Artificial Intelligence is everywhere, whether you realise it or not.

Customer service, emails, ad copy, blog posts, social media posts, audio and even video content are all now saturated with AI-generated content.

And as a marketing trying to automate and improve efficiencies both personally and professionally, I’ve been trialling a number of AI platforms to experiment and see if there really is a golden bullet.

What did I try to create with AI?

Here’s a brief summary of some of the things I’ve attempted to generate with AI support:

• Blog posts
• Social media posts
• Microsites
• Marketing campaign proposals
• Paid ad copy
• Lead magnets
• Personal goals
• Youtube video scripts
• SEO keyword lists
• Sales emails
• Sales call scripts
• Performance dashboards
• and more…

You can see that a large portion on the outputs were playing into the traditional strengths of a Large Language Model like ChatGPT which is: copy.

A few outliers, like performance dashboards and SEO keyword lists were taking advantage of Claude’s ‘Artifacts’ and Perplexities Research capabilities, respectively.

But in all cases, there were a few very important lessons…

Three Key Lessons

Lesson 1: AI-generated copy is recognisable

If you open up a ChatGPT, a Claude or a Gemini and ask to get a fancy new blog post –even if you’re following a solid Prompting framework — you’re immediately going to notice writing techniques and expressions that will often be repeated throughout your content production.

Some common themes for me include: beginning every article or social post with heavy exposition.

In the modern age of [[insert topic]]

Or a sheer, and unadulterated obsession with bullet-points.

Or.. this tendency to start a point, and never finish it. There’s usually always a certain lack of cohesion as you navigate from paragraph to paragraph.

Some prompters will argue that the key here is: provide initial writing content, including style guides and examples of your writing to emulate, and writing techniques to avoid. But not everyone has the time nor the resources to upload a load of training material when they just want to get a blog post out and be done with it.

Lesson 2: LLMs have their strengths and weakenesses

This seems like common sense, but it’s worth repeating: different large language models have different strengths and weaknesses.

Some models are better at interpreting complex questions that others, handling content writing, or being creative.

But just as important as experimenting is our next lesson…

Lesson 3: Iterate and improve

My advice here is simple: experiment and have fun with the different models. Draft a prompt then execute it within different LLMs to see which one gives the best output.

With every output from an LLM, you have an opportunity to learn and reverse engineer what happened.

Be extremely intentional with how you modify and improve each message from your LLM, and make it a part of your usage.

Never think of your first prompt output as the final — it’s a draft, and should be thought of as such.

Read it back: are there errors? Is it obviously AI? Then instruct the LLM on how to adjust and improve.