Why AI Won’t Replace Human Writers Any Time Soon

When generative AI first hit the scene in late 2022, I, like many other technophiles, was excited by the possibilities of an AI that could write. As a professional writer, I was also more than a little freaked out. Could this thing replace me? Or do what I do well enough to make my work less valuable?

Now that the dust has settled a bit—a year and some months later—the technophile in me is a bit disappointed. And the writer in me is more than a little relieved.

Automation has always raised fears of job replacement. Proponents have tried to calm such fears by framing automation as a way to reduce the drudgery of work so workers can focus on higher-level tasks.

In the case of my profession, I think they’re right.

AI for Writing

McKinsey estimates that AI could automate 70% of business functions by 2030, causing upheavals in just about every related profession.

That’s the big picture. But what does it mean for a given profession and the people in it? What are the implications for people in, for example, the marketing writing business?

McKinsey has an idea:

“Imagine a marketing leader uses a gen AI application to write a creative brief that previously would have been developed by a more junior marketing associate.”

Okay, let’s imagine it. In fact, we don’t have to imagine. We can jump on a chatbot and ask for a brief for, say, an article on why AI won’t replace human writers.

AI for Content Briefs

Asked to start with a compelling statistic to support the claim that AI won’t replace human writers, a leading AI-powered chatbot gives me this answer:

“Despite advancements in AI, 85% of readers prefer content written by humans over AI-generated text.”

That sounds perfect for our article. But the AI didn’t provide a source. And since my work requires me to back up every claim I make, I have to find one. The trouble is, I can’t.

The first hit I get in a web search is a Forbes article titled “Humans Prefer AI-Generated Content, New Research Suggests.” In other words, instead of support, I find evidence contradicting the AI’s claim.

The research, from MIT, had human writers and a chatbot facing off to write 100 words of persuasive copy. Human readers, it turned out, were more persuaded by the AI-written copy. At first glance, that looks bad for my thesis.

But dig deeper, and you find that the bot was given free rein to say anything to make its case, with no need to worry about accuracy, potential legal liability, or whether the copy could actually drive sales. Just like the stat I got when trying to write my brief with AI.

These aren’t isolated incidents. AI routinely makes things up, putting you on shaky ground if you try to rely on it for factual content.

Strike one for AI.

Telling Stories with AI

Here’s another bit from our AI-generated brief:

“Share a poignant human story that demonstrates the unique ability of human writers to connect with audiences on a deeper level.”

That sounds good. Let’s ask the AI to do that. Here’s the result:

“Every morning, Eleanor would sit by her window, overlooking the tranquil meadow, and write letters to her late husband, James. They were letters filled with memories, dreams, and the quiet ache of loneliness.”

This goes on for six paragraphs I can’t use because it’s not relevant to my article. I have to try again.

Strike two for AI.

Refining AI Prompts

I give the bot more direction, asking it to limit its output to 100 words and target a business audience. It created a character named Sarah, a copywriter at a tech firm. Here’s some of what it wrote:

“Drawing from her childhood fascination with technology, she penned a heartfelt narrative infused with authenticity and passion…. In a world dominated by algorithms, her story reminded us: nothing sparks emotion like the human touch.”

At first glance, this may seem okay. But if you look a little more closely, you’ll find it tells you nothing specific about Sarah. Why does she like tech? What makes her writing heartfelt? How does she spark emotion? We don’t know. It’s all tell and no show, breaking a cardinal rule of good writing.

This piece, too, is unusable. It isn’t compelling and doesn’t convey any real information.

Strike three, and AI is out.

Saying Something New

AI generates text by analyzing large amounts of training data (i.e., human-created writing) to predict the word most likely to come after the one before it.

By definition, AI text represents the lowest common denominator of what’s gone before. Hence, cliches like tranquil meadows and generalities like heartfelt narratives, but few specifics.

Want to talk about what your company can do better than any other? Highlight the unique insights of one of your executives? Tell a real story about how your product improved someone’s life? You can’t do it with AI.

But what about humanizing AI content? Can you take AI-generated output and somehow make it sound more human?

Unfortunately, not in my experience. Not if you want something unique and accurate. Try, and you’ll end up revising so much that you might as well just write a new piece from scratch.

How about the reverse: can you use AI to improve your writing? The answer to that question is definitely yes.

Writing with AI

At the beginning of a writing project, AI-powered search engines help me get up to speed on technical concepts, summarize research, and gather sources.

At the end of a project, AI proofreading tools catch grammatical and spelling mistakes and streamline my writing, for example, by flagging lengthy sentences and suggesting different word choices. Currently, my grammar checker shows 41 issues in this article for me to address—a pretty typical result for me for a piece this length.

But for writing itself, there’s still no substitute for hard work. And writing is notoriously hard. As various writers have said in various ways over the years, “Writing is easy. You just open a vein and bleed.”

So, while AI can automate some writing-related tasks, it won’t automate marketing writing jobs out of existence. At least, I would venture to say, by McKinsey’s estimated timeframe of 2030 for business function disruption.

That’s because, with any writing project, you must answer the question, “What can my readers get here that they can’t get anywhere else?” After that, it’s all about the bleeding — something AI, of course, can’t do.