Slinging Ai-Written Pulp Fiction On Amazon For Fun And (Very Tiny) Profit

At 83 copies for a grand total of $829 so far. Here’s what I did.

I took another sip of bulletproof coffee and tapped the trackpad once again.

Green highlighted text filled my screen.

I squinted.

That’s as good as it gets with very little input and a half-baked prompt on the other side of too much Modafinil and grass-fed butter coffee.

Terrible writing.

But a good enough starting point. 

I’ve dabbled in fiction for years. Only finished a handful of novels, started and abandoned around forty others, and still maintain a collection of hundreds of scraps, napkin scribbles, and index card notes.

Characters. Plots. Scenes. Dialogue. 

One weekend, as I was knee-deep in applying various AI tools to solve marketing problems for a client, I took a break and figured GPT-3 could help me write something funny.

It quickly evolved (or devolved) into a variety of scene ideas for a modern-day pulp fiction story, with entrepreneurs, online guru scammers, vaporware, boiler rooms, crypto, internet marketing weirdos, stacks of cash, supplement scams, law enforcement, legal drama—you get the idea.

A hodgepodge of pieces. A plethora of potential. 

I needed a structure and specific framework for turning the bits into pulp.

I’ve preferred Dan Harmon’s Story Circle over The Hero’s Journey.

And being a fan of pulp fiction but with no clue how to write pulp, I set out to cobble it all together. 

I started feeding GPT-3 ideas:

Not great. But you can turn anything terrible into something great.

I experimented with taking chunks from existing pulp fiction work and seeing what GPT-3 would give me back:

So weird. But interesting. And you can do something great with interesting.

Through my work combining AI and marketing, I knew full well at this point that you need to keep refining, editing, and re-prompt GPT-3 to get useful text. 

You get something like the above, which is a quintessential scene in pulp fiction (damsel in distress type thing, usually a murder, corrupt police, etc.)

But then you need to edit and refine it. 

After a couple of weeks worth of prompting, rewriting, editing, writing, prompting, and editing some more, I had a first novel in a series.

About 120 pages. Around 100,00 words.

I’d never written a pulp fiction story before but I have read a lot of it. I knew what to look for. I had a list of scenes that are typical. A general sense of what characters are common. Plot ideas. Antagonists. Sidekicks. Twists and turns. All the usual ingredients that most fiction shares.

I’m being vague on purpose here because the book is live and selling on Amazon (and has been selling since about September 2021). 

The experiment is to see how many copies it sells (as of today, 83), what kind of ratings it gets (4 stars), and if I can keep writing and adding to the series. 

All without anyone knowing a robot is my co-writer. 

Once I publish two more (for a total of three), I’ll give you all a link. But once everyone knows, it’ll put an end to the experiment and I’ll have people buying one or two for novelty, ratings will either be 1-star negative or maybe 4-star but all miss the point.

The second and third books are in production. And in recent weeks, I’ve come across James Hull and his Subtxt app (along with his Maven course, AI-Powered Storytelling).

This has given me more structure to fill with GPT-3 pulp.

And I’ve ended up with scene ideas like this: 

It’s raw. But you can do something with this.

Or this: 

All of this highlights one thing that’s critical for making GPT-3 work for you:

You must have a structure and intention behind the prompts and what you do with it all. 

Perhaps GPT-4 will blow us all away even more and make magic possible at the click of a button.

But for now, to write anything with GPT-3, you must have a mold to pour it all into. 

Talk soon,
Samuel Woods
The Bionic Writer