In his latest post for B&T, Time Under Tension co-founder Tim O’Neill explains why, despite all the hype, generative AI isn’t quite smart enough to do marketing.
In the past 18 months since we started Time Under Tension, I have met with hundreds of marketing people, and the common theme is this: they want to (need to!) do more with less. They have been told that generative AI can help, but they are only scratching the surface of what’s possible.
The reason for this is simple: it’s not simple.
It’s really easy to get started (e.g. ask ChatGPT to write some content) but, with the current pace of change it’s impossible to master.
New AI chatbot models are being released every month, each with their strengths and weaknesses. Without a lot of experimentation, it’s difficult to know which AI chatbot is best for a given task (e.g. I find Claude and Gemini have a more natural tone of voice than ChatGPT for creative writing).
Google recently released The Art of the Prompt, a 54-page guide to best practice prompting. It’s a great guide, but I have never met a marketer who has the time to study a 54-page instruction manual.
Best practice data security is possible, if you know what to do. But it’s complicated if you’re using multiple tools (e.g. ChatGPT, Gemini, and Claude all treat data security differently, let alone the image models like Midjourney).
Let me give a personal example.
Yesterday I was set the challenge of building a gen AI tool that would write natural-sounding LinkedIn Posts, mimicking my writing style.
Yes, I could get a 50 per cent good result with a simple prompt in chatGPT. But to get to a 90 per cent good outcome (I was always expecting to add 10 per cent of polish by hand) required experimenting with three different AI chatbots, experimenting with prompts, and experimenting with multiple AI review steps.
Two hours later I completed the challenge with a tool that works really well* (and is reusable). Very few people have the experience or the time to persevere, and many make the assumption that generative AI will never be able to do that particular thing.
With these complexities, it’s not surprising that marketers are getting some quick wins, but generative AI is not revolutionising their business like I believe it can.
So what is the solution?
I suggest you start with a problem you wish to solve. What are frequent tasks that are completed within your team? Choose one or two, and spend some time researching and experimenting with gen AI tools until you get the best possible result. You might spend three hours to find a good process, but this might then save you three hours every week from now on. If you get stuck, get in touch and I will be happy to help!
Once you have nailed down the process, all of the main AI chatbots now have ways of saving prompts for reuse: ChatGPT has GPTs, Gemini has Gems and Claude has Projects.
Do this three times over, and you will now be close to spending the recommended (by AI researcher Ethan Mollick) 10 hours of experimentation, after which you will be comfortable with the benefits (and limits) of AI chatbots – “No amount of reading and research can substitute for spending 10 hours or so with a frontier model, learning what it can do.”
In some cases, an AI chatbot won’t be the right solution, and there are thousands of single-purpose AI-powered tools that can help with a specific use case. For example, I now have a LinkedIn Post writer 🙂.
With some experimentation and perseverance, you can work wonders, and start to really get the best from this amazing technology.
Please share with us what you build!
* I ended up using Anthropic Claude 3.5 Sonnet, with multiple steps of review, and providing an example LinkedIn post of mine as a style guide. This was built using Peak, our SaaS product for marketing teams.