STEAL THE LESSONS I LEARNED FROM BUILDING AN INFINITELY SCALING AI COPYWRITER - THE SHAKESPEARE OCTOPUS

TLDR

• Focus on research not copywriting
• Don’t rely on LLMs, leverage classically proven schema 
• Make it multiagentic to optimize at multiple levels
• Model the customer at scale

> Why’d I build an AI copywriter, when I could just, you know…chatGPT it?

I’m passionate about copywriting.  Real copywriting.  There is a long, proud, and successful history of copywriting driving real sustainable business returns.  

And many current legends: Craig Clemens , Nicolas Cole 🚢👻 and the Category Pirates , Harry Dry , etc.

More - copywriting isn’t just the words on the page, but customer-centric, psychology-driven mode of business thinking.  An argument could be made that it’s one of the most powerful modes of business thought for customer-driven businesses.  

In fact, in my experience, copy is frequently more responsible for business success than design.  Especially in non-luxury markets and startups.  

Unfortunately, powerful copywriting is hard, time consuming, and rare.  

As a result, copywriting was one of the first things LLMs attempted to improve.  Unfortunately, it was and continues to be generally pretty bad without deft prompting (as of Aug 2024, Gemini is the clear copywriting leader in my opinion).  

But even with its limitations, I’ve found that using LLMs increased my productivity by several orders of magnitude.  The potential was there even if the tools weren’t.  In my head, I knew I had to eventually build a copywriter.

Fast forward a few months, a CEO friend of mine and I were discussing the sorry state of social media consultation.

I left that call and realized it was time to build the AI copywriter.  

What has emerged isn’t the tool of my dreams yet, but it produces very good copy at scale.  

Here’s what I think you can learn for your own AI projects:

• Think beyond the post - Use a research-based framework
• Don’t rely on your LLM - Enhance their intelligence with proven schemas
• Get insight your customers’ heads - Create cognitive-based models of your customers (and/or foundational world-model)
• Don’t just write, deconstruct - Optimize your results at multiple levels

Research-based Framework

The tool doesn’t produce one blog post or social post, it produces a copy-focused (for now) research plan.  

This means that it generates a series of research questions then creates the copy needed to test and answer those questions.

As a result, it doesn’t just produce high-quality copy, but generates customer insight.

TAKEAWAY: If your AI tool isn’t producing and improving from strategic insight, it will always be of limited value.  Kinda like hiring a new copywriter on Upwork for every new blog post. 

Proven copy schema

Copywriting is an ancient artform.  We’ve had many, many…many years to figure out what might work.  Many people have forgotten or never bothered to learn those lessons, but they are there if you look.  

This tool incorporates well known, battle tested schemas.  Eventually, it will generate and test new ones.  

TAKEAWAY: While I don’t know what happens behind the scenes at OpenAI, Google, or Anthropic, I’m guessing they are not optimizing for copywriting.  And while they’ve clearly ingested great copy, I’ve found your results will significantly improve if you provide, test, and optimize frameworks for your individual audiences.  

Of course, this requires that you know those frameworks for your individual use case, which is where human experience and research are rewarded.  It’s like the difference between hiring an expert marketer vs. someone who’s just been marketed to their whole life to craft your marketing strategy.

Cognitive-customer segment modeling

Everyone’s different and yet we often (sometimes by necessity) write to them as a group.  

There isn’t a copywriting book or expert or podcast or tweet or…that doesn’t advise some version of: “get to know your customer…”

But how?

Over the years, I’ve developed a model that blends cognitive science with the jobs to be done framework.  I use this model to create a digital twin of the customer segment.  Future generations will become more granular.  

TAKEAWAY: If your AI system can learn from and inform a model of your value chain, your output is likely to improve significantly.  Like walking a new city with a map vs. winging it.  

MultiLevel Optimization

When chatGPT generates a blog post it creates a whole post.  Duh.

However, good copywriting obsesses over words, sentences, paragraphs, narrative arcs, etc.  

While some amount of granular optimization is possible using chatGPT or Claude (say improving the headlines, subheads, etc.) at some point, you’re basically re-writing the whole thing and not saving yourself any time.

To truly be effective, the research, schema, and customer models must impact multiple levels within an individual piece of copy and potentially across copy touch points to build a holistic narrative.  

As far as I can tell, at this time, this level of multi-level granularity is only possible with an agentic approach

TAKEAWAY: If your application benefits from multiple levels of analysis, agentic AI is likely your best option.  I’ve found Langchain to be a straightforward and quick solution.

> While the app doesn’t replace me fully yet (especially since writing is part of my thinking process), it has already allowed my intention to scale exponentially.  It also allows me to address a much larger audience.

I hope this helps - Sean

#AI #ArtificialIntelligence #Copy #Copywriting #CustomerInsight #CustomerModeling #Python #MVP