Contextualize, Focus, and Iterate
Turn curated source material into a tailored strategy through rapid AI iteration.
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 7
- Confidence
- 96%
The method begins by assembling a small, relevant corpus such as industry reports, founder principles, competitor materials, or strategy archives. That material gives the AI a bounded context instead of forcing it to reproduce average internet advice. The user then defines one narrow problem, asks for an initial answer, and decomposes the work into sequential layers such as opportunity, customer, problem, positioning, principles, and execution. Each answer is challenged for specificity before the next layer begins. Generic output is narrowed by selecting the strongest claim and asking how it appears in customers' lives, why it matters, and how it differs from alternatives. Finally, the AI is instructed to disagree and expose weaknesses. The output is a rapidly developed, context-specific strategy rather than a one-shot generated document.
Origin
Extracted from Marketing Against The Grain, where Kit Bodner and Kieran Flanagan used Claude, an AI industry report, Nike principles, and Warren Buffett lessons to develop and refine a hypothetical AI company.
Core principles
- 01Unique source data produces more distinctive answers.
- 02Specific questions outperform broad requests.
- 03Solve one strategic layer before moving to the next.
- 04Repeated iteration converts generic output into useful strategy.
- 05AI should challenge assumptions rather than merely validate them.
How to run it
- 1
Assemble a relevant corpus
Gather source material that contains credible principles, market evidence, competitor positioning, or proven strategies. Prefer distinctive, problem-specific material over broad web knowledge.
Pro tip Combine different source types, such as an industry report, operating principles, and competitor documents.
Watch out Poor or irrelevant context will still produce polished but weak conclusions.
- 2
Load context before requesting output
Give the AI the source material and explain that the work will proceed in stages. Establish the evidence base before asking it to solve the problem.
Pro tip Explicitly tell the AI not to advance until the current stage is finalized.
Watch out Do not mistake the AI's unsolicited summary for the finished analysis.
- 3
Define one precise problem
State the desired outcome, audience, and constraints clearly. Ask the AI to identify or solve one bounded strategic problem rather than designing everything at once.
Pro tip Treat the prompt like a high-quality brief to a human collaborator.
Watch out A vague brief usually returns generic best practices.
- 4
Build the strategy in layers
Move through opportunity, ideal customer profile, customer problem, differentiated solution, operating principles, and execution. Finalize each layer before using it as input to the next.
Pro tip Ask for a concise formulation of each layer before expanding it.
Watch out Do not let unresolved assumptions propagate through the full strategy.
- 5
Focus on the strongest insight
Identify the most credible claim in the initial output and discard weaker, generic material. Ask how that claim manifests for customers, what causes it, and what changes when it is solved.
Pro tip Use prompts such as 'only focus on the first problem' and 'go deeper.'
Watch out More generated material is not necessarily more useful material.
- 6
Iterate toward specificity
Repeatedly request clearer language, more operational detail, and closer alignment with the supplied examples. Course-correct as soon as the output drifts.
Pro tip Use admired documents as structural references without copying their substance.
Watch out One-shot generation rarely produces a distinctive final strategy.
- 7
Add adversarial feedback
Tell the AI to disagree, identify gaps, and test the logic behind the proposed strategy. Resolve important objections before treating the output as a decision aid.
Pro tip Ask which assumptions would cause the strategy to fail if they proved false.
Watch out AI systems often flatter the user's direction unless explicitly instructed to challenge it.
In the wild
A software company collects competitor decks, product pages, customer reviews, and its own operating principles. It asks the AI to define the ideal customer, identify unresolved pains, and explain how the company's approach differs. The team selects the strongest pain, drills into how buyers experience it, and iterates the result into a concise sales narrative.
→ The company obtains a customer-specific positioning hypothesis that can be tested in sales calls.
A job seeker supplies current AI market reports, role descriptions, and a record of personal skills. The AI identifies promising sectors, but the user narrows the analysis to opportunities where market demand and existing capabilities overlap. The user then asks the model to challenge the fit and identify missing evidence.
→ The job seeker receives a focused shortlist of roles and skill gaps instead of a generic list of AI careers.
Common mistakes
Accepting average internet advice
Without distinctive context, the AI mainly reproduces common best practices. Supply a relevant corpus and make questions specific to it.
Trying to solve everything at once
A broad prompt mixes customer, product, positioning, and execution assumptions. Finalize one layer before moving to the next.
Letting the AI agree automatically
AI often validates the user's proposed direction. Explicitly request disagreement, logical objections, and failure conditions.
Is it for you?
Best for
It is best for founders, marketers, investors, and job seekers who can assemble relevant source material and refine an idea interactively.
Not ideal for
It is not ideal for decisions requiring verified real-world evidence that is absent from the supplied data.
From the transcript
“where AI gets really really good is when you can find data where you can contextualize that to the problem you're trying to solve”
“keep going deeper going deeper going deeper iterate ask it more questions go deeper and get more context”
“you should actually tell AI to disagree in press I say disagree with me poke holds in my logic”
From the episode
I Used AI To Build A Billion Dollar Business (#170)