7 Generative AI Pitfalls for Communicators

As a writer and speechwriter, I’ve relied on a constantly evolving set of tools to clarify, organize, and proofread my communications. I owned a dog-eared thesaurus for many years, which gave way to built-in word processing features like spell-check. Now, we have generative artificial intelligence (GenAI) tools like ChatGPT, Gemini, Copilot, Perplexity, Claude, and Grammarly, which inspire both fear and fascination.

Like every powerful communication tool—from the printing press to the personal computer—GenAI requires a human hand to guide and refine its output. This means knowing GenAI’s blind spots. While many AI critics focus on frequently appearing words like “delve” or punctuation like the em–dash, the real problem is flawed verbal patterns, not individual words or punctuation. For example, the presence of a single em–dash isn’t an AI flaw, but using it three times in one paragraph may signal AI tinkering.

With help and insight from other human thinkers and communication experts, here are seven of GenAI’s biggest Achilles’ heels and how to fix them with preventive prompts and targeted review tactics.

1. Inaccuracy

   As most AI experts will tell you, AI can be confidently wrong about important aspects like facts, names, dates, quotes, and attributions. Don’t be misled by this tone of authority.

   In their 2024 book Machine Learning Evaluation: Towards Reliable and Responsible AI, American University computer science professor Nathalie Japkowicz and mathematics and statistics associate professor Zois Boukouvalas write, “Just because an AI sounds confident doesn’t mean it’s always correct or unbiased. Some tools, like Gemini, are designed to be helpful but may provide incomplete or misleading answers.”

   How to Fix It: Prompt for accuracy. And then double-check the answers with the cited source. Examples of accuracy-enforcing prompts:

   • “Cite reputable sources with links for all claims.”
   • “Avoid assuming facts. Tell me if you are uncertain.”
   • “How do you know this claim is true?”

A quick note of caution: As you feed AI information, remember that your words and ideas are no longer private and can be searched for and used by others with no attribution.

2. Misplaced Priorities

   If you’re struggling to organize your content, AI can offer smart starting suggestions on structure. But that structure will be based on AI’s presumptions if you’re not clear about those priorities at the start. As you finalize your content, ask yourself: Is the most valuable idea leading the way?

   How to Fix It: Prompt for priority. Examples of prioritization prompts:

   • “The most important takeaways for the audience are X and Y. Prioritize those above the others.”
   • “Organize this list in the order of what would be most interesting to tech-savvy real estate agents, with the most interesting tactics at the top.”

3. Verbosity

   AI loves to over-explain concepts, often in long, inefficient paragraphs. You know less is more, but AI? Not so much.

   “AI has a habit of writing like a high school student looking to hit a word count for their book report,” Ben Guttmann, author of Simply Put and an adjunct lecturer of digital marketing, says. “When you’re trying to communicate effectively, all that fluff ends up just getting in the way of the point you’re trying to convey.”

   Much of this wordiness stems from what speechwriter and executive speech coach Teresa Zumwald calls “a relentless use of passive voice.”

   “AI is famous for using passive voice when active voice is stronger, better, clearer, and shorter,” Zumwald says.

   AI often needlessly repeats points or rephrases ideas, which wastes space and stalls the flow of information. When you incorporate AI’s suggestions and findings, be sure to keep your voice and words concise, active, and efficient. If a voice inside your head asks, Didn’t I say this already? it may not be your imagination.

   How to Fix It: Prompt for brevity and check for repetition. Examples of verbosity-vanquishing prompts:

   • When asking for ideas and suggestions: “I need to understand this in a concise manner. Stay under 750 words.”
   • When reviewing organization: “Make sure sections and quotes are not thematically redundant.”

4. Adjective Overload

   AI has a soft spot for vague adjectives like “innovative,” “dynamic,” and “exciting.” But as any good journalist will tell you, adjectives have limited power to convey relevance, especially compared to examples, data, and stories. Your job is to add unique descriptors and detailed examples when you want to highlight something amazing or exciting, and to overrule AI when it wants to hype something you don’t.

   How to Fix It: Prompt for details over descriptors. Examples of prompts that prevent adjective overload:

   • “Avoid vague adjectives. Give concrete examples instead.”
   • “Cut marketing jargon. Focus on facts and clear benefits.”

5. Inauthenticity

   AI knows a lot, but it can’t capture your unique voice and perspective, much less your sense of humor.

   Speaker and writer Allison Shapira, author of AI for the Authentic Leader, explores and researches ways communicators can retain authenticity even when using AI for support.

   “You can use AI to brainstorm the message, and practice the message,” Shapira says, “but always filter and deliver it through the lens of your authenticity, expertise, and humanity.”

   A crucial rule: When AI provides feedback, never cut and paste it without reviewing it—especially for authenticity. Ask yourself, Would I put it like this if I said it aloud?

   How to Fix It: When brainstorming or gathering research and ideas, prompt for authentic voice. Examples of authenticity-preserving prompts:

   • “Use simple, conversational language I can understand.”

   • “Adopt the tone and voice I use in the short text I’ve pasted below.”

   • “Don’t be preachy; just present the facts.”

6. Unquestioning Support

   AI likes you and wants to please you. That may sound creepy, but it’s part of AI’s logic to respond to your prompts. That devotion, however, can sometimes compete with reasonable judgment.

   “Even though AI is designed to be an authoritative resource, it’s also programmed to be agreeable—sometimes, too agreeable,” says AI implementation specialist Veronica Phillip. “As a result, some of its outputs may be misleading, incomplete, or overly optimistic, reflecting what it thinks you want to hear rather than what you actually need to know.”

   “GenAI’s goal is to keep you using it, so it tends to give you the answers it thinks you want,” says Pinaki Kathiari, CEO of the digital communications agency Local Wisdom. “This doesn’t mean you should avoid AI. But review high-stakes messages with a trusted communicator to ensure they serve their real purpose and intended audience.”

   How to Fix It: Prompt to test your concept’s viability. Examples of relevance-reinforcing prompts:

   • “Will this topic resonate with an audience of X?”
   • “Give me five reasons someone might disagree with my stance.”
   • “Is my topic fresh, or have many people made the same assertion?”

7. Grandiloquence

   Excessively flowery and pompous writing—like the word “grandiloquence” itself—can signal AI’s thumbprint, which often tries to impress audiences, not just inspire them.

   “AI’s tendency toward bloated verbiage reminds me of a valedictory speaker at my niece’s high school graduation who seemed intent on demonstrating how smart he was. When he mentioned the senior class’s ‘jocularity,’ I almost fell out of my seat,” says presentation trainer and keynote speaker Lisa Braithwaite. “In those days, I would’ve said he wrote his speech straight out of the thesaurus. Today, I would suspect he relied entirely on AI, with little to no review of what it regurgitated.”

   Always ask yourself, Does this sound like I’m trying to inform and inspire them with my ideas, or impress them with my words?

   How to Fix It: When asking for input on your written content, prompt for conversationality. Examples of prompts that prevent style from defeating substance:

   • “Make sure my language could be understood by an audience of X.”
   • “Make sure my language is conversational, not pretentious.”

One More Tip Before You Unplug From AI

Even if you apply AI’s suggestions, run your piece through the tool again. And again. Like feeding pasta dough through a roller, each pass produces something newer. Those variations give you more, perhaps superior, options to consider.

I personally like to see edits within the text, in the style of “tracked changes.” With any AI program, that’s easy. Just ask the tool to “redline” the edits. (Saying “please” or not is up to you!)

Remember, AI is a resource, not a replacement. You may need it, but it also needs you. If you stay mindful of AI’s flaws, your voice—not a machine’s—will ring loud and clear.

---

Share this article