AI Copywriting: Why Writing Jobs Are Safe

In the future, we’ll all be editors.
Hal Koss
May 13, 2021
Updated: August 19, 2021
Hal Koss
May 13, 2021
Updated: August 19, 2021

The 2002 film Adaptation stars Nicolas Cage as a screenwriter who has a terrible case of writer’s block. He sits down to type, but the words never come. He second guesses every thought that crosses his mind. He can’t even squeeze out a single sentence before daydreaming about the kind of muffin he wants with his coffee.

Many writers — not just neurotic Hollywood scribes — will tell you that the hardest part of the process is getting started. And writers who work in marketing, in particular, don’t have the luxury of waiting to pick up the pen until inspiration strikes; they often write in high volumes and a variety of formats.

Services have arrived on the scene to help these writers conquer the blank page. And they all share something in common: they’re powered by artificial intelligence.

These tools include Copy AI, Conversion AI, Anyword, Copysmith, Writesonic and Shortly AI. They are designed to transform simple human inputs (like brainstorm notes and bullet points) into clean prose — full sentences that pass off as human creativity.

It’s all but inevitable that AI writing assistants will help shape the future of content marketing and copywriting.

The matter of how much, though, is still up for debate.

More on CopywritingWe Had an Award-Winning Copywriter Analyze Toilet Paper Ads

 

How We Got Here

Marketing teams rely on the written word for just about everything: product descriptions, social posts, website copy, email marketing, blog posts, paid ads and webinar scripts — you name it. 

While the average reader might conclude that the web has enough content already (hello, fellow doomscroller), the truth is, businesses looking to drive awareness about their brands are scrambling to produce more of it — and with increased scale and efficiency.

After all, more than half of companies are thought to spend over $10,000 a year on content marketing, and content budgets are expected to keep climbing in the years ahead. AI writing assistants, the thinking goes, may be able to help businesses keep up with the quickening pace.

If you feel like the natural text generation industry exploded overnight, you’re not wrong — and there’s a good reason for it. Several of the AI writing services on the market rely on Generative Pre-Trained Transformer 3 (GPT-3), which was created by OpenAI, an artificial intelligence research laboratory, and released in June 2020.

For the uninitiated: GPT-3 is a language prediction model that is surprisingly good at producing text that could pass for being written by a real person. It was trained on a massive dataset of crawled web pages, books and Wikipedia entries. If you feed it a paragraph, it can spit out what it predicts is coming next (based on what it knows).

Although GPT-3 copy isn’t human grade yet, many companies see it as a way to help solve the empty page problem. Give the machine a few words, a few sentences, a few paragraphs, and out comes ad copy, social captions and blog posts.

 

How It Works

Although many of these services run on GPT-3, they usually have a mix of homegrown tweaks and adjustments that make them better suited for specific use cases.

From the user’s point of view, they follow the same basic structure. Open a window, then select the kind of text you want to generate: Instagram caption, landing page, Google ad and so on. Each one is calibrated for the context. (You wouldn’t want to be given a blog post introduction when you’re asking for an email subject line.)

Then you key in relevant information to give the AI a little something to work with. It can be as simple as a few sentence fragments or stray thoughts. Nothing gun-shy or time-strapped writers need fret over.

Finally, there’s a button to press that generates the machine’s suggestions.

In my brief experimentation, it seemed like these services generally spit out more misses than hits. Occasionally, suggestions are downright nonsensical. But usually, there are at least one or two outputs that seem like they were genuinely written by a human mind, which could be copied and pasted into its destination without much, if any, adjustment.

I’ve noticed, too, that, the more specific the input, the less volatile the output. (Some services allow users to drop in links to their websites, which the AI scrapes for information — in case typing is too laborious.)

With one such service, I created Google ad copy for a fictional product, a vegan leather baseball cap. I wrote that it was a “classic billed unisex hat made from 100 percent genuine vegan leather; it’s functional, stylish, great for all occasions and good for the environment.”

After a brief performative whirring, the machine deposited about a dozen suggestions, the majority of which struck me as minor, often inferior remixes of what I had already written. But there were a couple exceptions.

One suggestion had a very specific and chaotic energy: “Wedding season is here... and this hat is PERFECT.” Not usable, but certainly amusing.

Another stood out, but for positive reasons; it seemed simple and clever. Something I could definitely work with, if not copy and paste directly. 

“Make a statement that is also sustainable,” it read.

The ad copy captured my fictional vegan leather hat company’s essence better than I had anticipated.

More on Natural Language GenerationLights, Camera, AI: How We Made a Movie with an AI-Penned Script

 

The Future of Writing Is … Editing?

The release of GPT-3, and the subsequent adoption of its abilities by copywriting services, has, perhaps inevitably, reinvigorated the perennial moral quandary

Will AI replace writers?

I posed this question to Dave Rogenmoser, co-founder and CEO of Conversion AI.

“Yes and no,” he told me.

“Low-level copywriting and content creation is in serious jeopardy,” he said. “It doesn’t replace really good people.”

But give it 10 years, Rogenmoser said, and eventually, services like his probably will be able to write feature-length articles as good as mine — not just “B+” ad copy.

“It doesn’t mean that you’re gone,” he said, perhaps sensing my unease over our video chat. “It just turns everyone into an editor.”

Rogenmoser imagines a future in which AI tools augment writers, instead of replacing them. The tools will fill the blank page, then writers (editors, in this scenario) will polish them up.

If you’re a business that would hire a marketer to run content in house today, you wouldn’t use one of these tools instead of hiring that person, he said. What you would do, ideally, is hire a person who brings an editor’s mindset to the writing process, and who uses these tools to take AI-generated copy from good to great.

Rogenmoser compared the dynamic to that of software engineers, many of whom, he said, don’t write code from scratch, but copy and paste from elsewhere.

“Writing sentences from scratch,” he said, “will be gone in the future.”

“Writing sentences from scratch will be gone in the future.”

Inbar Yagur is the VP of marketing at Anyword (formerly Keywee), one of the few AI writing services whose language models predate the release of GPT-3.

Her vision for the future of copywriters is one that requires less editorial judgment and more data-driven decision making.

“The future of this is not generation. The future of this is curation,” Yagur said. “It’s helping the person make a choice, and understand which of those millions of words they can generate will actually work.”

She explained Anyword’s software to help make her point: Anyword provides users with a numerical score for each AI-generated text suggestion it makes, in addition to the text itself. The grade, on a scale of one to 100, is meant to represent how well the particular copy would perform in its given use case, based on the historical data the AI has been fed about ad performance.

In other words, I don’t need to rely on my gut to pick out which AI-generated vegan leather ad copy suggestion is “best.” Anyword’s service will tell me which one will, say, get the highest click-through rate (with 76 percent accuracy, Yagur said).

Marketers who use AI writing tools typically modify the suggestions before using them in their content, Yagur said. But these tweaks are based on their own taste and preferences, not data. So they might be choosing poorly (if conversion is the goal).

“The problem with that human touch is that [it] comes with bias,” she said. “Getting AI text generation right is going to be about helping bridge that gap between your bias and what you actually ended up delivering.”

If AI tools eliminate the need for human touch when it comes to both producing and picking what words to use in marketing, where does that leave the writer?

“It’s not here to replace us,” Yagur assured me. “It’s here to help us.”

I’m not so sure. I hear there’s money in vegan leather hats though.

Great Companies Need Great People. That's Where We Come In.

Recruit With Us