Generative AI as a Writing Tool: Cautions and Tips

To use AI effectively you have to frame and reframe your prompts, evaluate multiple drafts, and then fact-check, rewrite, revise, and edit. In addition, to get usable text, it takes several prompt iterations that require in-depth subject matter and writing knowledge. In many cases, it is less work to create your own first draft.

Professionals use writing to connect to others, build their brand, and network. You need to develop your own voice and create an authentic connection to others. Author and sociology professor Tressie McMillan Cottom defined voice as “that elusive fingerprint of all textual communication” which requires “a relationship between the reader, the world and the writer.”

Writer, editor, and professor of journalism William Zinsser emphasized in On Writing that “the product” that all writers have “to sell is not the subject being written about, but who” they are. AI tools, in most cases, don’t know who you are and cannot write in your voice to sell your ideas to your reader.

Writing isn’t a static recording of your developed thoughts; writing is part of the process of thinking deeply about a problem, concept, or work of art. This type of deep thinking trains your brain to perform even deeper thinking.

According to Peter Elbow, foundational scholar of writing instruction, “writing…is a way to grow and cook a message.” In addition, writing can also contribute to the formation of new ideas.

Other research, published in Language Assessment Quarterly, shows that writing can even increase gray matter volume—more gray matter means enhanced cognitive performance.

Allowing an AI tool to think for you outsources critical thinking practice and robs you of the opportunity to develop a skill that employers desire, according to surveys by the National Association of Colleges and Employers.

In competitive fields where most people have the same degree, creative thinking and problem solving is how you’ll stand out. If you have Gen AI do your writing for you, you may be losing years of cognitive development that you’ll need later.

AI tools like ChatGPT do not provide many sources even when prompted. They also confidently make incorrect claims. They are built to predict the next most likely word, not to provide vetted, reliable information. For example, Fast Company asked ChatGPT “What was the first TV cartoon?” and it gave a different answer every time the question was repeated, most of them incorrect. CIO’s report on AI errors that led to major issues for corporations from Air Canada to Zillow shows that the problem goes beyond cartoon trivia.

Issues with accuracy may get worse; according to The New York Times, when AI-created content is used to train generative AI tools the output deteriorates, becoming increasingly repetitive, inaccurate, and incoherent.

Margaret Mitchell, an AI researcher and entrepreneur interviewed by The New York Times, further noted that AI bots, trained on internet content, reproduce all “the biases and toxic information” their sources contain.

A study described in Scientific American found, for example, that ChatGPT included stereotypical language in recommendation letters, associating men with words like “expert” or “integrity” and women with words like “beauty” and “delight.” If you do not edit AI generated content, you could inadvertently reflect bias and ruin your own reputation at work.

This situation is unlikely to change; the biases built into large language models are reinforced by biased prompts added by many users, and, as one computer scientist explained, removing biases could create additional problems. It is up to the user to exercise caution and critical thinking.

Drafting is a key part of writing and a skill that requires practice. As John Warner noted in The Writer’s Practice, “we get better at writing the same way we get better at anything else: by doing it.” Widespread use of calculators didn’t mean that no one had to learn how multiplication and division work. Being a skilled writer yourself is going to make it easier to use AI assistance effectively in the future; as users are discovering, getting the most out of chatbots requires you to write clear, detailed, and specific prompts.

Copying and pasting client or proprietary data into a third-party AI tool raises serious security risks. The experience of using an AI tool trained on limited internal data is different from using a tool trained on centuries of text vacuumed from the internet. Becoming too dependent on open-source AI tools for your writing could cause you problems in the future if your workplace restricts or limits their use.

AI-generated messages have a generic tone unless you repeatedly refine your prompting questions and push for greater and greater specificity through careful reviews of multiple drafts.

Writing that is narrowly tailored to your reader will be more effective in communicating your message than a message without personalization. Think of spam email and mail addressed to “current resident.” Many AI produced messages have the same feel and will be equally ineffective at reaching an audience.

If you outsourced a report to AI and your boss has questions about it, it will be awkward if you can’t answer them because you didn’t write something you presented as your own work. Drafting and revising help you develop a deep understanding of a problem or issue, which will be necessary if you’re asked for additional explanation.