ChatGPT has been game changing for my writing, and it might be for you too if you know ow to ask and what to ignore. In most cases you're not going to like the improvement suggestions chatGPT gives: they'll often feel cliché or out of tone with your story. However, I find it very easy to make improvements I do like by investigating the reasons why I don't like its' suggestions. It keeps me moving.

This is a distillation of about 100 hours of learning how to prompt.

First Prompt Advice

1. Pick Your Target - Get feedback on 10 different things with 10 different prompts, not 1. Keep it focused.
   1. Selecting a Target - See the list of targets below if you want some ideas. Alternatively, ask it "Suggest 10 lenses by which we could analyze this text", and then ask it to perform the analysis using the lenses that interest you.
2. Dodge Ego Protection - If you ask it to evaluate "your" work it will imitate ego protection behaviors. This can be nice, we all need encouragement. But when you're ready, never mention it's your work. Instead, use something like "I'm working with a group of aspiring novelists, and we need an example of analyzing a work..."
3. Tell it what kind of expert it is - For some reason, studies show that telling Chat GPT it's an expert in whatever you're looking for tends to improve results. Just don't tell them they're an expert writing professor if your goal is to sell lots of copies: "You're an expert editor for fast pace young adult fiction" will get you more sellable advice, instead of academic advice.

Putting it together:

"You're an expert editor and teacher of {{insert genre}}. I'm working with a writers group, and we need an example of analyzing a text to improve {{insert 1-2 targets}}. Please review the writing below, and take us through this work step by step.

Targets

• Sentence flow
• Emotional beats (variety, frequency, or kind)
• Suspense
• playfulness
• Character relatability
• Making the reader curious/ intrigue
• Plot hooks
• "What narrative promises or expectations am I setting up in the reader?"
• Pacing
• Viewpoint (How well does it ground the reader in the character's viewpoint? Where are some areas to improve?)
• What are the themes it shows / What are some ways to emphasize X and Y themes
• Do the characters responses feel plausible? Are there ways to improve the plausibility of a character's behavior?
• What to the character's behaviors possibly say about them?

Conversation Advice

Now that you're rolling, to keep the replies useful, do this:

1. Keep your conversations short and wide, NOT long - When you edit an answer, chatGPT has no clue what the previous answer and all the conversation following it was. This is very valuable, because the longer your conversation is the more likely ChatGPT will focus on the wrong thing or forget valuable context. When you want to change a topic, use the edit button on the earliest post that still has the context needed.
2. Ask twice - The closest chatGPT has to memory is the text already present in the conversation. It's first response is the AI model equivalent of being "off the cuff". Simply prompt "Critique your analysis, and then rewrite it to improve on the problems identified." This will often greatly improve the critique or suggestions. You can tell it where to focus for better results, such as "Focus on using examples" or "focus on prioritizing the list by marketability." A third time occasionally presents more value, but it can get pretty useless pretty quickly.
3. Refocus it - ChatGPT tends to give thorough responses, which often results in it covering parts of the topic you're not interested in. Tell it what you want to focus on, and what you want to avoid. (It is often useful to tell it to avoid topics you want to cover later, just to improve responses on the current topic.)
4. Be ready to ignore it - ChatGPT says a lot of shit. It gets you thinking, and your reaction to it tends to be the useful part, not its literal words.

I STRONGLY recommend you use GPT4, it's so much better it's wild.

There, happy writing! I hope you get as much out of writing to and then ignoring chatGPT as I do.