list of rulers ChatGPT gave me alternated between Kings and Queens on a 1-to-1 basis. And no matter how many times I tried to tell it to reorder the list, it still gave me a list that alternated between Kings and Queens.
One of my more mystifying attempts was when I asked it to write what, in-universe, was a poem written by ancient Elves. I wanted it to sound mystical and perhaps a little stilted (due to the translation), so I repeatedly told ChatGPT that I didn’t want the poem to rhyme.
And every single time, whatever poem ChatGPT produced wound up rhyming.
Again, these are little things. As a Game Master, the simple fact that this tool could save me hours of trying to write an ancient Elvish poem from scratch, and instead just have me spend minutes cleaning it up, is incredibly impressive.
So what does that mean for marketers?
Should Marketers Use ChatGPT?
Maybe it’s just me, but I feel like ChatGPT has a “tell.” It might be the voice, it might be the quippiness or the certain types of analogies it likes to use, but I think it’s not hard to tell when a piece of content is written by ChatGPT.
And if I can tell, surely the great algorithms at search engines like Google can tell, right?
Well, maybe. Maybe not.
But given that search engines already penalize people for plagiarized content – as they should – it’s hard to think that AI-generated content won’t, someday, be considered similarly.
That’s not to say that if you want to feed blog prompt after blog prompt into ChatGPT and flood your site with AI-generated content, you shouldn’t be allowed to. If you want to, go nuts. Certainly, given that Google and Microsoft are both investing real money into these systems, it would seem silly to think that there will be an abrupt about-face anytime soon.
But here’s the thing: As we mentioned earlier, LLMs like ChatGPT and Bard don’t actually think. They’re incredibly, wonderfully, advanced text predictors. A world where all content is written by AI is a world that ceases to advance, because the only thing we’re ever doing is pulling text from somewhere else and repeating that as gospel.
In academia, for instance, ChatGPT has developed an infamous habit of citing research papers that don’t actually exist. This Twitter thread dives more deeply into why it does this, and it seems to rely on its text predictive nature – when asked to cite a popular economic paper, its research gets snarled. This is by no means the only example of it doing this.
Again, it is possible that all of these errors might get ironed out within the next couple of years, and ChatGPT blogs and websites embark us on a new age of instantly-generated content for business and personal use.
But I don’t think we’re there yet. And I don’t know if we ever will be. “Invent a computer that thinks and creates like a human being” is a long way off.
So what can ChatGPT and other AI tools actually do for marketers?
How Can ChatGPT Help Marketers?
Well, if you’ve noticed, all of the examples in this post have had one big thing in common: They’ve all required a human to help them get across the finish line. Whether it’s researching electric lawn mowers or changing a list of fictional rulers so that they don’t always just alternate between genders, ChatGPT has gotten the ball rolling – and it’s been up to me to continue it.
There, I think, is ChatGPT’s greatest strength for marketers and, indeed, anyone who wants to use it: It’s excellent at generating ideas and inspiration, even if it isn’t the best, yet, at following through.
So using ChatGPT to create an outline or some blog topics that a human then fleshes out might wind up being the best of both worlds – you get to use real human creativity in the process, but you also save a lot of time on the backend. It’s also handy for content that might not ever see the light of day on Google, like email campaigns.
So, About That AI Optimization (AIO)…
Is AIO going to be a thing? When will it be a thing? How will it differ from current SEO best practices?
The answer to that is: Hell if I know!
In all seriousness, consider the questions before – the research into electric mowers, looking for a place to eat. All of the responses were pulling from exactly what you would find if you searched these things on Google or Bing without using fancy AI.
At the end of the day, AI tools are scraping the web the same as the normal search engines are. Therefore, is AIO meaningfully different from SEO in terms of how you create your site? No, probably not. The same SEO best practices that you’ve been using this entire time – meta descriptions, alt image text, proper use of keywords in headers, and so on – are likely to be AIO best practices.
It is possible that due to the more conversational nature of tools like ChatGPT, you will want to make sure that you are answering questions in a conversational format that ChatGPT and other AI tools for marketers can easily parse and repeat. Similarly, since ChatGPT mentions multiple times that it uses review sites like Yelp, it is all the more important to be sure you’re getting good customer reviews on major sites as often as possible.
ChatGPT won’t be able to do everything marketers (or screenwriters, or creators of fantasy worlds) can do, nor should we want it to – because that represents the death of human progress. But it is a very fascinating, very helpful tool at generating a starting block, and that’s where I think ChatGPT and marketing will most closely align.
But for the time being, you’ll probably be stuck going over the damn Elvish poem to make sure it doesn’t rhyme.