As an agency that specializes in content marketing, we are often asked if we use artificial intelligence (AI) in our copywriting production process. Our response is unabashedly unequivocal: “Absolutely!”
Many of our colleagues in the marketing industry are still hesitant about admitting this, even in the face of overwhelming external and internal pressure. They worry that admitting to using AI tools would somehow compromise them as marketers, given that marketing trends for the last decade have consistently emphasized “authenticity.” How can you authentically use AI? Isn’t that “lazy” and “easy”?
This existential struggle ensnares not only marketing agencies, owners, and copywriters but also marketing designers, strategists, and account managers. Besides the anxiety of being labeled inauthentic, there is also the very real fear of layoffs and job losses for talented marketers who may be overlooked for this shiny new technology that seems bigger, stronger, and faster.
But bigger, stronger, and faster doesn’t always equal better. From our very first usage of AI chatbots in 2022, it was clear they had a long way to go. While they’ve gotten much better since then—and will undoubtedly continue to improve—it seems many of the initial fears were unfounded or, at the very least, overblown.
This isn’t to say that AI chatbots are not useful. They’re incredibly useful. In many ways, they’re no different from many other marketing tools—tools that many marketers continue to use for automation, illustration, project management, and more.
But every tool has its limitations, even AI. And this is where the true artistry of human innovation comes into play.
AI Chatbots: A Clarification of the Term
Before we dissect the proper usage of AI in copywriting, we should first make sure we’re working from the same definitions. When we refer to AI today, we mostly talk about chatbots like the first-to-market ChatGPT and ChatGPT’s main competitor, Claude.ai. There is also the ever-evolving Google Gemini and, of course, Microsoft Copilot, which is frantically playing catch-up to its peers.
Recently, social media platforms have unveiled their own versions of AI chatbots, most notably X’s Grok and Meta’s Meta.ai. While these platforms are still in their infancy (relatively speaking), their infancy is much more advanced than the earliest days of the earliest chatbots.
Outside of these natural language generators, other AI tools help marketers with copywriting. One of our favorites, Grammarly, helps analyze large blocks of copy and provides grammar and spelling suggestions. Grammarly also checks your copy for plagiarism, a major plus when working with contractors, freelancers, and vendors.
Another interesting tool we recently discovered and now use regularly is SurferSEO. Surfer uses your topic input to offer suggestions for keywords and keyword density and applies an optimization score to your content for SEO. While it doesn’t generate the copy for you, its keyword suggestions help provide direction as you write.
There is undoubtedly a plethora of other AI tools that we’re not including here and many that are coming to market at this moment. For the purposes of this article, however, we’ll focus our attention on the copywriting generators that produce copy for long-form articles, blogs, and website copy.
Effectively Using AI Prompts
As chatbot writing has increasingly embedded itself into the world of marketing, the copywriter’s job has evolved accordingly. At Headline Consultants, we work with contract subject matter experts (SME), freelance copywriters, and other copy-producing vendors all the time. It’s getting easier to see who uses AI and how.
Prompt writing is the new art form in content production. Prompt-writing positions can fetch hundreds of thousands of dollars in annual salaries, and agencies are starting to include “prompt writing” as a key element in many of their copywriting job descriptions. Because the push toward authenticity continues to dominate the marketing industry, many brands are happy to pay top dollar for prompt writers who can effectively produce relevant copy that resonates with their audiences.
So, how do they do it?
It’s mostly trial and error, but it’s also about understanding chatbots’ strengths and weaknesses. Incorporating the proper perspective, the right tone of voice, a realistic word or character count, and clear directives into the prompts takes refinement, patience, and a clear idea of what you want the machine to spit out.
Here are the key items every copywriting prompt should include:
- Usage (“I’m writing an email newsletter for a client…”)
- Topic (“… about engineering week…”)
- Tone (“…from the perspective of [name of engineering company].”)
- Purpose (“The goal of this newsletter copy is to congratulate our firm’s engineers and thank them for their contributions to our company.”)
- Request or Ask (“Please produce …” It may sound weird to say “please” when prompting the AI chatbots, but when our robot overlords finally take over the world, you’ll be glad you did.)
- Word (or Character) Count (“… between 75-100 words …”)
- Specificity (“… of email copy emphasizing their importance to our team and the value they bring to our clients.”)
- Exclusions (“Please do not include a greeting or salutation…” – for some reason every chatbot wants their newsletter copy to read like a personal email)
- Inclusions (“…but please do include a subject line, preview text, an interior headline, and an interior subhead for the copy.”)
You can adapt these prompts to fit your specific needs. For social media posts, you might want to add requests for emojis and hashtags (be sure to include the number of hashtags you want, or you’ll get 50) or specific character counts (“no more than 156 characters (with spaces)”) for meta descriptions.
You can also ask the chatbot to omit certain phrases and words to reduce the number of AI “tells” the chatbot will undoubtedly reveal.
AI’s Copywriting “Tells”
Every chatbot we’ve used comes with certain “tells” that indicate it is a piece of AI-generated content. Weirdly, most of these tells are the same regardless of platform and generally involve two specific types: overused words and unique sentence construction. These tells repeatedly appear throughout the copy, cheapening the work and screaming that it was artificially produced.
Overused Words
AI’s most obvious “tell” is its reliance on words that appear repeatedly in articles, regardless of industry, purpose, topic, or tone.
Among AI’s favorite words are:
- Discover
- Diverse
- Elevate
- Enhance
- Enrich
- Ensure
- Explore
- Foster
- Harness
- Innovate
- Powerful
- Revolutionize
- Robust
- Transform
- Unlock
This is only a partial list, but they’re the ones that come immediately to mind. You may also see variations of these words (“ensuring,” “enriches,” or “diversity,” for example). There are plenty of others, and many that are more specific to certain industries and markets.
Interestingly, we asked Claude.ai (our preferred chatbot) what words it believes AI overuses and were surprised at its obtuse response. It included words like “however,” “additionally,” and “regarding.” While these come up frequently throughout an AI-generated article, they’re also words that human writers often overuse, so they’re not a true representation of the words chatbots tend to retreat to over the course of a 1,000-word article.
When we offered the list above, Claude seemed to gain a better understanding of what we were asking and produced this list (along with its profuse apologies):
- Empower
- Optimize
- Leverage
- Streamline
- Seamless
- Cutting-edge
- Game-changing
- Paradigm shift
- Synergy
- Robust
- Scalable
- Disruptive
- Holistic
- Sustainable
- Agile
- Bespoke
- Innovative (in addition to “Innovate”)
- Groundbreaking
- State-of-the-art
- Unprecedented
These are excellent additions to the list. The bolded words are the ones we encounter most often. Generally speaking, if you see these words appear in your copy (especially the ones in the first list and the bolded ones in the second), chances are pretty good someone or something will flag the copy as AI-generated.
Unique Sentence Structure
As part of their programming, chatbots are trained to provide a variety of complex sentence structures. This is a critical component of the natural-language model. Although sentence structure issues do come up periodically, they are rarely grammatically incorrect. For example, if you run AI-generated content through Grammarly, Grammarly will most likely flag examples of passive voice, areas of vagueness, and instances where clarity is lacking. Rarely, however, will it identify issues such as run-on sentences or subject-verb agreement.
One of AI’s more interesting copywriting tells is the relatively frequent use of the trailing present participial phrase. Here’s an example:
“AI tools can analyze lengthy specifications and highlight the most important or unique aspects, saving time and improving accuracy.”
By themselves, participial phrases are sophisticated grammatical arrangements, but trailing present participial phrases (the bolded phrase in our example above) take a great deal of confidence to produce as a writer. So much could go wrong with a construction of this complexity. What if the phrase fails to appropriately describe the main clause? What if there is an error in tense variation or punctuation?
What if—God forbid—a modifier dangles????
While these kinds of phrases aren’t rare in human writing, they are certainly uncommon. And yet, AI uses them all the time. Any piece of content with more than 1,000 words will undoubtedly have at least one trailing present participial phrase and likely two or more.
Recently, our team had the pleasure of discovering the most PERFECTLY AI-generated sentence ever:
“The groundbreaking software update went live, elevating system performance, enhancing user experience, and fostering a seamless transformation into the future.”
Simply delicious.
The Limitations of AI Checkers
There are in fact dozens of AI checkers on the market. There’s GPTZero, Copyleaks, Sapling, and so many more. At Headline Consultants, we use ZeroGPT and Undetectable. We do not, however, place a tremendous amount of faith in them. Why? Turns out, machines aren’t always great at flagging other machines.
While the article you’re reading right now won’t light up your AI checker, our company has written quite a few original articles that have. Even when every word in an article is human-generated, AI checkers will frequently flag it anyway. Sometimes it’s only 40%; other times, it can be as high as 60% or 70%.
Here’s what the AI checkers are looking for:
- Sentence complexity
- Error frequency
- Factual density
- Repetition
- Formatting peculiarities
- Stylistic Markers
This means that if your human-generated copy is relatively sophisticated, free of errors, well-researched, and has a consistent use of voice and tone, you’ll probably short-circuit the AI checker bot and reduce its feedback to a bright, flashing warning sign that blinks “ARTIFICIALLY GENERATED” over and over.
AI Watermarks
Believe it or not, chatbots tend to embed AI “watermarks” into the copy they generate and AI checkers look for these imprints when reviewing content.
These subtle and hidden markers appear in AI-generated text to let AI checkers know if the content has a copyright issue, obfuscates transparency, or includes some nefarious (yet nebulous) “misinformation.”
We discovered the watermark revelation by accident after researching why we were seeing so many trailing present participial phrases in AI outputs. Turns out, the trailing present participial phrase is itself a “syntactical watermark.” There are also “semantic watermarks” (subtly changing meaning or context) and “statistical watermarks” (where statistical properties of text are slightly and nearly imperceptibly altered).
While not every large language model is said to embed watermarks, the temptation for them to do so must be irresistible. And while we may never be able to fully remove watermarks from AI-generated content, recognizing them when you see them will go a long way toward helping you humanize the copy.
What’s a Brand to Do?
If you’re willing to spend time tinkering with the chatbots to get copy that resonates with your audiences and aligns with your goals, your time will be well spent. Chances are you’ll discover efficiencies you never knew existed. You’ll find topic ideas and strategies you may never have imagined.
But if you’re like most of our clients, whose day-to-day does not include nerding out on grammar, spelling, and sentence structure, you’re probably going to want to invest in a content marketing agency or find a prompt writer who can make the machine dance like you want it to.
And so, until ChatGPT becomes self-aware, brands will still need humans to help other humans create artificially generated content for humans everywhere.
***
Ready to thumb your nose at our AI overlords? Send your prompts to the humans at Headline Consultants instead. We use AI, not the other way around. We promise our copy is authentic, human-centered, and most importantly in alignment with your business goals. Contact us today, and let’s undermine the cyborgs together.
