13 Unlucky Perils of AI-Generated Content

Is AI-generated marketing content really that bad?

It can be if you let it. And unfortunately, too many well-meaning businesses are letting it …

The problem arises when companies view generative AI as a cost-saving technology without fully understanding that the content it creates often severely lacks depth, narrative, accuracy, and/or soul. It doesn’t connect with its audience, and Google may penalize it for low quality.

So the problem might not only be the AI but also the humans who overlook, aren’t aware of, or dismiss the shortcomings.

That doesn’t mean you should throw out the AI-generated baby with the bathwater. By understanding AI’s limits—including the 13 perils highlighted in this article—you’ll be better positioned to adopt large language models (LLMs) for your content strategy.

1. Repetition

An important fundamental to remember about generative AI is that it’s creating output based on what it thinks you want to see. So, for example, if you ask for 1,500 words on a certain topic, the AI might give you 1,500 words … and repeat and reword ideas, paragraphs, and whole sections to get there.

Good prompting can curtail the bloat, but then you risk another problem: The AI fails to reach the word count you want. Be realistic—some subject matter may not warrant the word count you’re seeking. But if you think the topic is worthy of a certain length, you may need to manually get it across the finish line.

2. Too Formulaic

Generative AI often relies on writing structures that fail to effectively serve marketing content. You don’t need sentences of copy previewing everything that will follow. You don’t need a long conclusion summarizing everything you just wrote. You can spin a narrative that doesn’t mention three key points, a la the five-paragraph essay you might have learned to write in eighth grade.

Unfortunately, AI falls back on these formulas and others. It also may struggle to write to your best practices, even when prompted. Essentially, the AI doesn’t know how to create copy that really stands out. The result is content that disinterests readers into clicking away.

3. Accuracy

Generative AI is only as accurate as the internet sources it has been fed and is drawing upon. An example of this I encountered recently: I saw the quote “How did it get so late so soon?”—attributed to Dr. Suess—and wanted to find out where it came from. I Googled it, and the AI result stated it was from Oh, the Places You’ll Go. 

Just one problem, the quote isn’t from that book—it’s just a random Dr. Suess poem, and I couldn’t confirm an absolute source on the interwebs. However, it appeared on a webpage along with (but separate from) the entire text of Oh, the Places You’ll Go, and Google’s AI connected dots that never should have been connected. (To Google’s credit, when I ran this search again, it says “the exact source is debated.”)

Unfortunately, people believe that if AI is telling them something, it must be true. The potential for inaccurate information or, worse, bad advice slipping into content, increases when you use generative AI and fail to double-check what’s being produced.

4. ‘No #%&!’ Copy

When I was a newspaper copy editor, my coworkers and I took care not to write what we called “no s—t headlines”—headlines that were so obvious they didn’t tell readers anything or inspire them to want to read the article. “Coach Proud of His Team” or “Player Looking Forward to Season” just screams, “Duh!” and cuts into the effectiveness of the article.

Alas, generative AI becomes this obvious as it draws from internet sources and yields copy it thinks you want to see. For example, it might tell you, when prompted to write content about advanced strategies on appealing to marketing personas, to—I swear I’m not making this up—to “understand your audience.” No s—t Sherlock! Generative AI often can’t resist the low-hanging fruit and gives you extreme basics when you’re looking for something deeper.

5. Non Sequiturs

Revisiting the idea of generative AI connecting dots, simply put, it stinks at doing so. It can pull in info from multiple sources and present differing points of view, but it can’t take info from those sources and effectively draw conclusions or decide what’s important and what’s not.

Moreover, AI might attempt to connect dots but fail so miserably that it essentially is making things up—thinking that, as a silly example, that one plus one equals rutabaga. Although AI excels in some analytical areas, unless you are specifically feeding what you want it to analyze, the results are hit and dangerously miss.

6. Real-World Examples

Great marketing and thought leadership content often incorporates real-world examples into the messaging. I’ve been incorporating digital storytelling throughout this article—demonstrating how something applies and putting you, the reader, into the story. 

Unless generative AI is taking an example from existing content or from your prompt, it struggles to create new scenarios … and may even make up examples that wildly don’t make sense. 

7. Not Sounding Human

Accuracy and logic issues notwithstanding, this has become maybe the biggest complaint with generative AI: It often sounds like a computer wrote the copy. You’ll encounter phrases, sentences, and other grammar constructions that make you think, “No one writes, speaks, or thinks like that.” Simply, the copy has no soul—and, increasingly, readers are thinking the same way, automatically discounting content as not being reliable, personable, or resonant. 

Unfortunately, good things AI is learning (slowly) to do with content are being dismissed as indicators that a human might not have written something. The recent controversy with em dashes exemplifies this distrust—dashes are an excellent grammar tool for compelling copy (including in this sentence, see what I did there!), but because not many people use them, people think it surely must be a sign of AI. 

8. Overdoing It

Generative AI really wants to please, so when you give too simple a prompt to produce content, it might go overboard and spit out a complete guide on a topic when you wanted a short LinkedIn post. I see this often with outlines—AI can be a big help organizing your thoughts, but it tends to drift away from a tight focus to an unnecessarily expansive epic.

Even with more specific prompts, generative AI may go the extra mile when you don’t want or need it to. Again, AI struggles with deciding what’s important and not important, and it often errs on the side of too much over too little—making you waste time paring down the output.

9. Underdoing It

Less common than generative AI giving you too much, not getting enough can also be a problem when you’re asking for content. Although this might happen because not enough sources are on the interwebs for AI to draw from—and better that the LLM comes up short instead of making something up—it may also occur because AI just doesn’t know where to look for the information that you’re prompting. 

The result? You end up with content with serious knowledge gaps and the aforementioned “no s—t” copy. Readers barely derive any value from it, and your reputation as an industry authority takes a hit.

10. Wandering

As much as people may benefit from using AI to organize notes and thoughts into an organized narrative, the AI sometimes doesn’t quite get there. For example, AI copy might include everything it’s been prompted to include, but without really specific instructions, it may not figure out what’s most important, what relates to what, and how to effectively get from X to Y. Rambling content doesn’t impress readers, potentially confusing them into wondering if your business can help them.

11. Using the Same Expressions

Use generative AI enough and you’ll see the same words, phrases, cliches, and expressions used over and over—to the point it’s a red flag that the content was AI-written. Some of these ad nauseam fallbacks include:

• In conclusion

• To sum up

• It is imperative

• As we’ve seen

• Game-changing

• Part of the tapestry

• It’s not just about … it’s also about 

• It’s important to note

• Let’s dive into

• When it comes to

Unfortunately, generative AI has swept up good words and expressions—I love “furthermore,” “moreover,” and “as a result” but might get dinged for using them. That said (I promise, a human just wrote that sentence intro), most of these cliches are not the best writing, but because AI is taking its cues from bad writing off the web, it thinks that’s what you want to see. 

12. No Impact

Marketing and sales content should impress, educate, and inspire. Unfortunately, these are minor considerations for AI, which is more concerned with generating a response to your prompts. Even if you ask, “Please write a blog that will impress my audience,” AI may struggle with understanding what impressive means. The resulting content barely makes a ripple with readers. 

13. Unnecessary Changes

An occasional frustration I’ve encountered in the last few years is the occasional client running professionally written content through AI to see how it can be “improved.” The client wasn’t not trusting us so much as believing that AI is smarter than humans and whatever it suggests must be better. Unfortunately, the LLM usually comes up with “improvements” that reduce the quality of the copy.

Again, this happens because generative AI wants to please so much that it nitpicks despite having nothing to really nitpick. Grammarly is even guilty of this, falling back on its best practices that don’t actually help much. Stare at the ceiling of the Sistine Chapel long enough and you’ll find faults while straining your back. Unless you’re willing to question AI’s suggestions, you can irreparably damage copy to the point you almost need to start over.

Do these perils sound too perilous? Unretained can help you navigate using AI to help with content, or we can take all the worry out by writing content for your organization. Connect with us to learn more.