AI in social media

Image by Tara Winstead, Pexels

By Anja

How do you use AI? It’s become an honorary member of social and marketing teams almost everywhere. It’s embedded in Canva, where Reels, text and art can be created with unique prompts, as well as Hootsuite, where text generation can be used for new content. Generative AI is available on social media platforms too, where LinkedIn now encourages users to “Start a post, try writing with AI” at the top of the main feed.

What we know for sure is that AI use will skyrocket this year. Interest was so high that from 2022 to 2023, topics on learning about AI increased by 550%, according to an analysis conducted by Hootsuite of over 15,500 news articles and blogs.

Is AI human-sounding enough?

Use ChatGPT a few times and you might notice there are some things that are missing. Things that make for truly human writing, like colloquialisms, slang, abbreviations, quips or witticisms, and perhaps most importantly in the work we do, nuance, are all absent. What you will find a lot of is hyperbole. Because ChatGPT is used by so many for marketing – and is essentially an amalgamation of everything it is prompted to do – most of the text it produces sounds like a really wordy sales pitch. This means that ChatGPT produces text that implies that everything is amazing (or, a ChatGPT favourite, a “gamechanger”). Here’s an example of what ChatGPT produced when asked to provide a blog intro about AI in social media:

“From the heart of vibrant South Africa, where innovation meets diversity, we delve into a captivating exploration of the burgeoning influence of Artificial Intelligence (AI) in our social media landscape. As the rainbow nation embraces the digital age, join us on a journey through the dynamic intersection of technology and culture, as we uncover the transformative role AI plays in shaping our online experiences.”

Vibrant, captivating, burgeoning, dynamic AND transformative. That’s a lot of adjectives.

It’s best to think of generative AI as your overly enthusiastic, newly-instated office assistant. One that can synthesise, simplify and guide a social media strategy or communications strategy. It’s a great tool to have, but it will generally produce generic, unnecessarily zealous content if it isn’t prompted clearly, toned down a bit, and then adjusted for your audience and made purpose-specific. There’s also this version of the blog intro that I prompted ChatGPT to write using distinctly South African colloquialisms.

“Howzit folks! Pull up a chair and let’s chat about a wild ride happening right here in Mzansi – the rise of AI in social media! From Jozi to Cape Town, AI’s flipping the script on how we do things online. So, grab your biltong and let’s dive into this tech wave that’s got the whole country buzzing!”

Biltong and stereotypes aside, it did get one colloquialism right, but assuming that the whole country is currently buzzing about AI would be a bit off the mark (at present, Home Affairs is actually trending on X).

Dos and Don’ts

Throughout our discussions about AI at Pomegranite, we’ve prioritised authenticity and credibility. If you recall some of the social media posts you’ve seen from really well-rounded South African brands, they probably have a well-defined voice, far from the cookie-cutter, impersonal tone often associated with automated content. Here are some basic dos and don’ts when using ChatGPT that will help with creating good content, while fostering a genuine connection with your audience and their needs.

Do:

• • Remove sensitive information from your ChatGPT briefs (this a warning from the ChatGPt software itself).
  • Prompt or “brief” ChatGPT with as much information as possible like the tone you’d like to see and the purpose of the text.
  • Cross-check information from ChatGPT. This includes dates, names, and the information embedded in the answers that ChatGPT generates.
  • Use ChatGPT to save time. If something can be automated, like typing out lists, short paragraphs or explainers, then go for it.
  • Use it to kickstart creative ideation. Even if you aren’t happy with the answers you see, you can use it as a starting point.

ChatGPT makes assumptions about the information you provide it, and will then generate answers based on these assumptions. This can range from incorrectly assuming the meaning of acronyms, all the way down to deeper, implicit bias.

Don’t:

• • Copy + paste from ChatGPT and leave the work as is.
  • Rely on ChatGPT to provide historically accurate or impartial information.
  • Expect brand new creative content ideas.

This one may be common sense but I’d like to add that you should not let AI put together an entire event for you (like the recent Willy Wonka debacle that was “inextricably tied to AI”, with the whole premise, script, marketing materials and very minimal visuals all seemingly AI-generated).

Bias in AI

While AI has clear limitations, it can be incredibly helpful. Here’s an image that was generated by Open AI’s Dall-E website for a Pomegranite blog I wrote in 2023.

I was really impressed with this, but had to explicitly ask for “multiracial hands” holding the plate, or the hands generated were all white, every time. Artificial Intelligence mimics human bias, which includes historical and current social inequalities. In order to prevent perpetuating this bias, AI prompting might need to be very specific, at least until machine learning reflects more equitable systems in the real world.

AI at Pomegranite

Amidst the jokes made about using AI to write our AI policy (we didn’t), we’ve used our policy to outline our commitment to transparency. The policy takes into account our values of honesty, efficiency, professionalism and providing a personal experience. We drafted it with the purpose of being a “living” document – one that’s updated frequently and reflects how AI usage will rapidly change as it evolves. Our focus is on creating work that is distinctive, impactful and well-thought out. Ultimately, we want our clients to feel heard. And for that, you definitely need a human.