How We Solved our Copywriting Conundrum with ChatGPT

I’ve spent a significant proportion of my career copy-editing blog posts for non-native speakers of English. And it’s not something I particularly enjoy. While I do like writing my own blog posts (albeit in fits and starts that typically last several years), it isn’t very rewarding to spend hours sprinkling “ands” and “thes” throughout a blog post written by a Czech programmer whose mastery of their technical subject matter is tempered by the Czech language’s stubborn lack of definite and indefinite articles.

This is particularly irksome because I’m running a company and have plenty of other demands on my time. Unfortunately it’s proven to be a daunting challenge to find someone who understands highly specialized subject matter, is a native English speaker and can write. We’ve tried in-house copywriters, external agencies, secret voodoo rituals… but I still spend several hours a week cleaning up blog content, rather than drinking champagne in a hot tub full of supermodels like a proper CEO.

All of this changed a few weeks ago, when I succumbed to the increasingly omnipresent chatter about ChatGPT and gave it a try. I was blown away. My thoughts turned immediately to how this would affect our marketing activities, and on a lark I asked it to write a funny dialog between me and Razi, our indefatigable marketing manager:

Boss: Hey, [Marketing Guy], can I talk to you for a minute?
Marketing Guy: Sure, what’s up?
Boss: I wanted to let you know that we’ve decided to replace you with a marketing bot.
Marketing Guy: What? Why?
Boss: Well, the bot can handle all of our marketing tasks more efficiently and at a lower cost.
Marketing Guy: But I’ve been with the company for [number] years and have contributed so much to our success.
Boss: I understand that, and I appreciate all of your hard work. But the reality is that the market is changing and we need to adapt. The bot can handle tasks like email marketing, social media, and content creation more efficiently and effectively.
Marketing Guy: But what about my job?
Boss: Unfortunately, we won’t be needing your services anymore. We’ll be letting you go and offering you a severance package.
Marketing Guy: This is really disappointing. I thought I was doing a good job.
Boss: I understand how you feel, but this is just the way things are going in the industry. I’m sorry.
Marketing Guy: Okay, I understand. Thank you for the opportunity to work here.
Boss: Thank you for your hard work. We’ll make sure to provide you with all the necessary support during this transition. Take care.

I chuckled and posted the chat in our Slack, tagging Razi. He took it in stride and replied with another ChatGPT-generated dialog between the Boss and Marketing Guy, this time emphasizing the limitations of the bot and the need for it to collaborate with its human counterparts, rather than replace them. This exchange of dueling dialogs continued for some time, to the great amusement of our colleagues.

In the end, Razi embraced the bot and quickly became its biggest proponent in the company. He speaks amazing English but he’s not a native speaker, so he’s been frustrated for years, literally, by the fact that he needs external help to clean up our blog posts. Suddenly he has a tool that performs better than any human copywriter we’ve worked with. It writes perfect English, is an expert on every imaginable subject, is at his beck-and-call 24 hours a day and can even write cogently on the topic of rodent productivity:

I was chatting with one of our suppliers the other day, and he laid out the case against using ChatGPT for blog posts. His argument rested to a large degree on Google’s long-standing efforts to stamp out AI-generated content. In the words of Google Search Advocate John Mueller:

“…people have been automatically generating content in lots of different ways. And for us, if you’re using machine learning tools to generate your content, it’s essentially the same as if you’re just shuffling words around, or looking up synonyms, or doing the translation tricks that people used to do. Those kinds of things.”

In other words, if ChatGPT writes it, it’s spam.

It’s worth emphasizing that creating a good blog post using ChatGPT is not just a matter of entering a prompt then blithely pasting the response into your content management system. Maybe this would work for short, simple posts, but sophisticated results require a lot more hands-on interaction with the bot. Razi told me he spent several hours working through his post on the role of marketing in launching a 3D product configurator. He fed it prompts for each section separately, pointing it to relevant articles where appropriate, and asking it to rework and rephrase when necessary. The result is excellent. I did give it a read-through to be safe, but it only took me a few minutes, and I hardly changed anything.

But more to the point, I would argue that the prose generated by ChatGPT is qualitatively different from anything an AI could have produced 20, 10 or even 5 years ago. With that in mind, maybe it’s time for Google to revisit their policies on automatically generated content, which Mueller says have been in place “since almost the beginning” (by which I assume he means the founding of Google in 1998, and not the formation of the Earth from a cloud of gas and dust about 4.5 billion years ago). If an AI produces a post that is as good or better than most of the human-generated content on the web today, in what way is that spam?

We’ve probably watched too many Terminator movies. After all, seemingly rational people apparently worry about a superintelligent AI coming back from the future to torture them in their sleep. However, the immediate rush to put negative spin on every advance in AI strikes me as counterproductive. It is still highly speculative that we will ever create machines with true consciousness, or virtual simulations that are as rich and detailed as the world we live in. But that’s a topic for another post. Suffice to say that it strikes me as at least as plausible that future AIs will be our benevolent servants, in the mold of Star Trek’s ship computer or Tony Stark’s J.A.R.V.I.S.

Those of us hailing from the Anglo-Saxon world forget sometimes that not everyone speaks English. And not everyone with great ideas can write well, even in their native language. If ChatGPT allows a broader slice of humanity to say interesting things in perfect prose, I say more power to it! After all, one person’s spam is another person’s delicacy.