Around the turn of the 21st century, Web 1.0 came online. (Well, not the 1.0 part since no one knew it would have versions.)
Everything centered on the “net.” You logged onto the net using the Netscape browser. You made calls with your net phone. Microsoft created dot-net software. ZDNet and CNet published tech news. Digital music was called “net music.” Sandra Bullock starred in The Net, a movie about hackers.
The net was everywhere in the late 1990s and early 2000s.
In 2024, the net doesn’t get many mentions. Today’s conversations center on prompts — prompts for generative AI. “You need to get better at prompts.” “You need prompt engineering.” “You need to be the best prompter in all of prompting to get good at creating content.”
But you know what? Knowing how to create prompts for generative AI is akin to knowing how to create HTML codes in the early days of the net.
That’s what CMI’s chief strategy advisor Robert Rose says and what prompted him to give his take on the topic. Watch the video or read on for his thoughts:
AI-generated content has gained traction in marketing content these days. I see more and more images and text clearly generated by AI. The images are easier to spot because the AI generators create a certain look and appear frequently on Facebook and LinkedIn.
Play “Where’s Waldo?” with them, and you can usually see relatively odd components. See what I mean in this image accompanying a blog post about teams working together to edit a document.
Seventeen people sit and stand in an office-looking room with big windows. Eight of them sit at a big table with two open laptops, marker-filled cups, a tablet, coffee mugs, books, and papers. Most of the others look at the table. Nearby sits a flip chart, a few more desks, and posters on the painted brick walls as sunlight streams through the windows.
I have so many questions:
Why do all the men have beards?
Is that a picture of a raccoon on the back wall?
What does that poster next to the raccoon say?
Is that a small person with no legs sitting on top of that table in the back?
What is wrong with the woman in the front right?
Does the woman on the left have a computer mouse surgically attached to her left forefinger?
But I digress.
What about AI-generated text? It’s replete with adjectives and flowery language. When I prompted ChatGPT to write about content marketing and open with a story to establish context, it created:
“Once upon a time, in the bustling boardrooms of a Fortune 1000 company, there was a vice president of marketing, Alex, who faced a daunting challenge. Sales were stagnating, and traditional marketing methods were losing their luster.”
Yes, ChatGPT does love alliteration. In this case, it used three in two sentences.
In any event, many would suggest the problem with AI images and text lies in the prompts provided to the generative AI tool.
A hot topic among marketers, prompts have become a commodity as thought leaders sell or give away their best classic Mad Libs fill-in-the-topic format. Some suggest you tell the generative AI who it should be, such as “Pretend you’re a librarian” or “Pretend you’re a world-class aviator coming out of the Top Gun school in San Diego.”
(I tried that last one and asked for the No. 1 piece of advice. That prompt earned this response: “Embrace the challenge.” Whew, that’s good stuff and no threat to Tom Cruise.)
At the Content Marketing Institute and The Content Advisory, we have dived deep into the world of generative AI, testing the tools and understanding the strengths and weaknesses of the learning models to generate content.
My conclusion: Creating the best value from generative AI has NO relevance to being an expert at prompting.
That’s not to say that learning to ask better questions — what you really want to know — won’t get better and more valuable answers. That’s possible whether you ask AI or humans.
But aside from asking for specific looks for images and providing words to help the AI understand jargon, prompting diminishes returns beyond the most basic levels. If generating good AI output is relegated to only those who can “code” good prompts, then it’s not the disruptive technology everybody believes it to be.
The Dunning-Kruger effect also emerges to create a problem with AI-generated content. People overestimate their ability or knowledge. But knowing how good you are at something requires the same skills as being good at it in the first place. With AI-generated content, people think it’s higher quality because they don’t understand what “good” looks like.
I know a company that recently replaced its customer content-enablement team with generative AI and a freelancer. After they prompted the generative AI, it created scores of new customer-enablement content the freelancer published to the website. There was only one problem. The content gave wrong information about the products, technology, and how things worked. The freelancer couldn’t discern the content’s accuracy. While the content was impressively written, it was just plain wrong.
As you consider integrating generative AI into your content strategy, remember my prompt: Do not become prompting experts. Drive hard to become experts in your subject matter and create awesome human content with an AI assist originated from simple, easy-to-remember prompting.
HANDPICKED RELATED CONTENT:
Cover image by Joseph Kalinowski/Content Marketing Institute
Get the week’s best marketing content We spent ~$80 to purchase five premium ChatGPT prompts and ran a blind test among the members of our marketing team to see if they were worth it. Long story short: They aren’t. The experiment Here’s what I did: I signed up for a service selling ChatGPT prompts and selected five...