GPT-3: Jack of all trades, master of none

No, artificial intelligence isn’t coming to take your copywriting job. And here’s why.

Photo by Markus Winkler on Unsplash


Is artificial intelligence making you nervous? If the conversations that I’ve had with copywriters in recent months are anything to go by, many of them are concerned about the impact of AI on their careers.


Anxiety levels shot up last week with the latest reviews of GPT-3, the new open source language model. I’ve read extraordinary claims for the software: As well as crafting spreadsheets, it can write poetry, business memos, even news articles.


Shouldn’t content marketers and copywriters be worried?


I don’t think so. GPT-3 is quite good at lots of things. It excels at some repetitive tasks. But it lacks the essential human qualities for copywriting and storytelling. All the learning data in the world are no substitute for a sense of purpose, or the empathy required to connect a business with thousands of readers.


Or as Wired magazine puts it:


"GPT-3 often spews contradictions or nonsense, because its statistical word-stringing is not guided by any intent or a coherent understanding of reality."


What does this mean in practice? If you feed GPT-3 the opening lines of your article, there’s a good chance that it will generate a paragraph or two of coherent prose that builds on your initial premise. But stretch it further, and it quickly loses the plot – in every sense of the word.


That’s why you need a ‘human in the loop’ to rein in the software as soon as it starts to stray. You can actually see this at work in Adweek’s review of the software. GPT-3 was used to write every other paragraph of the article, while a professional journalist wrote the rest. The results are impressive until you remember that every human paragraph ‘resets’ the software and pulls it back into line.


That said, I can see two areas where GPT-3 and other AI tools have a role to play in copywriting and content strategies.


Less brain, more brawn


Firstly, at the ‘brain-storming’ stage where you’re still searching for headlines or even a mission statement for the project. GPT-3 doesn’t suffer from writer’s block and can throw out statements non-stop, some of which will inspire the room.


You’ll still need human participants to curate the output, but I can see how the technology could inject energy into the debate and even steer the group towards some kind of consensus.


The second area applies to data and reporting, especially when it comes to digital advertising. AI can find connections between copy and visuals, advertisements and audience that help inform future campaigns.


Software houses such as Bidalgo can analyze the client’s advertising history by copy, images, colours and concepts. If a particular word works better than another, the AI recommends it to the designer along with other creative elements.


Stop worrying, start learning


The advice I give to marketing clients and trainees is to find out how AI can boost your copy career. Then learn how to harness these tools to your advantage.


You can see analogies in other industries such as journalism. The reporters getting ahead are those that augment their core human skills with AI tools. For example, using software to sift through thousands of documents has helped reporters to break front page news, such as the Panama Papers scandal.


The Radar News Service, in the UK, is another good case study. Its journalists use AI technology, to draft articles based on data sets released by the UK government. Using their investigative skills, the journalists identify data from which they can derive a story and then build a template into which the data and standard phrases can be assembled.

For journalists to flourish in this environment, being able to work with automation templates is essential. Some simple programming knowledge is also required. Skills such as these can also help copywriters and content marketers remain relevant and rewarded.

In a future post I’ll look in more detail at the AI skills required to flourish in the 2020s. Until then, stop worrying about GPT-3. It’s great at processing massive amounts of data and generating standalone copy. But a creative campaign, with all the headlines, blogs, social posts and captions threaded together in a coherent narrative? That's still a long way away.

Feel the connection?

© 2020 Peter Springett.