How AI Made My Stories More Human

Claude told me that a story I was writing was not worthy of my craft.  Despite telling Claude to be blunt, that was still a slap in the face.  Perhaps I should back up a bit to show you how I got here.

I had a story about falling down a 600’ hole that I wanted to tell, but it was not coming together.  Something was missing.  It lacked emotional impact.  It was highly descriptive, but gave nothing to care about. I knew there was a great story in there, but it did not want to emerge.

I’m not talking about writer's block.  I’m talking about the difficulty of bringing a good idea over the finish line to where it’s something I would be proud of.  I needed to blend my inner critic with the honest work of bringing forth my best.

I trained AI to give me the direct feedback that I needed, without having it create a word of the AI-slop we hear so much about.  I’ll put the prompt at the bottom.

I gave Claude the links to about five stories that I already presented and believed that they represented the quality of work that I want to repeat.  I had way more stories I could feed it, but I chose to start with a manageable load to start so that I could confirm that Claude was on the right track, and to ensure the amount of data it had was manageable for a single pass.

AI then gave me two types of output.  The first was a synthesis of my craft, as I have been using it.

• My structure (staccato)
• Where I put the turn (at a crisis moment)
• Structural loops (tie the beginning to the end)
• Humor mechanics (deprecation without pity)

This was not feel-good praise.  It used elements of my own stories to explain and support the assessments that it made.  This was like walking around with your fly open until someone is kind enough to say XYZPDQ.  I thought each of my stories were unique, only to learn that they are remarkably similar to each other.  My stories are about times where I should have died, but didn’t, and they all followed a common formula that I knew existed, but could never articulate.  Ironically, at the same time that Claude told me my work needed improvement, it also offered the 988 help line based on the content it was reading.  (I’m fine.  Really.)

The real value lived in the next set of outputs. Claude gave me a set of “writing rules” that I employed.  These were not generic "best practices,” but were instead the rules that I actually demonstrated through my best work.  They were personally relevant.  AI then proposed a set of memories that it could use for our future work together.

I accepted most of the memories as-is.  One of them, while accurate, represented a pattern that I wanted to modify.  Because I’m so often the expert, I fail to ask questions that would make me personable.  I present my expertise in a way that makes me unapproachable, when what I want is to offer my experience as a gift and to form connections. By reflecting my implicit communication patterns, Claude gave me the power to choose whether to amplify or correct them. 

So I first acknowledged the pattern, then added instructions that AI should show me where I can ask questions that make me more approachable. Essentially, I was using this modified rule to boost my self-awareness to correct a practice that I wanted to use less often.

After that, I fed it three more stories and asked it to re-evaluate.  In all cases, the earlier rules were confirmed, but it was able to deduce an additional rule for me to consider.  I had to distinguish when the new rule applied and what made it different, then accepted it.

Even after feeding it four more stories, Claude was pretty clear on the rules that I have implicitly followed in my own storytelling craft.

The Final Test

I had a new story that I was working on, which I had not shared or presented yet.  I was pleased with the general content and arc of the new story, but it was not up to my standards.  I knew the story felt flat and that something was missing, but I was at a loss to identify how to address it.

So I fed AI the new story, telling it that it was a work in progress, and asked it to evaluate the new story in light of my established patterns.  AI did not write a single word of the story for me, but instead 

• It pointed out common elements that I employ, and had not used in this case. 
• It identified points in the new story where those elements would belong. 
• It identified where I was describing events in a detached voice, rather than an embodied voice.
• It told me that my conclusion was cliche and not worthy of my craft.

Essentially, AI told me “If you want this story to reach the level of quality of your other works, here’s what you need to change.”  Its feedback was completely structural, and based on rules we already established. The suggestions it made were not “say this.” Instead they were context based: “In the story about X, you used elements like A, B, and C.  None of those are present here.”

The net result was that I had trained my AI to be the writing coach I always wanted.  Within a couple hours, I was able to modify the story substantially and bring it up to my standards.  I did this without AI writting a word on my behalf.

So what about that 600’ hole?  I’m not going to give that away, but I can share that the key to getting out was knowing how to ask for help. 

If you’d like help with your business implementation of AI or Notion, schedule a conversation.

Walking Through The Process

The Training Prompt

The Memory Prompt

The Testing Prompt