If you’ve been keeping up with ITS social media for the past week, you may have noticed a certain new flair. After a month of training and fine-tuning, our AI-powered assistant, Maizey, embarked on its first mission. Alongside a few tweaks to the AI settings, the trial of implementing this tool into our workflow began. Though it became evident that the tool is only a supplement to our actual social media specialist, Tommy Valdez, we gained valuable insights into how this tool can increase efficiency and provide a new lens of creativity to what is already being created.
Prior to the inauguration of this tool to our social accounts, there was a necessary update to ensure the accuracy and efficiency of the chatbot. Per the advice of Emerging Tech Analyst, Ben Andries, I updated the chat settings from Conversational Retrieval Qa to Retrieval Qa. This setting change meant the chatbot didn’t remember the prompts while in the same chat, and though I could no longer ask follow-up questions, the speed and value of responses were greatly increased.
With improved productivity and trial runs complete, I decided it was time for Tommy. I added him to my MCommunity group, published the Maizey, and waited around 20 minutes for it to update fully. He started with a simple prompt I’ve shown him: ”Give me a content schedule for the next week for each platform”. But, to our surprise, it returned an obscure message saying, “The data provided does not include a forward-looking content schedule.” I had never run into this issue while testing the Maizey, so I once again turned to Ben for ideas on how to navigate this. While awaiting advice, I too tested the prompt and, interestingly, it returned an analyzed content schedule to me immediately. With some back and forth between me, Tommy, and Ben, we were able to resolve this issue by slightly changing the prompt and opening a new chat. Though a minor setback, the process exemplified the iterative nature of GenAI tools, showing us that achieving comprehensive responses is heavily dependent on the user’s ability to work with the tool.
With the amended settings, Tommy embarked on a day of deploying Maizey’s AI-driven content suggestions. By asking the Maizey bot to develop a content plan based on the audience and performance summaries provided, the plan paralleled what our audience is likely to resonate with.
The content plan suggested by our Maizey included ideas that became the baseline for nine posts throughout the week. On three platforms–Instagram, LinkedIn, and X–Tommy followed Maizey’s subject recommendations. The content schedule drafted was based on analyses of typical content themes for each platform and the past performance summaries provided. Through guidelines like this, Tommy’s social expertise and artistry were still very necessary. Yet, he was able to invest more time into this as the time spent combing through the past posts and analyzing metrics was reduced to how long it takes the Maizey to load a response.
The post that garnered the best engagement from the recommendations was from a LinkedIn recommendation that called for “a detailed post about job opportunities or career advice for students in tech.” Though this post did better than the others, there wasn’t a notable increase in engagement from those not AI-informed. Despite this, Tommy spoke on how it offered a different perspective on content strategies. But, as always, social media is heavily subjective. Both Tommy and Maizey acknowledged that metrics are affected by various factors, especially given the seasonal engagement fluctuations of college posts.
Looking forward, as Maizey becomes increasingly intertwined with ITS’s social media, we anticipate an enhanced degree of customization in our campaigns. As we post more and consistently upload metrics, our Maizey will continue to improve. In the coming weeks, I intend to focus on how the data quality can be optimized and plan to experiment with the system prompt and other AI settings. With more to come, be sure to keep up with ITS Socials, and see if you can spot the next AI-informed post.
