First things first: I’m pro-AI. I worked for an AI startup for a couple of years. I use AI a lot. AI helped me write this post. But using AI in communications work has unique risks, and those of us who communicate for a living – in words, pictures, or other media – need to be careful of them.
To see why, first think about other uses for AI – and how easy it is to know when the AI gets it wrong: A self-driving car crashes. An AI investment model loses money. A lawyer gets caught citing fake cases in court documents. The errors are obvious.
Let’s stick with the last example for a minute. A legal brief is much closer to communications work than a statistical model is. A brief is a piece of persuasive writing with some factual claims woven in. And it turns out that AI gets briefs wrong remarkably often: A study by the Stanford Institute for Human-Centered AI found that even bespoke AI tools for the legal profession, created by the likes of Lexis and Westlaw, produced hallucinations about one time out of six – and sometimes a lot more.
What, then, are the risks in using AI to create content – like marketing or communications content – that’s often more abstract and less falsifiable than legal writing? Content that’s designed to be inspiring, or funny, or project a certain brand identity?
There are obvious ways for marketing or communications content generated by AI to be wrong, of course, but there are several more subtle risks:
Let’s be honest: A lot of marketing and communications content – written by humans – is jargony and generic to begin with. It’s an empty box wrapped in beige paper. You’ve forgotten it in minutes. There’s a vast amount of that stuff out there, and that’s what large language models (LLMs) are getting trained on.
Here, I’ll prove it: I just asked ChatGPT to write a corporate purpose statement for a healthcare company. Here’s what I got: “Our purpose is to advance health and well-being by delivering trusted care, innovative solutions, and compassion—so that every life can thrive.” Is that so different from real purpose statements you’ve seen?
AI creates content that sounds plausible because AI is probabilistic: It looks for what’s most likely to be “right” based on what’s most common in its training data. To put it mildly, “most common” is not what you’re aiming for as a communicator. Yes, you can tweak the model – its “temperature” – to get a wider array of answers, but the principle is the same.
(Incidentally, this is one of the dangers of widespread use of AI at the macro level: a phenomenon called “model collapse,” in which AI is increasingly trained on content created by other AI in a grand race to the average.)
You have little to no chance of breaking through the clutter with content written this way. But you have a good chance of devaluing your brand by making it sound like everyone else. As George Tannenbaum says of the ad profession, “In advertising we'll see brilliant prompt engineers making millions and insipid work costing brands billions.”
This is the next level of damage created by “sounds OK” content. Everybody’s busy, right? We’ve all seen someone give something a quick read, say “sounds good” – meaning “nothing bad jumps out at me” – and move on. (They can also say, “run it by Legal,” but that’s a circle of hell for another post.)
If you’re creating bland content and people are approving it, you’re gradually teaching the organization that bland is acceptable. Before long, “sounds OK” will be what people aim for. After all, it’s what got approved last time.
I once took an all-day course on the philosophy of the great designers Charles and Ray Eames, who were pioneers in everything from architecture and furniture design (the Eames chair) to film. One of their principles was, “Never delegate understanding.”
It’s impossible to create good content about something you don’t understand. You can create something superficial, but you’ll have to gloss over what you don’t know – and everything you do know will be on the page. That lack of depth will kill you when a leader asks a follow-up question, or a teammate ask you to explain the topic you just wrote about. You will be a less knowledgeable, less valuable team member.
A new study from MIT proves the point when it comes to AI. Researchers asked groups of students to write essays with or without the help of a large language model like ChatGPT. They found that, “LLM users struggled to accurately quote their own work,” and “Over four months, LLM users consistently underperformed at neural, linguistic, and behavioral levels.” To put it in simpler terms, they displayed less brain activity. Less thinking.
One of the major challenges already facing communications teams is being seen as tactical content creators instead of strategic partners to senior leadership.
Overreliance on AI will make this problem worse.
Consider the cumulative effect of the points above: If communications professionals create lots of “OK” content, the organization will start to see churning out boilerplate as what Communications actually does. Leaders will stop expecting any higher-level thinking, and the challenges to Comms getting a seat at the table in strategic discussions will become much worse.
You already know the obvious points: Use AI as one tool among many – as a muse, a thought-starter, a monitor of the conventional wisdom you may want to challenge. Learn the techniques of prompt engineering. Fact-check what the AI tells you. Think for yourself.
The deeper answer is that your most important task as a creator – or really, a valuable team member of any kind – is to learn. Dig deep. Be curious. Then start creating – and you’ll quickly realize there are still things you don’t understand. Then go learn those.
The result of true learning is a deep understanding of context. You don’t just understand what it says in the deck; you understand the topic, and how it fits into the big picture of your organization and your industry. You can make connections. The architect and designer Eliel Saarinen famously said, "Always design a thing by considering it in its next larger context – a chair in a room, a room in a house, a house in an environment, an environment in a city plan.”
The more you learn – the less you delegate understanding – the better your content will be, and the more valuable you and your Communications team will be you your organization.
