Should Copywriters Be Afraid of AI?

AI copywriting tools have flooded the market, claiming to handle everything from long form blog posts to Facebook ads. Should you, as a copywriter, be worried?

This post is a crash course in how copywriting AI works and what it can do. You’ll also find some pointers on how you can use artificial intelligence to adapt, and hopefully stay ahead of the curve, so you don’t end up searching for a second career any time soon.

Who’s afraid of the big bad machine? 

If you’re a copywriter, you may feel like your career is on the chopping block. People experimenting with ChatGPT have achieved some impressive outcomes. And Capterra currently boasts more than 70 AI copywriting tools claiming to tackle everything from long form blog posts to Facebook ads.

Social media, blogs, and the press are abuzz. Most writers are optimistic. Start searching, and you’ll find hundreds of articles arguing that AI is here to help copywriters do their jobs more easily and efficiently, not to replace them. 

This is a good time to pause and look at bylines. The top-ranking articles for “Will AI replace copywriters?” are written either by a.) copywriters, and b.) companies trying to sell AI copywriting tools. No wonder they’re trying to make the best of a dicey situation.

Plus, most of what’s being written in this sphere looks at how AI is performing now, circa 2023. That’s helpful, to some extent. But what about in 2028, or 2033? What kinds of leaps and bounds can we expect? And how do we brace for impact?

To approach something like objectivity, we need to understand what AI is, and what it isn’t.

What’s under the hood?

AI comes in two flavors: Artificial General Intelligence (AGI), and Artificial Narrow Intelligence (ANI.)

AGI is a computer—or a network of computers—that thinks and learns like a human. It broadly mimics intelligence as we understand it, asking questions like, “What is… love?” and calling the scientist who invented it “Father.” AGI is still science fiction.

Turns out, there’s a certain je ne sais quoi to the human mind machines currently can’t capture. A lot of very clever people are trying to change that—they’d love to have their own C3PO or Data to hang out with. But true AGI is a long way out.

ANI does exist. ANI is trained to do one thing well, like write responses in a chat window, or generate images of people with too many fingers and teeth. ANI is what we talk about when we talk about copywriting AI. 

What is GPT-3? Is it programmed to kill?

GPT stands for “generative pre-trained transformer.” GPT-3 was the precursor to ChatGPT. Both were created by OpenAI. (The most recent version of GPT-3 is GPT-3.5.)

GPT-3 is a language model: A program trained to look at a series of words and predict the next word in the sequence. For instance, if you gave a language model the word series Sitting on the couch there was a, it might predict the next word should be cat, dog, or person. 

(You could also give it a sequence with a word missing: Sitting on the ____ there was a cat, and the language model might predict couch, chair, or bench.)

How does the language model know to predict a cat, dog or person sits on the couch, rather than a bulldozer, passionfruit, or subpoena? It’s been trained on a huge corpus of written English, millions and millions of words. It used that training to build a model of how these words are typically associated with one another.

GPT-3 works pretty well (and much better than its predecessor, GPT-2). You can instruct it to write a short passage on a subject, and it can come up with a grammatically correct piece of writing that makes sense. But its output is still easy to distinguish from a human’s.

ChatGPT is trained specifically to engage in dialogue with people, and it does a good job. The team at OpenAI achieved this by adding a stage to its training: Reinforcement Learning. It consists of having teams of humans review and rate the AI’s output. The AI looks at these reviews and uses them to build a model of what types of answers are most useful and appropriate for humans.

How do apps and services use AI to write copy?

Tools like Jasper and Writesonic each use a variety of language models licensed from OpenAI to power their services. 

These language models aren’t used exactly as they are, out of the box—AI copywriting companies use their own training models on them, teaching the AI to produce output for particular use case scenarios. 

GPT-3, working on its own, isn’t an expert at writing Facebook ads, or headers and subheaders for blog posts. Once it’s been trained by looking at many examples how to do so, however, it’s ready to be used by copywriters.

What can an AI copywriting tool do?

Whether you’re mildly cynical about AI’s ability to write copy, or you’re building a Faraday cage in your home office so robots can’t read your mind, you have to admit that creators of AI copywriting tools have done thorough jobs designing their products.

For instance Jasper, one of the most popular copywriting tools, lists over 50 templates—each one a copywriting product you might prepare for a client.

It can create small, specific pieces of copy, like

• Social media ads
• Amazon product descriptions
• Headlines for blog posts
• Title and meta descriptions for web pages

And it can handle medium-length writing assignments and summaries, like

• Writing intro and conclusion paragraphs for blog posts
• Expanding sentences to make them longer and add context and style
• Writing answers to questions on Quora
• Succinctly summarizing longer texts

It even claims to handle broader, almost strategy-level tasks, like:

• Coming up with marketing angles
• Creating brand and product stories
• Brainstorming ideas for blog posts
• Writing unique value propositions

Should you be afraid? Are you about to be replaced? Not quite.

In any of these use case scenarios, it’s a human who initially drives the work. AI writing tools—just like AI image generators—need prompts. 

What are AI prompts?

A prompt is an instruction for an AI tool. By now, you’ve probably seen people on social media sharing funny or bizarre AI art generated by tools like DALL-E or Stable Diffusion.

Vague or confusing prompts create vague or confusing output. On the other hand, the more specific and concrete you’re able to make a prompt, the closer the output will be to what you’re looking for.

Coming up with effective prompts for AI is an art unto itself, a field known as prompt engineering. You can even take AI prompting classes. More advanced techniques in prompt engineering require some knowledge of computer code, but the simpler forms—coming up with prompts to enter in ChatGPT or AI copy tools—are accessible to all skill levels. (Check out Learn Prompting to get started for free.)

AI prompts in practice

To get a sense of how AI prompts affect output, let’s try an example with ChatGPT.

Suppose you’re writing a newsletter for a skincare company and you need a short blurb about organic exfoliating scrubs.

Here’s what ChatGPT gives us when we prompt it with the question: Why is it important to exfoliate?

Not bad—certainly a thorough answer. But it doesn’t talk about exfoliating products. To get a more specific result, let’s try: Why is it important to exfoliate, and what types of products can you use?

Let’s try to focus more on products, without talking about the potential harms of over-exfoliation: Tell me why it is important to exfoliate and what types of physical products can be used, without providing any information about over-exfoliating

More specifically, leaning into the benefits of organic scrubs: Tell me why it is important to exfoliate and what types of organic scrubs can be used, why it is important to use organic scrubs and not ones containing microbeads, without providing any information about over-exfoliating

Just about there, but a bit too long. So, let’s try: Tell me why it is important to exfoliate and what types of organic scrubs can be used, why it is important to use organic scrubs and not ones containing microbeads, without providing any information about over-exfoliating, and do it in five sentences.

There’s our blurb, more or less. You may need to edit it for voice and tone, and add a specific reference to the client’s product. But our final output—prompted by a long description, being specific about what information to cover, what not to cover, and the length—is much closer than the first one we produced using a short, vague prompt.

Learning to write effective prompts is an important skill no matter what type of work you’re doing with AI powered copywriting. As you’ll see below, it’s one of the tools you can use to set yourself apart as a copywriter.

True or false: Is AI already replacing writers?

In December 2022, Buzzfeed laid off 12% of its workforce in an effort to reduce expenses. In January 2023, a memo from CEO Jonah Peretti surfaced in which he announced plans to leverage AI to produce online content and quizzes. 

Investors reacted positively. Following the publication of the memo by the Washington Post, the value of Buzzfeed’s stock climbed nearly 300%. 

Peretti said Buzzfeed would still rely on humans to provide “cultural currency” and “inspired prompts.”

Is this the first volley in a war between AI and human writers? Can we expect ad agencies to start cutting staff, or tech companies to kick their content writers to the curb?

The Buzzfeed memo noted that AI would be brought in to help generate custom content—online quizzes—for users. Possibly, it could even tackle hard-hitting journalism, like 25 Nostalgic Things From Your Childhood That Will Bring You Comfort As An Adult.

Notably, ChatGPT has not been tasked with writing articles about international finance or diplomatic relations between countries, or crafting long pieces of investigative journalism covering global warming or the opioid crisis.

AI is being assigned low stakes tasks because, frankly, it’s not effective or reliable enough to handle anything important. That could change in the future, but for now, humans are better writers.

It’s important to keep that in mind when reading articles that gleefully announce AI taking over previously human-powered sectors of the communications industry. For instance, CNN recently announced that ChatGPT has become indispensable to real estate agents, who use it to write their listing descriptions.

But if you spend more than ten minutes reading real estate listings—or if you’ve ever been tasked with writing them—you’ll understand why they’re a match made in heaven for a soulless machine.

It doesn’t take a profound understanding of the human condition to see the words “1200 square foot rancher” and substitute “cozy.” You don’t need a Masters in English Literature to receive the prompt “extensive water damage” and respond with “handyman’s dream.” AI powered copywriting is good for jobs like writing real estate listing descriptions because they’re simple and repetitive.

Communicating with potential buyers, learning about their hopes for the future and what they want from a home, addressing their fears and anxieties, and helping them make the right decision for themselves and their families—all that messy human stuff—is still far outside the capabilities of AI. Even if they’re writing the listings, machines aren’t ready to earn their REALTOR certificates. For that, you need people.

We know AI does a pretty good job of mimicking human language and, with some fine-tuning, it can handle certain copywriting tasks. But what does it fail at? And, if you’ll be competing with AI in the years to come, how can you set yourself apart?

What AI doesn’t do well (yet)

Despite recent advances, there are certain tasks large language models don’t do well. Understanding what AI fails at can tell you where to focus your energies as a copywriter, and how to work with AI rather than against it (as covered in the next section.)

Crediting sources

If you ask ChatGPT what the average March temperature is in Pensacola, Florida, it will happily tell you:

Great. But how do we know that’s true? Where did ChatGPT get its information?

Ask it, “How do you know the average March temperature in Pensacola, Florida,” and you get:

That’s kind of cool. But still, no link to an authority stating the average temperature is 63*F, or even a chart you could use to calculate the average yourself. 

If you’re writing an article about vacationing in Pensacola during early Spring, and you want to cite a source for your temperature info, you’ll need to ask ChatGPT where it gets its information. Here’s the response you can expect:

So, in a word, “Google it.”

If you want to link to sources, in almost every case, you’ll have to do the research yourself. Because of that, AI is limited in terms of the kind of content it can produce, and the amount of time it can save you.

ChatGPT can’t google. That’s important to remember if you’re using it to write content. In its own words:

Finding original or up-to-date sources

Since ChatGPT and other AI tools are limited to referencing the data they’re trained on, they can’t provide you the most up-to-date information available. You’ll have to do that yourself.

And in terms of finding original sources—interviewing customers, for instance, in order to get testimonials or create marketing strategies—your average copywriter AI is nowhere close. 

Reliable factual accuracy

It goes hand in hand with being limited to a fixed set of data: AI’s factual accuracy is always in question. That doesn’t only apply to data that may be out of date. It’s been widely remarked that ChatGPT can provide incorrect answers to questions. If you’re using AI to write content, it’s up to you to do your own research and check that everything it tells you is correct.

A Telegraph article about the Buzzfeed layoffs notes that CNET recently used a copywriter AI to write 75 articles about personal finance, and published them. Not long after, they were forced to take all of the articles down due to factual inaccuracy. When it comes to the important stuff, like telling people how to manage their money or pay their taxes, AI is hardly to be relied on.

The finer nuances of voice and tone

Part of being an adaptable copywriter is being able to take on different voices and tones according to the clients you’re working with and the audiences you’re addressing. 

In a broad sense, AI can do that. Here’s what happens if you tell ChatGPT to explain the First Law of Thermodynamics in a “serious, academic” tone:

And here’s what happens if you ask it to do the same thing in a “bubbly, fun, and lighthearted” tone:

And, just for fun, here’s “Borat voice”:

Don’t feel obliged to read all the way through those three examples. If you do, however, you can see what ChatGPT is doing. 

For a serious, academic explanation, it doesn’t shy away from ten dollar words, and it throws in a formula to up the brainiac factor.

For a fun, bubbly explanation, it peppers in slang like “chilling,” makes extensive use of exclamation marks, and seems to favor the adjective “crazy” as a catch-all.

For Borat voice, it provides phonetic spellings of a vaguely offensive fake accent, draws upon the noble Kazakh donkey to explain energy conservation, and signs off with “Jagshemash.”

But what would happen if you copied and pasted descriptions from a client or employer’s voice and tone guide, and asked ChatGPT to write copy in that style? Would it be subtle, nuanced, and ready for delivery? Or would it need some heavy editing to make it deliverable to the client?

Humans switch their tones and voices all the time—you don’t speak to your mom the way you speak to your college roommate—and it’s something we’re really good at. Effective copywriters have an especially developed sense of this switching, and use it to their advantage. When AI attempts it, the results are crude at best.

Large-scale planning

Impressively, a number of AI tools promise to generate marketing angles and ideas for blog articles. But they’re still miles away from humans when it comes to taking on larger scale marketing tasks.

While you might be able to draw on AI to write social media ads, it won’t plan a campaign for you. If a client needs to completely revamp their website, an artificial intelligence copywriting tool may be able to fill in the blanks, suggesting copy for certain sections. But it likely won’t be able to tell you what those sections should be, or what kind of information the client’s customers are looking for.

When it comes to looking at things holistically and getting a bunch of moving parts to work together in tandem, AI just isn’t there yet. 

Scripts

Given the right prompts, AI can write a script for a video or podcast—up to a point. Leave it to its own devices long enough, and it will write something not entirely human.

For an example of the kind of surrealism you can expect from a script written by AI, check out Nothing, Forever, an AI-generated Seinfeld episode that plays 24/7, ad infinitum.

The fact is, an AI talking to itself for an extended period quickly drops any pretense it may be human. There’s a whole genre on YouTube of bizarre movies, cartoons, and stand up comedy generated by AI. The draw isn’t wit or originality; it’s weirdness.

For the time being, even if AI tools claim they can help, it still takes humans to write scripts for other humans.

Skills for working with AI as a copywriter rather than against it

Many of the skills that help set you apart as a real life human copywriter as opposed to an AI are also the ones that can help you use AI effectively. It’s AI’s shortcomings that create the opportunity to stand apart. Also, it’s those shortcomings you’ll need to make up for if you’re using AI to help you write copy and content.

The goal is to develop skills you can both advertise to potential clients or employers, proving you’re keeping up with the latest developments in AI and ready to adapt; and skills you can use in your day-to-day work to make you more efficient and productive. 

There is a growing market in certifications and training for AI-assisted creative work. Besides teaching you the ropes, certification gives you a handy, LinkedIn-friendly way to advertise your skills.

Think of AI copywriting in 2023 or 2024 as SEO in 2008. In 2008, SEO was still relatively new. If you could offer clients SEO expertise, you set yourself apart from most other copywriters. Today, it’s taken for granted that every web copywriter or content writer has some SEO experience. In ten years, it could be the same case with AI.

Copywriting prompts

This article provides a rough overview of how prompts affect output, and what factors make a prompt more effective. Taking an online class can teach you the finer points. Some AI copywriting services offer their own certifications and classes—see the Jasper Bootcamp for an example.

If you’re using AI to generate copy, you’re ideally able to come up with an effective prompt right away, with a minimum of additional fine-tuning to get the copy you need. Courses and certifications can help with that, but so can taking the time simply to experiment with AI, and developing your own approach.

Media prompts

As you school yourself in the finer nuances of AI prompts for copywriting, take a crash course in writing prompts for image generators like DALL-E and Stable Diffusion.

Some artificial intelligence copywriting tools already boast extensions for generating images. Working for future clients or employers, you may be required to generate artwork to accompany your blog articles, social media posts, or ads.

Cross-training in image generator prompts will make your copywriting prompt engineering stronger. Plus, there are other types of AI-generated media—music and video, for instance—on the horizon. Before long, you may be writing prompts for video ads, complete with original scores.

Copy editing

The role of copy editor goes back at least to the 19th century and the globalization of the publishing trade. It’s as relevant today as it ever was, but as a distinct skillset, it’s often neglected by content strategists and writers. If you’re planning to use AI to write content, or you want your own work to stand head and shoulders above AI-generated content, good copy editing is essential.

Copy editing is distinct from proofreading, which simply checks for errors in spelling and grammar. It consists of both mechanical editing—making sure published content fits the house style guide—and substantive editing (or content editing)—structuring content to make it readable and internally consistent.

AI doesn’t do either of these well. It takes human skill—at holding both small details and the larger overall picture in focus at once, at understanding intuitively how other humans digest and retain information—to be a good copy editor. 

If you’re using AI to write content en masse, one of your tasks will be copy editing it so it doesn’t read as though it was produced by a hoard of bots. And if you want your work to stand out from writing generated by hoards of bots, you’ll need to know how to copy edit it well.

Community colleges as well as larger institutions offer copy editing degrees and certificates, and you may be able to find online courses.

Deep research

Current copywriting AI is seriously hindered by its inability to look for information outside what it was trained on.

Whether you’re drafting Google ads or long-form content, the more deeply you can research a particular topic, the more effectively you’ll set yourself apart from copywriting artificial intelligence.

On the other hand, if you’re planning to use AI to your benefit, knowing how to pick up where AI leaves off and incorporate high quality sources allows you to flesh out AI-generated work and introduce unique angles and information competitors might lack.

Deep research can take the form of:

• Conducting customer interviews and surveys
• Researching the careers, lifestyles, and worldviews of personas you’re targeting
• Accessing databases, periodical archives, non-digitized books, and other data that can’t easily be found on Google
• Interviewing experts in relevant fields and providing direct quotes
• Joining discussion on social media, Discord, or Reddit

All of these resources fall outside the typical reach of AI.

What’s next for AI copywriting?

It’s hard to say exactly how AI will change the future, or what copywriting artificial intelligence will look like in five years. All we know for sure is that more large language models are on the way. 

As of this writing, GPT-4 is on the horizon. A few predictions from Alberto Romero, an analyst in the sphere of AI, based on the most recent rumors:

• GPT-4 will be able to accept audio and video input
• It will be much larger than GPT-3 (in terms of the number of connections it uses to generate output), but will use them differently, in a way that’s closer to how brains work (only using a portion of its available “neurons” at any one time)
• It may be able to pass the Turing test (during a conversation with the AI, it will be impossible for the average person to determine it’s an AI, and not just another human being)

Current and as-yet-unlaunched copywriter AIs are sure to scramble for access to GPT-4 to power their apps. What’s unclear is how that will affect their performance, and whether they stand any chance of seriously competing with flesh-and-blood copywriters.

One thing is certain: Even if the next generation of AI offers more competition in the field of copywriting, copywriters who learn to use AI to their benefit now will be most able to benefit from new developments.