It's easy to generate bland dribble with generative AI tools. No more!

This short essay will show you how to consistently generate rich and unique output from large language models. This technique can be applied to many domains, but none show how effective it is than writing marketing copy.

The Baseline: Mediocre and Uninspired Copy

Let's pretend that we are selling a new blue lipstick. Asking ChatGPT to "write marketing copy for a blue lipstick" comes up with:

• A statement, not just a shade.

• Vibrant and long-lasting.

• Inspired by deep sea magic.

This is not going to make much of an impact.

The Secret: Ask ChatGPT to List Interesting Facts First

We saw yesterday that due to the transformer architecture, ChatGPT looks back at its own output when generating new text, something called "autoregression". This simply means that we need ChatGPT to generate text that will help it generate better copy!

What is easier? Come up with good slogans out of nowhere, or come up with good slogans after getting a list of striking details?

We should first ask ChatGPT list vivid facts about our blue lipstick:

• Celestial Midnight: Deep cosmos-inspired shade.

• Iridescent Finish: Shimmers like twinkling stars.

• Oceanic Undertones: Depths of the deep sea.

Generating interesting copy is now a breeze! Here are a few examples.

• Midnight Meets Ocean: Depth redefined.

• Nebula Nuance: Stellar shade, sea-depth subtlety.

• Ocean Echo, Star's Shadow: Where deep meets dazzle

The Takeaway: Prompt for Intermediate Output

We will encounter many variations of this technique over the next few weeks, but its structure is always the same:

• Ask for intermediate information before the actual answer

• Make sure ChatGPT generates high quality intermediate information

• Enjoy the final result!