Don’t make it obvious you’re using AI to write
Subscriber Benefit
As a subscriber you can listen to articles at work, in the car, or while you work out. Subscribe Now
When ChatGPT and similar AI clients burst onto the scene, one of their capabilities that captured the public’s attention was the perceived ability to write. Ask ChatGPT to author an article for your company’s blog or a description of how jet engines work, and you’ll have several detailed paragraphs in a matter of seconds.
People often asked if I was worried these exciting new systems would put me out of a job. My reply has always been the same: not yet. Maybe five year from now, AI writing will reach the point at which it’s time for me to hang up the keyboard and take up a hobby. But as for today, the words these sophisticated systems generate don’t quite match human cognition and abilities.
It’s true that AI systems are fast writers – much faster than me even when the caffeine in my bloodstream makes it difficult to keep my fingers on the keys. The problem is that no matter how carefully you create the prompts that start the process, the result is copy bordering on robotic in its pacing and voice.
Clients will send me copy they claim to have developed on their own and are shocked when I tell them it’s obvious ChatGPT wrote it for them. Whether you ask for a blog post on the use of home equity credit or a full-length article on which SUV is right for you, what’s generated becomes remarkably similar. Most often, it sounds like a high school sophomore’s attempt to sound smarter than they really are.
These systems nearly always begin copy with trite phrases like “in this fast-changing economy.” What they write is structured like the short paper that sophomore is writing, with clearly defined paragraphs ticking off the subjects listed in an unseen outline. Nearly every piece ends with a conclusion that’s invariably titled “conclusion,” just in case the language isn’t obvious.
People who are much smarter than me attribute these traits to the fact that AI systems are incapable of true thought. AI grabs chunks of data from all sorts of sources and knits it together instantly in response to your prompt – sometimes well, sometimes not so much.
It’s extremely critical to remember that AI programs can’t differentiate between right and wrong. They sweep up language they’ve seen elsewhere that appears to be appropriate to the task at hand, and then they regurgitate it. Often, what they pick up along the way is inaccurate or inappropriate. A case in point is when my professional bio needed to be updated, and thought I’d ask ChatGPT to do the honors. The first couple paragraphs were pretty good, but after that, the AI started talking about work I don’t ever remember doing and awards I’ve never won.
Sometimes, I’ll use ChatGPT to crank out a rough first draft of an article or to help me craft an explanation of a complex concept. It always responds with that similar, semi-robotic approach … but often brings “facts” that aren’t accurate or even germane to the discussion. It automatically assigns equal weights to all the topics it wants to include (usually in the structure of paragraphs following heading tags or subheadings), even if those topics deserve more or less space and exposition.
For example, the second point I’m raising might warrant a couple paragraphs to make the information relevant and understandable, while the fourth point may require just a single sentence. The AI tool will try to make both points identical in length, shortchanging the subject matter of the second point while adding nonsense to stretch out the fourth.
Fortunately, there’s a trick you can use to dramatically improve whatever your favorite AI tool cranks out upon your command. It’s called editing and rewriting. Read it carefully for accuracy, but also to make sure it’s interesting, the style matches that of your company or organization, and factual errors haven’t crept into the copy. AI systems are fond of using clichés and adjectives like “thought-provoking,” so make sure you eliminate those overused phrases and words. Beware of subheadings and bulleted lists that sound like awkward keywords. Most important, eliminate the fluff AI loves to cram into copy.
There’s nothing wrong with using ChatGPT and similar tools to help you develop written materials, as long as you don’t turn around and use exactly what the machine generates. For all its wonders and potential, a lot of the writing coming from AI is pretty lifeless. Make sure yours isn’t a prime example.
Scott Flood creates effective copy for companies and other organizations. To learn more, contact him at sflood@sfwriting.com or 317-839-1739, and visit his blog at sfwriting.com/blog.