One of technology’s great promises has been that it would free us from rote tasks and leave us more time for creative thinking. Now, generative artificial intelligence (GenAI) promises to do some of the heavy lifting when it comes to the written word, drafting correspondence, such as email messages. But that raises some questions: Should you use it—and can the recipient tell if you’ve relied on a bot to do your writing for you?
TO BOT—OR NOT?
Kathleen Day, a lecturer in business communication, finance, and corporate governance at Johns Hopkins Carey Business School says GenAI-created correspondence should never be used without a human reviewing and contributing original content to it; also, that the content created entirely by GenAI needs to be labeled as such.
“For anyone to send me something saying, ‘This is from me,’ and it was really from ChatGPT, it’s not different than if they sent me something saying it was from them [but someone else] had really written it,” she says. She adds that lack of disclosure that something has been written by GenAI is dishonest and if something is submitted for publication, it’s plagiarism.
Franklin Orellana, chair of the computer information system and the data science programs at Post University, says that GenAI-written correspondence may have flaws ranging from syntax issues to false information, which can damage credibility if they are passed along as the author’s work. In addition, Orellana says that AI can be used for nefarious purposes, such as phishing or other cyber crimes or to spread fake information. So, it’s a good idea to always read correspondence with an eye for exactly how much was authored by a bot.
I SPY AI
But can you really tell if AI has authored an email or other correspondence? While a slew of AI detectors are available, there are questions about how accurate they are in determining AI. In June 2023, a team of researchers tested a dozen publicly available AI detectors and two commercial platforms. They didn’t mince words about their findings, stating that the platforms were “neither accurate nor reliable.”
But there are some telltale signs that can belie bot-written copy, says Madeleine Lambert, director of marketing and sales at Originality.ai, an AI detector. Here are some of the key signs:
1. The language is bland
Without a human editor, GenAI-created language tends to be “too perfect,” Lambert says. Also, the language may be overly formal or stiff. If you have had conversations with the sender before, you may notice that the email doesn’t reflect how they typically use language. In other words, it doesn’t “sound” like them. A recent “sentiment analysis” of generative AI-written content found that bot-written content lacks emotion.
2. The structure is uniform and repetitive
Writing sentences that vary in length typically makes your content more engaging for readers. Some sentences are short, others much longer, and you may string together a few ideas, descriptions, or other elements.
Generative AI doesn’t really do that, Lambert says.
Sentences and paragraphs are typically of similar lengths and tend to run on the long side. You may notice that sentences have the same or similar structure throughout the work. “Sentence structure and length are typically very consistent, almost down to the number of characters,” she says.
3. Periods dominate punctuation
When humans write, colons, semi-colons, and (too many) em-dashes often pepper the page. The confident writer may even throw in the occasional ellipsis or “scare quote.” GenAI tends to stick to the basics, with mostly periods and commas, Lambert says.
4. Hallucinations may happen
Day, of Johns Hopkins’s business school, warns that so-called AI hallucinations—instances where the GenAI platform simply makes up data or information that isn’t true—are relatively common. And relying on AI to write correspondence for you without carefully fact-checking the contents could be professionally damaging if you’re passing along bad information. “If there’s anything wrong with it, it comes back to bite you. It’s not going to be an excuse to say, ‘Oh, gee, a machine wrote it,’” she says.
These attributes might help you spot AI-written email messages and avoid those that are spam, or worse. But they may also help you know what to look out for if you use such tools to get over the paralysis of staring at a blank screen. Adding your own voice, fact-checking rigorously, and ensuring that your correspondence has personality and warmth can help turn problematic AI content into something that works for you.