Read this article in our brand new Speed Reading Mode
AI is a fast-evolving and complicated phenomenon. Businesses want to know if AI tools like ChatGPT are worth their attention or are just a passing fad.
Many of you will have experimented with an AI tool in recent months. We’ve been putting them to the test, too. Today, we explore the results of this experimentation and our opinion about AI, specifically the applications of AI for marketing.
Back in the mid-sixties, Joseph Weizenbaum wrote a chatbot called Eliza. It simulated in a very crude way the conversation that you might have with a psychotherapist. He was alarmed by the fact that one of his colleagues asked him to leave the room so she could have a private conversation with the computerised therapist. To our eyes now it isn’t particularly convincing, but for the time, it was quite the cultural phenomenon (some of you may remember!).
Then, in the mid-seventies, there was Zork — a text adventure game where you could type in your answers and tell it what to do. Again, not terribly convincing, but still quite an exciting use of AI or a “simulated AI”, for the time.
It was written by a program called Racter that could write poetry. It’s not great poetry, but it’s not terrible either.
‘More than iron, more than lead, more than gold, I need electricity. I need it more than I need lamb or pork or lettuce or cucumber. I need it for my dreams.’
OpenAI’s ChatGPT is only a little better at putting together rhymes, 40 years later (in our opinion).
Today and tomorrow
Let’s fast-forward forty years to today. Now, there's a huge outpouring of hype about ChatGPT and generative large language models. The human ingenuity and human ability and human desire for computers to do this sort of thing slightly blinds us to the reality of what it actually is.
Colin Powell said, ‘Never believe the first thing you hear.’ What we have are stochastic parrots that use lots of data and machine learning to do interesting things. Not super-intelligent AI machines pandering to our every need, moving society towards abundance beyond scarcity, Ian M. Banks-style.
In reality, a lot of AI hype is marketing and it’s designed to drive or defend large companies’ share prices. So everyone's building similar systems. Microsoft’s CoPilot, Google’s Bard, Meta’s Llama. They want to be perceived as being at the cutting edge of technological innovation. They want a slice of the pie.
Really, it's too early to say what the full implications and impact are going to be. Chairman Mao met Henry Kissinger, who asked, ‘What do you think the impact of the French Revolution is?’ And Mao said, ‘It's far too early to tell.’ We're still living through the impact of the internet revolution and that's 30-something years in. Most technology change is overestimated in the short term and underestimated in the long term. This is also true for AI.
Nicholas Carr, a man who argues that Google is making us stupid says, ‘IT doesn't matter.’ If everybody has personal computers (that’s a big “if”. ‘The future is already here - it’s just unevenly distributed.’ — William Gibson), then having personal computers makes no difference to productivity, strategy or competitive advantage. Of course, you are at a disadvantage if you don't use IT. Excel has firmly defeated the abacus and pencil. So, some of these AI tools will become table stakes. You'll need to be using them to be competitive, but by themselves they don't give you a competitive advantage.
That all said. Let’s talk about AI in the context of marketing.
What has changed in the last year?
First off, ChatGPT-4 (specifically) is useful. HubSpot did research on 1,300 marketers and many are already doing research, generating ideas and creating content with ChatGPT regularly. More on this later.
The second thing that has changed is it is being built into lots and lots of tools. It’s ubiquitous. You’ll find it popping up in all sorts of marketing and project management apps and tools like Notion, ClickUp, Grammarly, chatbots. Every business is an AI business. Every app is an AI app.
The third thing is the price has changed. Cheap has been an option for years, for marketing. You can buy dozens of blog posts a month on Fiverr or Copify or Guru. There are risks to your reputation and your SEO, but the option has always been there. That’s not what’s changed. What’s changed is, with AI, you can get those dozens of blogs for pennies, not pounds. That’s quite a differential. But again, beware. Hazards. Warning.
Theoretical risks of AI
These are risks in the category of “interesting, but you can't do much about it”. The more speculative risks that people like to talk about, but for which there is little practical, immediate resolution.
1. AI is amoral
‘ChatGPT and its brethren are constitutionally unable to balance creativity with constraint. They either overgenerate (producing both truths and falsehoods, endorsing ethical and unethical decisions alike) or undergenerate (exhibiting noncommitment to any decisions and indifference to consequences).’
Of course, while you can say ChatGPT isn't that creative, you can also say the same of a lot of marketers and marketing agencies… Basically, everyone has an opinion here and it is a bit more of a tangible subject to grapple with in a sense, so we’ll talk about this more in the next few sections.
4. AI will cause the Job apocalypse
‘Eighty percent of the U.S. workforce would have its work-related tasks at least 10 percent affected by language models. One in five would see at least half of daily tasks affected by artificial intelligence.’
There are some genuine things about AI and ChatGPT that you need to be conscious of as marketers. This is stuff that concerns you, now, and that you should be mindful of.
1. AI is out of date
By its own acknowledgement, ChatGPT has ‘limited knowledge of the world and events after 2021’ because that's when it did its data processing.
2. AI is insecure (not like that)
This is directly from their website: ‘Conversations may be reviewed by our AI trainers’. If you are worried about confidentiality, privacy and GDPR compliance, this is a serious, legitimate risk.
3. AI hallucinates
If you are relying on it for research, the fact that it hallucinates is a serious problem. It will make up statistics, sources and quotes that sound real, but don’t exist. Always check your sources. Test this by asking ChatGPT who you are. (I asked ‘Who is Matthew Stibbe?’ The bio I got back was riddled with errors.)
4. AI censors itself
ChatGPT 3.5 got in a lot of trouble for not censoring, so this version is very much PG. This is a concern in that, if it censors itself in some regards, what else is it not telling you? What biases could be impacting that filtering?
5. AI risks your reputation
There are four angles to look at this from:
Watermarking: This is when the content you’ve positioned as your own thought leadership is discovered to have been written by ChatGPT. That makes your position as an ‘expert’ null and void.
Errors: This is when ChatGPT cites something false (a hallucination) and you take it as accurate. There was a case of two lawyers who submitted a brief to a court citing cases in their defence, all of which were completely made up by ChatGPT. It backfired, massively.
Plagiarism: Stealing is wrong, obviously. But what happens if the AI was trained using copyrighted material? There is a risk you could be done for plagiarism.
Inauthenticity: Imagine you write a personal letter as a CEO to a business partner or client, and they find out it was just ChatGPT talking. That’s very damaging to your relationship.
6. AI adds to the noise, but doesn’t cut through it
If your marketing strategy relies on high volumes of cheap, low-quality content, you're just going to face a lot more competition. And, put like that, it doesn’t sound like a great strategy anyway? Not to mention the fact that Google is optimising its algorithms to detect fake content. You need to position yourself as a genuine thought leader to differentiate from competitors.
7. AI is (too) easy to rely on
Think of this like muscle atrophy. If you rely on AI to do the thinking for you, you lose the skills that are actually important in life, and for marketing, like source criticism and logical thinking.
Applications of AI for marketing
With all those risks out of the way, let’s look at the opportunities. These are ways you can use AI for marketing that we think are worth trying out.
1. Less drafting, more polishing
Norman Nielsen group found that in a business context, professionals who used AI spent less time on the drafting of text and more time editing. So they spent more time on things that were producing better quality work and less time on drudge work. You can spend the same amount of time working on something, but devote more of that time to high-value, high-payoff activities like research, structuring, editing and refining.
2. Streamline writing chores
There are some writing tasks that are prosaic and time-consuming. Summaries. Meta descriptions. Social media posts (to some extent). Questionnaires. You can tweak and edit the copy a bit, but having AI on hand is helpful to tackle the bulk of this work.
3. Adjust your tone of voice
Use ChatGPT and AI tools it to adjust the tone of copy. Emails, particularly. If you want to ensure your email is on brand and isn’t coming across too blunt or something, AI can help you achieve that. Always review the email before you send it, though!
4. Gain a creative sidekick
We asked for five alliterative words that explain why marketers might want to avoid using ChatGPT. It came up with impersonal, inaccurate, inconsistent, insensitive and inflexible. Great. Now we might not necessarily use the rest of the copy it suggested, but all writers love alliteration and AI will save you agonising over the alphabet.
Plus, you can use it for things like connector sentences to bridge paragraphs, or to suggest interesting word choices for ledes or kickers. it has a huge vocabulary, so use it for that.
5. Take minutes and action items
With Microsoft Teams it’s easy to get a full transcript of a meeting. But that can be a huge amount of text to sift through. AI, then, can extract from a transcript an accurate summary of the points and action items, in seconds.
What humans can do (that AI can’t)
Let’s look at the other side of the coin, which is what AI can’t do. O, at least, where it takes more of a backseat.
Similarly, if you want to do thought leadership, you need to be different. You need to sound different. You need some personality, some opinions, some genuine experience, some originality, some fizz and ginger. AI cannot be you. Thought leadership is an area where good quality marketing and copywriting has a really important role to play.
Storytelling and suspense
There is a craft to storytelling in non-fiction writing and journalism. Check out ‘Writing to Deadline’ by Donald Murray. ‘Control of suspense’ is a phrase that Alistair Cooke, the famous journalist and broadcaster, used. These things are at the heart of good writing and AI doesn’t do them well.
The New Yorker is not going to use ChatGPT to write its articles. This is why. Anything that requires deeply understanding a client or your business, products or customers beyond the most superficial or generic level needs the human touch. AI can’t do original research and it can’t interview people.
Going beyond the prosaic
AI writing is like McDonalds. Looks good in the ad; tastes like cardboard. It’s a bit dull. Bland. It might pass the Turing test, but it's not Alan Turing. Any writing that requires something more than the prosaic, such as product naming, taglines, high-impact website copy, brand-specific copy, headlines… they are going to need human imagination.
AI as your personal autopilot
Microsoft talks about co-pilots. Reid Hoffman, who was the founder of LinkedIn, wrote a book with AI and he thinks of it as an undergraduate research assistant or as a reasonably smart intern. Steve Jobs said ‘computers are like bicycles of the mind’ — an accelerant for human power. And Doug Engelbart talks about intelligence augmentation (IA) rather than artificial intelligence (AI). We like that idea.
To add our own metaphor to the mix, we think of AI as an autopilot. This can fly the plane and do some pretty clever stuff. But it can’t choose where to go. It can’t use its own judgement and it doesn’t have the level of expertise, experience and judgement as the pilot. But it makes flying safer because it frees up the pilot’s bandwidth. British Airways won't let you land in bad weather without using the autopilot, in fact.
It’s up to us to embrace this technology and use it in a safe, clear way where there are clearly understood parameters and rules of engagement and if it doesn't work, it doesn't put you in an unsafe position. So, to finish, there are three ways of implementing AI in your business.
Optional: Use it if you want. In our agency, everyone can use it if they want, understanding the risks and best practices we’ve come to as a business. This is the default position for us.
Required: There are some cases where it may be required. Sending high-volume personalised sales emails is an example we’ve seen recently with a client. You couldn’t do it without AI.
Forbidden: Apple has forbidden AI use in the company for privacy and confidentiality reasons. There may be use-cases in your business where it is forbidden, too.
Agree the rules of the road, explore, adjust, don’t overreact, evolve as the technology evolves.