The AI-lephant In the Room

A person wearing a white jacket who looks futuristic

So many marketing teams we speak to are dabbling with using generative AI in their work. And their attitudes to it vary from the slightly secretive whispered admission to the loud and proud, “Yeah I get AI to do everything for me!”. I think the former reaction stems from the act of using it feeling like “cheating”, or perhaps not being completely confident in how they’re using it. And the latter worries me a little, being brutally honest, as it has a feel of lacking accountability.

As a self-confessed tech nerd, I like to test these things from time to time to see how they work out; here is what I’m seeing:

Copy and content writing

What did I test?

I’ve experimented with the paid and free versions of ChatGPT and Gemini, plus have created my own GPT and fed it everything I’ve written in the last couple of years (roughly 500 A4 pages) so it can learn my style. This is not a comprehensive test of the whole gen AI market, but more a targeted sampling to give you a feel for what’s going on.

Observations

  • My own GPT’s output is better style wise than either ChatGPT 3.5, 4 or Gemini free or paid. But it can struggle/default to average when it comes across a topic, format, or approach it hasn’t got referenced in its source documentation.
  • Personally, I find Gemini less responsive to my prompts, but maybe that’s because it’s newer to me and I don’t fully understand how to ask it in the right way.
  • The paid versions of both tools are an improvement on the free ones, but neither make me think “I’ll never hire another writer”.

What it’s good for

  • Writer’s block. If you’re having a hard time getting started or you’re someone that gets the jitters sat in front of a blank page, these tools are a great way to start. If I’m feeling uninspired, I’ll often get it to write me an outline or even the whole thing because the writer in me then gets fired up to improve the writing, correct everything, add things it hasn’t thought of, and often completely overwrite the entire thing.
  • Organising your thinking. If you’re the kind of person who has millions of ideas and points but struggles with structure, it’s pretty good at sorting through your brain dump and putting some manners on it. That doesn’t mean that you don’t need to edit the hell out of it afterwards, but it can help you see the wood for the trees.
  • Making sure you don’t miss the obvious. If you’re an expert in your field trying to write approachable entry-level content, you’re highly likely to skip some fundamental, basic stuff that you assume everyone knows. The AI isn’t an expert, so it tends to surface things that make you go “Oh yeah, I probably need to explain what that is…”
  • If you’re not a writer and you have to write things because you have no support. If you’re working in a second or third language, or perhaps you’re just really challenged by written English for reasons of neurodiversity or natural inclination, the AI can help. If what the AI produces is better than what you can do by yourself, go for it. But given the limitations outlined below, I’d prioritise finding some human help, whatever form that might take, a colleague, an agency, a freelancer, whatever’s possible for you.

Where the limitations are right now

  • Both ChatGPT and Gemini have a bit of a “house style,” even when you prompt them to write a certain way, they tend to fall back on similar vocabulary and phrasing. Chat GPT in particular has a bit of a thing for using 50 words when 10 will do and stating the blindingly obvious to start a piece.
  • They both get stuff wrong, although Gemini tries to correct this with its function to cross reference Google, it’s only checking whether the same claims exist elsewhere. So basically, if some other doofus made the same mistake and published it, it thinks it’s legit.
  • They have no new ideas or content. Literally none. They might rephrase an old idea, but you’re getting no originality. Remember they’re language models, not sentient, thinking creatures. They can’t infer one thing from another, share unique perspectives, provide real-world examples from their own experience, tell stories, and their analogies generally suck.

Image creation

What did I test?

I took a look at Midjourney, Dall-E via Canva, Shutterstock which leverages a few models including Dall-E, and JumpStory a startup making waves in the space. As with the copy and content tools, this isn’t exhaustive, it’s more of a sample of the ones I hear about most often. I used two prompts, one requesting an image representing the use of AI in marketing and one requesting a modern logo for a drone photography company.

AI Generated Drone Logo
Is it a drone or a car?

Observations

  • All tools performed better on creating images than logos
  • None of the tools that included text in their output got it right
  • Overall, the output was indistinguishable from standard stock imagery, except where it was obviously worse
  • All images were very predictable tropes of what had been asked for
  • Overall image quality was best from Midjourney, although having to use it via Discord is annoying and potentially limiting depending on your company’s IT policy.
  • Almost nothing I got out of the other tools tested was good enough, even for a low visibility/low priority piece.

What’s it good for?

  • If you need a supporting image for a blog or social post and can’t find anything on the stock libraries that fits the bill, this might help you get closer.
AI generated drone image
Nice try AI, but it's not going to win any design awards

What are the limitations?

  • It will probably take hours to be even vaguely on brand. You’re totally at the mercy of what the tool thinks is right for your prompt and refinements are slow and painstaking.
  • You can’t copyright anything you produce unless you substantially alter it yourself afterwards and even then, you can only protect the human element of the work.
  • Much of the quality of what is produced is terrible, both graphically and with type.
  • Outputs are so predictable it hurts, none of my drone photography logo suggestions veered away from a rendering of an actual drone and 95% of the AI in marketing images featured humanoid robots.
  • Logos (as well as being rubbish) were all overly complex, impossible to reproduce effectively across different media, and wouldn’t be viable for use in the real world.
AI Generated Image
Is this an image of AI in marketing, or a promo for the next Terminator movie?

Overall Verdict:

Generative AI is an amazing technology, but still very much in its infancy. I have no doubt it will improve and become more useful for marketers. There is some value in the text tools, but I’m not hanging up my quill just yet. The image generators are almost universally awful and to my mind should be avoided, with the possible exception of turning to Midjourney when there’s really nothing you can find in the stock libraries.