Dialectical Prompting (DP)
Judgment, not generationDP is an A/B testing method that forces you to compare outputs, log contradictions, and make craft decisions. It's not about getting the "best" AI output—it's about building your judgment muscle.
Core Principle
AI is a judgment tool, not a generation tool.
You don't copy-paste AI prose into your work. Instead, you:
- Test different approaches (A vs. B)
- Notice where they contradict
- Decide which choice aligns with your craft goals
- Document your reasoning in a Process Note
The DP Workflow
- Design a BRIEF+QA prompt (see checklist below)
- Run A/B test: Change one variable (tone, length, image density, POV, etc.)
- Log contradictions: What differs between A and B? Why?
- Make a craft decision: Keep A, keep B, or create hybrid
- Write a Process Note: Document what you kept, rejected, and why
BRIEF+QA Prompt Checklist
Every DP session starts with a well-designed prompt. Use this framework:
| Element | What to Include |
|---|---|
| Background | Genre, narrator type, tone range |
| Role | "You are a fiction writer specializing in..." |
| Instructions | Task + constraints (length, POV, image system, etc.) |
| Examples | 1-2 craft examples or stylistic references |
| Format | Output structure (paragraph, dialogue, scene beat, etc.) |
| Quality-bar | What counts as success? (e.g., "micro-tension in every line") |
| Audit | "Before outputting, check for filter verbs, clichés, etc." |
A/B Testing Strategy
Change one variable between A and B. Examples:
- Tone: Intimate vs. detached
- Length: 200 words vs. 400 words
- Image density: Sparse vs. heavy image system
- POV distance: Close psychic distance vs. far
- Pacing: Fast (short beats) vs. slow (long beats)
- Dialogue ratio: 70% dialogue vs. 30% dialogue
Why one variable? So you can clearly see what changed and why.
Contradiction Log
After running A and B, log at least 3 contradictions. Example:
| Craft Lever | Output A | Output B | Which Works? |
|---|---|---|---|
| Psychic distance | Close (internal monologue) | Far (external observation) | B—keeps tension higher |
| Image system | Recurring water imagery | No recurring images | A—reinforces theme of loss |
| Sentence length | Short, staccato | Long, winding | A—matches anxious tone |
Process Note Template
Every DP assignment requires a Process Note. Include:
- Tool/Model: ChatGPT 4, Claude 3.5, etc.
- Parameters: Temperature, length, any custom settings
- Prompts: Full BRIEF+QA for both A and B (or note the one variable you changed)
- Contradictions: What differed? (use table format)
- Craft Decision: What did you keep? What did you reject? Why?
- Final Output: The prose you actually wrote (not AI-generated, but informed by DP)
Example Process Note excerpt:
Tool: ChatGPT 4 (temp 0.7)
Variable tested: POV distance (A=close, B=far)
Contradiction Log:
- A used internal monologue ("She thought..."); B stayed external
- A had filter verbs ("She felt the cold"); B showed action ("She shivered")
- A explained emotion; B trusted image system
Decision: Kept B's external stance but added one internal beat
from A for character access. Rejected A's filter verbs entirely.
Why: Close POV felt over-explained. Far POV kept tension but
risked detachment—so I added one internal moment for grounding.Debugging Matrix
When AI outputs fail, diagnose the problem and adjust your prompt:
| Problem | Symptom | Fix |
|---|---|---|
| Vague | Generic language, no proper nouns | Add "specificity ratio 40%+" to Quality-bar |
| Hallucinated | Invented facts, names, or events | Add "only use details I provide" to Instructions |
| Flat | No tension, no stakes | Add "micro-tension in every beat" to Quality-bar |
| Over-literal | Follows prompt too rigidly | Add "prioritize craft over instruction" to Role |
| Messy | Filter verbs, adverbs, clichés | Add Audit step: "Remove filter verbs and adverbs" |
What DP Is NOT
- NOT copy-pasting AI prose into your story
- NOT asking AI to "write a better version"
- NOT using AI to avoid workshop feedback
- NOT generating workshop letters or peer feedback
- NOT a shortcut—it's a deliberate, time-intensive practice
DP Assignments (A1-A4)
- A1 (Thu Oct 2): First DP session—BRIEF+QA, A/B, Contradiction Log, Process Note
- A2 (Thu Oct 9): Debugging AI outputs using the Debugging Matrix
- A3 (Thu Nov 13): Use DP to revise one scene from your fiction or CNF
- A4 (Thu Dec 4): Reflection essay on your DP practice this semester
✅ DP Success Looks Like...
- You can explain why you rejected 80% of what AI generated
- Your Process Note is longer than the AI output
- You used AI to test an idea, not to avoid making a choice
- Your final prose reflects your voice, not the model's