Thought-Leader Revision Loop
Keep human judgment in control through repeated critique and adversarial review
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 6
- Confidence
- 96%
The Thought-Leader Revision Loop separates generation from judgment. After AI produces an answer, the user refuses to treat it as final and explicitly identifies what is strong, what is weak, and which changes matter most. The AI revises, and the exchange continues until the user reaches the limit of their own perspective. At that point, the model changes roles: first into a challenger skilled at finding insufficiencies, assumptions, cracks, and biases, and then into a realistic customer or stakeholder persona that evaluates the proposal from the receiving side. This sequence raises quality while ensuring the human remains the thought leader. It also creates cognitive demand because the user must evaluate evidence, articulate preferences, resolve criticism, and decide which recommendations deserve adoption rather than passively accepting fluent output.
Origin
Extracted from BigDeal, where Geoff Woods describes the refinement behavior he uses after CRIT and explains how he plays AI against itself to improve results.
Core principles
- 01AI supplies drafts while the human owns judgment
- 02The first answer is never automatically final
- 03Specific feedback improves subsequent iterations
- 04Opposing personas expose assumptions and bias
- 05Customer simulation tests usefulness before delivery
How to run it
- 1
Downgrade the first answer
Treat the initial response as an incomplete draft regardless of how polished it sounds. Approach it with skepticism rather than admiration.
- 2
Identify strengths
Tell the AI which elements are accurate, useful, persuasive, or aligned with your intent. Preserve those elements during revision.
- 3
Identify weaknesses
Explain what is inaccurate, generic, verbose, misaligned, or unsupported. Include omissions and undesirable stylistic patterns.
- 4
Prioritize changes
List the most important modifications and request an updated version. Repeat the feedback cycle until your own critique stops producing meaningful gains.
- 5
Activate the challenger
Switch the model into an adversarial role that stress-tests assumptions, biases, and weaknesses in the current result. Resolve the strongest objections it finds.
- 6
Simulate the audience
Have the AI evaluate the revised result as an ideal customer, board member, or other stakeholder. Use that feedback to make the final human decision.
In the wild
A leadership team develops a growth plan with AI, critiques several drafts, and then asks the model to act as an aggressive board member. After resolving the surfaced risks, it switches the model to a documented customer persona to test whether the recommendation addresses real needs.
→ The final proposal is more robust and better aligned with both governance and customer expectations.
Common mistakes
Submitting the first draft
A polished first response can still contain generic language, hidden assumptions, and factual errors.
Requesting an undefined improvement
Telling AI merely to make something better avoids the human work of specifying quality.
Using a passive challenger
Without explicit instructions to attack assumptions, the model may continue to agree and flatter.
Is it for you?
Best for
It is best for consequential writing, recommendations, plans, and decisions where quality improves through critique.
Not ideal for
It is unnecessary for low-risk deterministic tasks whose result can be verified directly.
From the transcript
“Viewed as a first draft. I told him, tell it what you like about it. What's good about it? What do you not like about…”
“So I really get AI to play against itself to elevate the quality of the results.”
“Remember whatever draft, whatever it gave you is just the first draft.”
From the episode
How to Use AI to Make Money, Grow Your Business, and Be More Productive