Can AI do copywriting? (And would you want it to?)

It’s feels like the biggest thing in the tech space since the internet– AI has been part of so many conversations with clients and contacts recently, and the big question is: Can AI do copywriting? I’d argue the bigger question is, would you want it to? 

Of course, generative AI has been built to create things. But whether you would want it writing the final draft of that blog, client report or case study is an entirely different matter. I’ll get into the details of what we know about how AI performs later in this blog, but first it’s worth understanding how it actually works. 

How does generative AI copywriting work? 

Fundamentally, generative AI is making a next best word guess whenever it writes something, it’s essentially a predictive text machine. The large language models (LLMS) that the AI has been trained on were fed mountains of content from the internet (fairly controversially, as it turns out) to help them “learn” what word should come next. Words are reduced to numbers within the algorithm and spat out based on the logic it has created.  

What this means is, whenever you get AI to write something you’re getting predictable content. If I read another article which contains, “In a world where…” “It’s important to note…” and “In conclusion…” I’ll be borrowing a rusty spoon to remove my eyes.  

Do you like reading predictable content? Does it make you want to read more? Would it make you think the writer is an expert? Clearly as head of a team of content strategists and copywriters, I have my own view; you can make your own mind up.  

But does it really matter whether people love your AI-written content if it draws in more traffic for your website? Good question.  

How does content written by AI perform? 

I’ll defer to SEO guru Neil Patel to answer this particular question. His team took 68 websites and created 744 articles. Half were written by AI and half by humans. The content was created to target similar keywords and difficulties, and was comparable in length to make sure the results were objectively as fair as they could be.  

Neil wanted to know how the content would perform to test whether it’s really worth getting AI to write web content.  

So what happened? Five months after posting all the content, AI articles were generating an average of 52 visitors a month, with articles written by humans pulling in 283. So far, so comprehensive. But what if you’re only bothered about the time investment for getting traffic and would be happy to just churn our five times more AI articles to get the same amount as the human content was attracting? 

Neil makes two points about this. The first, is that low-quality content damages the overall effectiveness of your site in SEO terms, impacting traffic and rankings. The second uses the time they spent creating both sets of articles and compares that to the traffic attracted. Although humans took an average of 69 minutes to create each article versus AI’s 16 minutes, human content still pulled more traffic per minute spent (4.1) than AI (3.25).  

You can read all of the details about Neil and his team’s experiment in his blog. It’s the best example I’ve come across which uncovers the real impact of using generative AI to write content for your website. 

But what if you want to use it to write other things? The Wordsmiths team crafts everything from 20,000-word reports on mental health to emails taking customers through a journey to get information, say if they signed up to know more about a housing development. None of this is written to perform online, so what would be the risks of writing those types of content with AI? 

For me, there are a few. The biggest is still the lack of original thought. AI can’t discuss a range of angles and hooks with the client to help you come up with the most impactful way to pitch information. It can’t apply lived experience to writing about anything, and it handles emotional topics in a clunky, robotic and obvious way. (Just read copywriter Eddie Shleyner’s account of seeing his child born and what AI offered as its own version.)  

A skilled team of copywriters and content strategists can add so much more value than the words we write for you. Discussing a simple proofing and editing project with a contact the other day, I explained how we could  

• contribute better by working with them to create an editorial calendar 

• write the articles their (non-marketing) colleagues didn’t want to 

• interview key industry figures to create valuable thought leadership articles which could be used both shortform in a newsletter and longform on their website 

• provide some social media content to support promoting those articles 

You might think that’s a clever up-sell. But that’s not what I intended. For me, it was about making the most of the client’s investment by thinking carefully about writing the right content and making it work as hard as possible for them. Sure, AI could have answered a question with a similar plan, but you would have to think about asking it the right question in the first place.  

Will AI kill copywriting? 

Honestly, I don’t think it will. There have always been cheap ways to get content written if you weren’t overly worried about quality. People Per Hour and Fiverr have never been competitors for Wordsmiths as our clients rate the way we add value as part of their virtual team just as much as the outputs we create for them.  

I can see how AI can be a valuable tool for research and for when you just need to not be staring at a blinking curser on a blank screen, but I’m not worried it’s going to make us all redundant at this point. I can’t predict how it will develop in future and there’s no doubt it can become more sophisticated, butI see it in the same way as automation. It took over the mundane tasks to leave more interesting, higher-quality roles for people rather than striking a large chunk of jobs out of the market entirely.  

Do AI detectors work? 

Our eldest, August, is 17 and gearing up for A Levels then heading off to university next year. There has been plenty of controversy about what AI means for universities and lots of them went heavy on using AI detectors to weed out the students looking for a shortcut to doing their assignments. But the results have been mixed and universities seem to have backed off relying solely on more technology to spot the use of technology.  

I have a personal example. August spent the summer working on a piece of history coursework. We had discussed their paragraph plan, and I’d read versions of it as they wrote it, chunk by chunk, over a series of weeks. Just before submitting it, August decided to run it through an AI detector. I don’t know why (because all that awesome work is now in an LLM somewhere) but the results were shocking. It said the essay was 94% AI-generated. And yet not a word of it had been written with AI. For good reason, it seems, I’m as suspicious about AI detectors as I am about using generative AI to write. 

(As an aside, do you think the above two paragraphs would have been written by ChatGPT if I had posed the question about whether AI detectors work? #JustSaying) 

What does ChatGPT say about whether it should do your copywriting? 

So, can AI do copywriting? Well, what does ChatGPT think? We asked it whether AI can match the skills of an experienced copywriter. It said:  

“While AI can handle many copywriting tasks effectively, it cannot fully replicate the depth, creativity, and strategic insight of an experienced copywriter. The best results come from a partnership where AI handles routine tasks and idea generation, and skilled copywriters apply their expertise to refine and craft content that truly connects with audiences”.  

So, while AI can do some of the work of a copywriter, the question of whether you would want it to remains. But I’ll leave that decision up to you. I’m sure it’s no surprise I’m firmly in the not on your nelly camp, so if you want a chat about how our lovely team of humans can write content for the humans you want to read it, you know how to get in touch!