A four-prompt sequence is circulating that promises to extract a real, actionable business plan from a large language model. The kind of plan that pushes past the generic “build an audience, find your niche, create content” advice that most AI tools default to. The promise is overstated. The underlying prompt pattern is genuinely useful, and worth understanding on its own terms.
The pattern is interesting because each of the four prompts solves a specific failure mode of asking an LLM for business advice. A single broad prompt (“help me make money with AI”) produces generic output because the model has nothing to anchor against. The sequence below works because each step gives the model a more specific anchor than the previous one, until the final answer has enough structure to actually be tested.
Here is what the four prompts actually do, why each one works, and what the pattern is and is not good for.
Why a single prompt fails
LLMs default to generic advice when given generic questions. Ask “how can I make money?” and you get the union of every business book in the training corpus, averaged out. The output sounds reasonable. It is also unfalsifiable, untestable, and approximately useless for actual decisions.
The fix is not to write a better single prompt. It is to build a small chain of prompts that progressively narrow the model into a specific, testable plan. The sequence below does that in four steps.
Prompt 1: leverage memory, force a single answer
The first prompt: “Based on everything you know about me, what is the fastest way to reach my income goal? One offer, one channel.”
Two things make this prompt work harder than it looks.
The first is the implicit appeal to persistent memory. On products that have a memory feature (Claude.ai has one, ChatGPT has one), the model already holds context about you across conversations: your work, your goals, your prior projects, your constraints. Most users do not actively invoke that memory when asking strategic questions. The phrase “based on everything you know about me” forces the model to use what it has.
The second is the “one offer, one channel” constraint. This is a forcing function. Without it, the model produces a buffet of options, which is the polite way of producing nothing usable. The constraint pushes the model to commit to a single specific direction. The commitment may be wrong. A specific wrong answer is much more useful than a buffet of vague right answers, because the specific wrong answer can be tested against.
Constraints are how you get an LLM to be useful for strategy. A specific answer the model is willing to defend is more actionable than a careful list of options. The forcing function is the prompt’s job.
Prompt 2: extract a specific persona
The second prompt, run in the same conversation: “Who is the most inspiring person to follow to get there? Give me one name.”
This is, on the surface, just asking for a recommendation. What it actually does is set up the next prompt. The model has just identified a person who, in its training data, exemplifies a particular approach to the strategy proposed in prompt 1. That person becomes the handle for the next move.
The “one name” constraint operates the same way as “one offer, one channel” did. It prevents the model from hedging across five inspirational figures. The handle has to be a single concrete reference for the next prompt to work properly.
The caveat worth knowing here: the model is recommending a person from its training data. That person may be a real expert. They may also be a popular figure whose actual playbook is more idiosyncratic than the public version of it. The model knows the public persona, not the private working method. The name it returns is a stylized reference, not a real coach.
Prompt 3: persona-driven planning
The third prompt, often the most useful in the sequence: “Act as if you were that person. Create me a 90-day plan with less than €1,000 investment to get my first paying client as fast as possible.”
Persona prompting has a complicated reputation. For many tasks it adds noise without helping. For this specific task, it works, because it shifts the model from “average across all advisors” to a single stylized perspective, and that specificity is exactly what the previous two prompts were setting up.
The three constraints in this prompt all matter individually.
The 90-day horizon is short enough to force urgency but long enough to allow a real plan. Asking for a five-year strategy produces empty pages. Asking for “tomorrow’s actions” produces task lists with no arc.
The €1,000 budget cap eliminates the entire class of plans that quietly assume venture capital or extended runway. It forces the model to produce something a person with a normal financial situation could actually execute.
The “first paying client” target eliminates the entire class of plans that mistake content creation, audience-building, or product polish for income. The plan must, by construction, end with money changing hands.
The combination is sharp enough to produce a plan that is testable. It may still be wrong. It is no longer vague.
Prompt 4: the Socratic move
The fourth prompt, the one most people skip: “Act as my entrepreneur coach. Ask me questions one by one until you find what is really blocking me. Reframe it and give me a two-week plan to prove it works.”
The first three prompts produced a plan. This prompt does the opposite. It interrogates the plan by interrogating you.
The reason this works is straightforward. The plan the model produced in prompt 3 is based on the model’s reading of your situation, which is based on its memory of you, which is necessarily incomplete. Most of the reasons a business plan fails are reasons that were not visible in the original framing: assumptions that turned out to be wrong, constraints that were not surfaced, motivations that were not examined.
The Socratic question structure forces those assumptions out. You answer one question at a time, in real conversation. The model adjusts its understanding based on your answers, then asks the next question. By the end of the dialogue, the model has a meaningfully sharper picture of your actual situation than it did after prompt 1, and the resulting two-week plan reflects that.
The “two-week plan to prove it works” is a separate forcing function. Two weeks is short enough that the plan must consist of testable actions, not preparatory work. “Prove it works” pushes the plan toward producing real signal from the market, not internal milestones.
The fourth prompt is doing structurally different work from the first three. The first three extract a plan. The fourth extracts the gap between that plan and your actual situation. The combination is much more useful than either alone.
Why the sequence works
Stepping back: each prompt in the sequence applies a specific prompt-engineering pattern that has been shown to improve LLM output quality in isolation. The interesting thing is what they do in combination.
Prompt 1 establishes context and forces a focused first answer.
Prompt 2 extracts a stylized reference frame.
Prompt 3 uses that reference frame to produce a constrained, testable plan.
Prompt 4 stress-tests the plan against your specific situation through dialogue.
The combination is more useful than any of the four individually, because each step makes the next step possible. A persona prompt without a defined focus produces generic advice. A plan request without a persona produces averaged-out output. A Socratic dialogue without a candidate plan produces a directionless conversation. Run in sequence, each prompt’s output becomes the next prompt’s anchor.
What this pattern is not
A few caveats worth being explicit about, because the framing this sequence travels under tends to oversell what is happening.
The pattern does not produce expert business advice. It produces a structured version of the model’s averaged view of what business advice looks like, narrowed by your constraints. That is meaningfully better than unstructured generic advice. It is not the same thing as a real strategy session with a real expert.
The “act as if you were [X person]” move produces the model’s stylized version of X. It does not channel the actual person. If X has insights that are not widely public, the model does not have them. The persona is a handle for narrowing, not a substitute for the real thing.
The plans this sequence produces should be treated as hypotheses to test, not instructions to execute. The two-week plan from prompt 4 is genuinely useful precisely because it is designed to produce signal from the market. Running it as a test, paying attention to what you learn, and updating the plan is the loop. Following it blindly is not.
The output of this sequence is a credible starting point for a business hypothesis, formatted in a way that makes it testable. It is not a recipe for success. The difference is the gap between any LLM output and a real outcome: you still have to execute, observe, and adjust.
When to actually use this pattern
The sequence is at its most useful when you have a genuine business question that has been stuck on “I am not sure what to focus on.” It collapses the analysis-paralysis loop into a specific, narrow, testable plan, which is exactly what stuck people usually need.
It is at its least useful when you already know what to do and just need to execute. In that case, you do not need a strategy session. You need a verifier sub-agent and a calendar, both of which are different patterns.
It is at its most dangerous when used as a substitute for thinking, particularly in fields where the cost of a bad plan is high. Use it as a structured way to surface and stress-test ideas. Do not use it as a replacement for the part of the process where you and the market actually meet.
Frequently Asked Questions
Does this sequence actually work?
It works in the sense that it produces a meaningfully more specific and testable output than a single broad prompt would. It does not work in the sense of guaranteeing business outcomes. The output of the sequence is a hypothesis to test, not a path to follow.
Why does the “act as if you were [X person]” prompt help?
It shifts the model from averaging across all advisors to producing a stylized version of one specific perspective. That specificity is what previous prompts in the sequence were setting up. The caveat is that the persona is the model’s reading of the public version of that person, not the actual person.
Why is the Socratic prompt the most important?
Because the first three prompts assume the model already understands your situation. The fourth prompt tests that assumption by interrogating you. Most business plans fail on assumptions that were never surfaced in the original framing. The fourth prompt is the move that surfaces them.
Can I run this sequence with any LLM?
Yes, with one caveat. The first prompt’s appeal to “everything you know about me” only works if you are using a model with persistent memory (Claude.ai or ChatGPT with memory enabled). On a stateless model, prompt 1 needs to be replaced with a quick summary of your context.
What is the biggest mistake people make with this sequence?
Treating the final output as instructions instead of a hypothesis. The plan from prompt 3 and the two-week tests from prompt 4 are designed to produce signal from the real world. Their value is in what you learn when you run them, not in their correctness on paper.
Is this sequence a substitute for actually learning business?
No. It is a structured way to surface and narrow business ideas using an LLM as a thinking partner. The judgment about which hypotheses are worth testing, how to interpret market signal, and when to pivot all still require you. The pattern speeds up the framing. It does not substitute for the work.
