A Human Touch Helps Generative AI Hit the Mark
It’s a rookie writer’s misstep my wife and I still joke about 23 years later. Back then, we were cub reporters at the Sandusky Register newspaper in Ohio, assigned to cover the weekend Milan Melon Festival. I filed my story on Saturday, and she filed hers Sunday.
As prideful young writers, even a local street fair was going to get our best artistic write-up. So, we were more than a little embarrassed to discover we’d ended up writing the same lead to our different stories, each of which made the front page: “Rain didn’t dampen any spirits.” Some artistes we were, eh? Turning out identically lame clichés!
Fast-forward a couple of decades, and these grizzled writers have taken turns leading creative teams, overseeing editorial efforts, and serving as content coaches and safety nets for new waves of writers and their unwittingly problematic prose. In the digital world, that puts us on the front lines of unleashing a creative force seemingly more disruptive than any rookie reporters out to memorialize deep-fried Twinkies and festival queens.
Generative artificial intelligence (AI) tools like ChatGPT and Stable Diffusion can whip up written and visual content on command. But the potential pitfalls are many, including inaccuracies, outright falsehoods, bias, and copyright infringement—not to mention a watered-down output that, while it may save time, does not exactly make for memorable or profitable content.
In a recent webinar, some of our TPT Digital team members working most closely with generative AI shared what they’ve learned so far—and how keeping a human hand in AI workflows has been key to improving output and carrying us closer to a more productive future.
Generative AI: What Is It? How Does It Work?
Fred Bane, Senior Manager of Data Science, admitted laughing at the results of AI-generated translations only a few years ago. As a trained linguist, he took pride in not only being able to say something in another language that conveyed the meaning of a source text but understanding enough about the subject and its context to deliver something ideally as nuanced as the original. Machine translation would comically miss that mark.
Until it didn’t.
Early AI models had the disadvantage of being new to the content they were emulating. By now, their training gains have been exponential—not just absorbing reams and reams of data but racking up repeated applications of the patterns they’ve learned into specific assignments and deliverables.
The data models that have been fed into AI solutions, like ChatGPT, are so vast that a mere prompt can get you other qualifier outputs. So, the focus of data scientists like Bane is on refining instructions to hone what AI delivers.
For example, a machine translation tool can be told to translate “cheese” from English to French. Adding examples of what you want to see come back, including coaching on tone, length, and style, will shape the response more beneficially.
Because generative AI can create content in a target language—without even the need for a source—Bane recognizes its evolving power to replace traditional translation workflows with native content creation. One drawback? AI does tend to make things up.
In one case, a popular home furnishings retailer needed to add product descriptions to its website pages. Fed information on the product catalog, ChatGPT dutifully came up with paragraphs about the items—including many attributes they did not, in fact, possess. But AI can’t think for you, Bane points out—it can often be just as much work honing clear instructions as producing copy yourself. You can’t achieve efficient, scalable AI until you’ve taken the time to run experiments: What works for certain applications? What doesn’t? And, how can the kinks be trained out of the process?
Smart Approaches to AI Content for Digital Marketing
Before recommending AI as a tool for clients, it’s important to understand a client’s goals for a project and weigh where generative AI can help, said Henry Barfoot-Saunt, Executive Director at TPT Digital.
Not all content is created for the same purpose. When quality is paramount, or strategy needs to lead the way, human copywriters might be the best bet. If it's quantity a client is after, and the text demands a workmanlike approach, AI—if properly trained and managed—could present an advantage. Breakthroughs in keyword research and suggested optimization could make AI an ideal SEO partner, but the final edit is still likely to be carried out by your flesh-and-blood specialists.
Barfoot-Saunt and Digital Account Manager Paul Carlier have put AI to the test on a pair of recent projects. In the first, creating blog content in Chinese for Taiwan, they relied on human expertise for keyword strategy and content optimization, but AI did the copywriting. In the second, AI generated content for product description pages in e-commerce while humans fed in the prompts and edited the output.
“We had to establish new workflows and comprehend the overall capabilities of the tool,” Carlier said. “We made adjustments on the fly, and the experience has been really exciting, especially when we discovered the right approach to achieve our goals.”
AI struggled, at least for now, with optimizing content with keywords and crafting meta content within strict character limits, Carlier noted. But part of the trial and error was seeing whether human editing time could be reduced to an acceptable level to make the AI effort worth it.
More and more clients are interested in raising the stakes—not just in text generation but pursuing AI-assisted projects for images and video. During our webinar, audience surveys revealed 60% of attendees were already actively testing or publishing AI-generated content while just 19% had yet to get started. Of the most popular industries testing AI capabilities, retail led the way at 43%, followed by media and entertainment, and travel and hospitality. Regulated industries like life sciences have been slower to adopt—so far.
The push for more content—and quality content—has Barfoot-Saunt convinced that AI will generate more jobs for digital teams, not fewer. Experienced writers now become the editors that are learning from, and teaching, the AI engines.
Carlier, a confessed early adopter, embraces that change.
“We’re still in the early stage with this technology, and its implementation will…(grant) us the opportunity to dedicate our focus to other areas that demand strategic thinking,” Carlier said. “I cannot predict the future, but I know we can all at least continue to follow the developments of these tools and keep experimenting.”
To learn more how you can better utilize digital marketing, visit www.tptdigital.com