The introduction of ChatGPT-4 suggests that generative AI could get closer to meeting our expectations in terms of quality. The latest iteration offers the ability to create ‘Customised GPTs.’ These are essentially more customised content engines that can be taught to replicate the unique voice, style, and objectives of brands and authors. By feeding it previously examples of articles, blogs, press releases etc. and brand guidelines the AI can learn templates and attempt to get closer to required outcomes faster. At least this is the promise.
Are customised GPTs any good?
We’ve been building a bunch of customised GPTs to test their ability to create press releases, CCGroup-led insight reports, CCGroup-driven blogposts and replicate some of our reports.
How effective you view customised GPTs depends very much on the expectations you had to start with. Are they ready to replace human involvement and effectively clone your best writers to instantly recreate their work? No. Do they create a more useful starting point than was previously possible with early versions of ChatGPT? Yes. Do these more useful starting points reduce the amount of time you’ll spend creating said content? Probably.
The more specific the focus, the better the outcome
Customised GPTs, like their previous, more standard predecessor will still miss the occasional nuance or misinterpret context. Not only do you still need careful human oversight to get the best result, whoever is creating the content needs to have a fairly clear idea of what they want to say and why, and a strong grasp of the technical and market context, to be certain the GPT is sufficiently accurate and appropriate.
We found that creating a specific GPT for a specific task delivered a better result than a more general approach. What do we mean by this? Well, say we wanted to create this blog using a customised GPT, we would create a GPT specifically on the opportunities and limitations of ChatGPT4 in creating content in PR, rather than creating a more generic CCGroup blog generation engine that could create a variety of blogs on different topics.
How worried should we be?
I’ve said before that I’m something of a PR purist and believe in the sanctity of human creativity. There have been several instances where I’ve heavily edited generated content to make it sound more like me. Plus, I’ll always prefer to originate my own content on issues and trends that are personal to me. Not everyone will feel the same way, I know.
The more I think of it. The more pervasive generative AI becomes, the more the marketing world will crave authentic authors and experts. I therefore feel very grateful to be testing AI on topics and trends I am expert in to test its effectiveness, rather than using it to try and make me sound and look more accomplished. I think it’s inevitable that we’ll become more and more reliant on generative AI for administrative content. The more we use it, the better it will get, so we’ll continue to use it more, for more things.
Never underestimate the human propensity to cut corners, especially if it becomes nearly impossible to detect. But then the counter argument to this is not to underestimate the human tendency for procrastination. In my opinion, customised GPTs create a more useful way to reconcile the two things and help busy minds to organise thoughts and create better, authentic outcomes more consistently and faster.
Hence why we’ll continue to look for ways to sensibly integrate it into our workflows.
