The Beta Reader prompt was the first one I built. It is still the most valuable.
Here is the problem: AIs want to tell you how great you are. You paste in a scene, and the response starts with, “This is a powerful and emotionally resonant piece of writing.” Every time. It does not matter if the scene is good, bad, or somewhere in between. The AI has opinions, and all of them are positive.
That’s useless.
What I need is someone to tell me where my writing is weak. Where I’ve let a bad idea leak into the scene. Where I’ve overwritten something that should be spare. Where I’ve included a paragraph that doesn’t need to exist.
Someone offered to be my beta reader. For a fee. A high fee. I don’t doubt a human beta reader would be better. I’m sure of it. But I need feedback now, not overnight, not next week. I’m publishing serialized fiction at a pace that doesn’t wait for anyone’s schedule. A human beta reader doesn’t fit into this process.
So I built a prompt that does it instead. Its instructions are blunt: Do not tell me what works. Identify only what needs to change.
It took iterations to get the tone right. LLMs default to encouragement. You have to be explicit about what you don’t want, or they slide back into praise. The current version tells the AI to flag problems only. No compliments. No padding to appear thorough.
The biggest value is when it tells me to cut.
We all have bad ideas. Mine show up as paragraphs that sounded important at 2 a.m. but do nothing for the scene. A character gives a speech that’s really me giving a speech. A description runs long because I fell in love with the setting. A transition exists because I couldn’t figure out how to skip it.
The Beta Reader catches these. Properly prompted, it calls me on my weak writing and tells me where to cut.
The prompt does not catch everything. Some edits, I believe no AI can do.
One is the cry test. If I am not crying by the time I finish a scene, it needs to be rewritten. That’s not a metaphor. If a scene should hit emotionally and I am dry-eyed at the end, something is missing. No prompt catches that.
The second is the wandering mind. I’m neurodivergent. My mind wanders. If it starts to wander while I’m reading my own writing, then it wanders for my readers too. When I find myself skipping ahead, that section needs work. An AI does not lose focus. It reads every sentence with the same attention. That’s a limitation, not a strength.
Supporting Libraries
I maintain three reference libraries that I update as the project develops.
Story Summary. A running summary of the entire novel, kept current. After each scene, I use AI to compress it to roughly one-twentieth its length. This gives the Beta Reader enough context to catch continuity problems without flooding the context window.
Character Library. Character references for each person in the novel. Physical details, relationships, voice patterns, backstory. When I load a scene for review, I include the character files for everyone who appears.
Fact Library. Canon facts about the world of the novel. Emily’s apartments. Dave’s house. Dave’s office in Nashville. Geography, layouts, details that have been published and cannot change. These prevent the AI from contradicting what I have already published.
Before each run, I load the summary, relevant character files, fact files as needed, and often one or two prior scenes for continuity.
Example: Wedding Preparations
My most recent scene, Wedding Preparations, needed only Dave’s and Emily’s character files. No fact files. I loaded one or two earlier scenes where Dave and Emily meet their future in-laws, because those scenes establish the family dynamics the Beta Reader needs to track.
I also wrote a character file for the pastor. To get him right, I had to get the church right, the denominational details, how a United Methodist pre-marriage counseling process actually works. That took research, all done with an AI and Google Maps.
The goal of that scene was specific. I wanted to model civil behavior. Not civil as in polite debate. Civil as in how people should treat each other in life. How Jesus would want us to be.
Dave is a white man from a four-generation Tennessee family. Emily is Hispanic, from a military family out West, working for a Democratic congresswoman. They’re getting married in his childhood church, sitting across from the pastor who baptized him. Race is in the room. Politics is in the room. Faith is in the room.
I wanted to address all of it. Not with a lecture. With two people who love each other talking honestly in front of someone who cares about them.
The Beta Reader helped. It flagged places where the conversation felt too smooth, where I had let Emily give a speech instead of an answer, where Dave sounded like he was performing for the reader instead of talking to his pastor. It pushed me to cut the parts that were me talking and keep the parts that were them.
That’s the value. Not grammar. Not style. The Beta Reader tells me where I’m being a writer instead of telling a story.
I wrote Wedding Preparations about two months ago, before I left for my last thirty-day travel adventure. I just read it again now. There are things I would change. Too much telling the reader what to think and not enough showing. I started writing this novel in October 2025 and have become a better writer with every scene. I read people complaining that AI is making them lazy. For me it’s the reverse.
The Prompt
The Beta Reader is Pass 3 in a four-pass editing pipeline. It handles everything a careful human reader would catch: continuity, pacing, character voice, clarity, and logic.
Each pass stays in its lane. The Beta Reader does not flag comma usage or passive voice. It does not fact-check military protocol. Those were already handled. It focuses on whether the scene works for a reader encountering it mid-series.
The output is structured. Every finding comes as an EDIT block or a NOTE block, each with exact text and a reason. I feed these into the Edit Markdown program I described in the first post and accept or reject each one.
Here is a link to the prompt:
These are designed to work with the Python program:
And if you want to see the project itself, the best starting point is probably the preface:
Preface — A Cold Civil War
Master Index
Substack Archive
All my posts related to AI are in this archive: