Using boxing as an analogy, Combat Writing is "title-fight" infrastructure. It is a production style for drafts that make decisions, move money, and have consequences for actual people.
An "exhibition fight" is when you only run the stages of this methodology that are relevant due to time sensitivity or limited scope. Most publications are exhibition, for content, compliance deadlines, staying active.
Pull punches or go for the KO. The match is yours to decide.
This approach takes you into the "gym" by synthesizing perspectives and orchestrating multiple AI models to reveal unforeseen results. Dialogue with LLMs and in-person discussions with humans is the "fight camp" for these key questions:
How does AI feedback complement human feedback in refining ideas?
When evaluating the quality of rough drafts, where do we agree and disagree with AI?
How does the Combat Writing approach engage with multiple AI to reveal hidden insights?
— WITHOUT AI —
How does feedback refine our ideas?
When evaluating the quality of rough drafts, where do we agree and disagree with one another?
How does the Combat Writing approach foster the conditions for insights to emerge?
STRATEGY STAGE
STEP 1
Select a meaningful project unique to your work or business. If it's not high stakes enough to require action, scratch it.
STEP 2 [SM: Source Material]
After reflecting on the desired outcome of the piece, write your first draft. Created entirely by you, the human. This becomes your "Source Material" and the foundation for all subsequent analysis.
STEP 3 [C: Context Brief]
Describe and establish clear context: Who you are (or your team), your purpose for writing this piece, what you're attempting to accomplish, and the stakes involved. Share this context with multiple AI models. We use Anthropic's Claude, xAI's Grok, and Perplexity as the main "crew". They earned their position by passing our custom challenges and standards.
SPARRING STAGE
STEP 4 [R: Solo AI Analysis]
Run this step in a private browser or incognito mode. This begins the sparring session with a cold read. Afterward, share the results along with your Source Material with the AI crew that has your context brief.
Upload your SM to an AI crew of your choice. We suggest at least three LLMs for identifying potential signals and patterns.
Prompt them based on your project's circumstances. For a starting point, use: "Read this rough draft and rate it on a scale from 1–10, you cannot use 7. Explain your reasoning with evidence."
STEP 5
Process the feedback from each AI model separately. Consider how the different LLMs evaluated your first iteration. If this is done without AI, consider the biases each person has when sharing their interpretations. Welcome various biases in a discussion as valuable learning signal.
STEP 6
Discuss specific points with human teammates or reflect on where you agree and disagree with AI. For example, if the critique involves topics uniquely human, like emotional responses or physical sensations, remember that artificial intelligence can be excellent at explaining yet not always at understanding.
STEP 7
Identify patterns among the multiple AI feedback and make revisions or edits. If the topic is too controversial or involves sensitive information, discuss it with human collaborators or with an offline, uncensored AI model. We run Llama 3 for this using the Ollama app.
BATTLE STAGE
STEP 8
Upload revised draft and ask specific follow-up questions: Would you recommend this piece? Purchase this service? Join this team? (AI has its own biases, responses can be subjective, and hallucinations do occur so take it with a grain of salt.)
STEP 9 [FQ: Focus Question Injection]
Create a focus question based on insights that have emerged. This might involve evaluating your work against specific criteria, research, outside references (books, articles, datasets, philosophies), or newly discovered frameworks.
STEP 10 [SFQ: Synthesis Focus Question]
Share the responses to your FQ from the AI crew with each model along with a new FQ. Each AI synthesizes their perspective while considering how their interpretations overlap and diverge with their AI peers.
Create a document organizing all crew responses to make sharing easier across models.
STEP 11 [N: Navigating Collaboration]
If synthesis hasn't fully revealed what matters, address the AI crew collectively with new guidance. Share pros and cons of their syntheses, redirect focus, or inject additional context to sharpen the next round.
STEP 12 [Optional SN: Synthesis Navigation]
Share each AI's synthesis with the other AI models along with your new prompt or refined focus. The crew reads each other's syntheses and responds to your updated guidance. Iterate as needed (SN2, SN3...) until structure and insights emerge.
STEP 13
Consider important details and results you may have previously been unaware of. Process the synthesis rounds (SFQ, N, SN) and apply signaling feedback — patterns, latent structures, contradictions — to reiterate your piece once more. If necessary, take any action you must to address emerging insights unique to your circumstances.
STEP 14
Prepare final draft, decide on the outlets to distribute your piece, and prompt your AI crew (or teammates IRL) with the final two prompts:
"What is the S/N on this final draft? (Signal-to-noise ratio)" Review feedback and make last minute adjustments if needed.
Then conclude with: "Read this final draft, are there any red flags before this is sent for publication?"
Fix any typos, omissions or other oversights and save your file for submission.
CHAMPION STAGE [ALL STEPS REQUIRED]
STEP 15
Publish your piece. Share it with your intended audience across chosen platforms.
STEP 16
Analyze metrics. Impressions, comments, business outcomes.
STEP 17
Acknowledge results, whether engagement is high or low.
STEP 18
Follow through with specific action. Reach out. Have the conversation. Close the deal. Make your next move.
Eventually, you may also host a "Roundtable (RT)" bringing in LLMs outside your AI crew to rate or critique your published piece as an external probe. Cold feedback from unfamiliar systems can confirm structure or unveil unexpected signals.
STEP 19
Reflect on what the Combat Writing process revealed. Document the insights and lexicon that emerged, this will become your "Insight Vault". This data becomes infrastructure components for your next project.
Monitor how your published piece performs in the field. If market response, reader questions, or new context reveal gaps, return to the piece. Update for stronger clarity and effect.
Publication isn't the endpoint. It's when your writing enters the real world and begins generating feedback. Use that signal to refine further.
BEFORE YOUR FIRST BATTLE SESSION
- Most AI platforms offer free tiers sufficient for Combat Writing. Claude, Perplexity, Grok and others provide enough monthly usage for several writing projects without paid subscriptions.
- Offline AI is always a recommended option for anyone with privacy concerns about using major AI models.
- Begin by uploading the Combat Writing framework to each AI model or copy and paste this entire post to your crew. Prompt: "I'm using this Combat Writing methodology to evaluate my drafts. Reference these steps throughout our sessions:"
- When uploading context before a session, it is best to upload word heavy documents as .txt files. For data, .csv or spreadsheets are fine. Avoid .pdf for text documents, save image slots for information that is actually visual.
- Create a document with the context of your project recorded in case you reach a chat limitation or it lags due to length/data constraints (then start a new chat).
- Organize a separate document for your FQ injection (excerpt or file reference to be copy and pasted).
- Prepare documents to save and track synthesis responses (SFQ, N, SN) to easily copy and paste across models. If sharing with team members, create a folder and organize all relevant information and insights.
- Use the speaker button (TTS) to hear AI responses aloud. This rests your eyes during long sessions and activates dual-channel processing for enhanced reflection.
- Combat Writing's iterative process invites seemingly endless depth to emerge; remember to stop when satisfied and publish!
- Professionals needing reinforcements or additional support navigating the CW process, join: The Writing Production