How to create content that is *actually* unique using GPT3
We've spoken with nearly 100 marketing teams over the last few months. While mostly all are excited about the prospects of using AI to create personalized and compelling content, one strong concern has emerged -- if everyone uses GPT3 won't everyone's content be the same?
Most of these teams are focused on using GPT3 for organic search, i.e. Google Search Engine Optimization (SEO). They want to deploy a strategy called *Programmatic* SEO which means publishing unique, high-quality pages at scale using a template and a database. The overarching concern is that Google will penalize pages written by GPT3 since the content would be duplicative of content elsewhere on the Internet.
(As a side note, GPT3 can be used to create all types of marketing content, from landing pages to custom emails that power email marketing campaigns at scale.)
This is where skill comes to play. You could put both me and Pharrell in music studios, but I won't be producing the same hits. The quality of its results are a function of what you put in. Any user that is using GPT3 out-of-the-box can create generalized content, but developing mastery of two skills can open a world of possibilities *and* ensure that you create content that is unique to you. What are those skills?
- Prompt Design. This is telling GPT3 exactly what you want. The more specific and the more data-driven the prompt, the better. For our purposes we’ll focus on the idea of feeding GPT3 data that only you have access to so your results can be specific to you.
- Fine Tuning. This involves providing GPT3 a lot more data on what you want it to say. OpenAI describes this as the “Ability to train on more examples than can fit in a prompt.” There are hundreds if not thousands of in-depth articles written about this topic, so I’ll keep it short.
Example: Let’s say you’re a real estate agent and you want to create a one-pager for your most recent home listing. Here’s an example of poor Prompt Design without Fine Tuning.
At first glance, this might seem like quality content, but it is very generic and factually incorrect, using dummy data to complete. The property is not 3 bedroom, not 2 bathroom, and in fact 3,466 sqft.
How would we design a better prompt and fine tune the response? Stay tuned for part 2 of this post where we go deeper into examples of fine tuning and structured data in order to create a stellar and unique pitch for a home.
Dataherald has helped dozens of companies create and deploy programmatic content strategies that use AI + structured data. We’re here to chat so feel free to Get in Touch.
