Image generated with DALL-E in July 2022 vs. image generated in Midjourney August 2023, using the same prompt.
A year ago, we embarked on a journey to let AI build the latest version of the Buzz Radar website. This was before the surging popularity of apps like ChatGPT and Midjourney had dragged AI into the spotlight; it was a cutting-edge idea at the time and something we were only prepared to undertake as we had years of experience with machine learning and AI.
It's astonishing how quickly the world adapts. What seemed like an out-there experiment a year ago now feels completely commonplace. Truthfully, we’re probably not far away from the point where using generative AI programs to create a brand’s web presence will be the default rather than the exception.
So, with a year gone by, we've decided to give our website a mini-refresh to see how AI tools have advanced in that time. The results were fascinating, revealing what has moved forward and what hasn't. Here's what we learned this time around.
AI Image Creation Has Massively Improved
Last year, we were amazed that we could create images using text prompts with DALL-E2, but it was very hit and miss. Sometimes you'd strike gold, and other times you'd spend hours lost in the uncanny valley, trying to create images that were both realistic and met your needs. Our designer would spend hours trying to correct the strange eyes and crazy hands. Often, it was quicker to simply create an illustration from scratch than endure the painstaking hours of tweaking prompts and adjusting variables that would eventually lead to an AI image that did the job we needed.
Shortly after the launch of the website the game changed dramatically with the launch of Midjourney V1. Producing much higher-quality images, it has since become a go-to tool for many of our team. It's now on V5.2 and has a number of hugely useful features that make creating relevant images incredibly fast and efficient.
Our previous DALL-E-generated header image vs. our new Midjourney header image.
Using AI-Generated Copy
Last year, our aim was to retell our story based on a data-driven understanding of our audience using Chat GPT-3, which had, at that point, just emerged and needed to be accessed via an API. It provided some great ideas and thought starters, but the results were frequently bland and needed quite a lot of reworking from our copywriting team to make them really sing. A year on, and ChatGPT has quickly become an industry-standard tool with almost endless applications. This has its drawbacks, however. When everyone uses ChatGPT, everything sounds like ChatGPT and we’ve found that we still have to work with it to create fresh-feeling copy. Some even started to suspect that GPT-4 was somehow getting worse as it ingested the bad habits of its users. Things improved when we created our own copy creation tool and added custom instructions so the AI learnt our tone and style. While it still requires human polishing, we were able to created updated website copy much faster this time around.
Connecting the Dots Between AI Tools
One of the original challenges we found was the lack of interoperability between these tools. This year, that's improved considerably through our own development efforts connecting tools and datasets, alongside the fact that AI developer platforms are far more well-rounded now. We now use AIs that can communicate with each other and our data, reducing the manual effort needed to 'connect the dots.'
Audience Intelligence is key in creating content that resonates
Our proprietary Audience Intelligence platform continues to be the cornerstone of our strategy. This year, we delved deeper into psychometric profiling, allowing us to tailor our content even more precisely to our target audience. The insights gained have been instrumental in guiding the AI to understand our audience and create content that resonates with them.
You still need human creative to stand out
While AI has come a long way, the 'Human in the Loop' (HITL) approach remains crucial. Our team of strategists, creatives, and coders still played an indispensable role in fine-tuning the AI-generated content and ensuring that it aligns with our brand messaging and audience needs.
So what have we learnt?
The age of AI has unarguably arrived, and there is no longer a question about the capability of these tools: they’re sophisticated, easy to pick up and have almost limitless applications. The question now is how to use them strategically, how to understand their drawbacks and how to integrate them. This is now the present, rather than the future, of content creation. The next frontier could very well be AI-generated strategies, where machine learning algorithms analyse vast amounts of social data to recommend comprehensive marketing plans.
AI technology is evolving at a breakneck pace, offering unprecedented opportunities for efficiency and personalisation in digital marketing. These are fantastic tools, but they are still only tools: humans remain essential for strategy and creativity … at least for now. With many of the useability challenges either solved, or on their way to being solved, the bigger concern is ensuring that data privacy and security are maintained when undertaking AI projects. Brands that want to take advantage of AI without sending sensitive data offsite still risk hitting roadblocks in trust and compliance. We managed to do this, but only with the help of our expert in-house engineering team … but that is a subject for another blog.
So, how effective were our updated AI-led efforts? Judge for yourself by visiting the new Buzz Radar website!#ArtificialIntelligence #SocialIntelligence #DigitalMarketing
Published on 2023-09-06 10:12:54