Role-Task-Format-Loop Prompt Framework
Structure expert AI prompts around context, work, output, and repeated analysis
- Difficulty
- Moderate
- Time to result
- ~days to results
- Steps
- 6
- Confidence
- 99%
The framework builds an AI prompt in four layers. First, the role supplies a credible expert perspective and relevant scale experience. Second, the task defines the company, target outcome, differentiation requirement, and strategic question. Third, the format specifies exactly how the answer must be organized. Fourth, a loop applies the same analysis to every proposed channel or option, requiring tactics, validation, differentiated proposition, risks, budget, necessary skills, and a step-by-step guide. This repeated structure makes recommendations comparable and exposes omissions that polished prose can hide. The user then inspects weak assumptions, adds missing inputs such as average contract value, and iterates. Domain expertise remains necessary because prompt quality and output evaluation both depend on understanding the strategy being requested.
Origin
Extracted from Marketing Against The Grain as Kieran Flanagan demonstrated a Claude 3 prompt for building a differentiated SaaS marketing strategy from $10 million to $100 million ARR.
Core principles
- 01Assign expertise before requesting strategic work
- 02Specify the business objective and constraints precisely
- 03Define the required response structure rather than accepting generic prose
- 04Repeat the same evaluation dimensions for every proposed channel
- 05Include risks, resources, and validation alongside recommendations
- 06Iterate using missing business inputs such as ACV
How to run it
- 1
Assign the role
Tell the model which expert it should emulate and specify the relevant type and scale of prior success.
Pro tip Match the role's experience to the business stage in the task.
Watch out A generic expert role may produce generic recommendations.
- 2
Define the task
Describe the company, market, strategic objective, growth target, and required differentiation.
Pro tip Include economic inputs such as ACV when they affect channel viability.
Watch out Missing business constraints can make unsuitable channels appear attractive.
- 3
Specify the response format
Define the sections, level of detail, and sequence required in the answer.
Pro tip Request a concise strategy summary before detailed analysis.
Watch out Leaving the format open encourages uneven or incomplete coverage.
- 4
Create the analysis loop
Require the model to repeat the same evaluation structure for each recommended channel or option.
Pro tip Use explicit tags or headings to make the loop unambiguous.
Watch out Calling a repeated structure an if-loop does not replace specifying exactly what repeats.
- 5
Demand operational dimensions
For every channel, request tactics, validation, differentiated proposition, risks, budget, required skills, and implementation steps.
Pro tip Ask how the channel can sustain the target scale rather than merely generate initial traction.
Watch out Recommendations without resource and risk analysis can sound strategic while remaining unusable.
- 6
Critique and iterate
Review channel choices and detail quality, then add missing context or deepen weak sections in subsequent prompts.
Pro tip Use your own strategic expertise to challenge recommendations rather than accepting them wholesale.
Watch out A sophisticated prompt does not guarantee factual accuracy or sound judgment.
In the wild
The prompt assigns the model the role of a marketer who has scaled SaaS businesses from $10 million to $100 million ARR. It asks for differentiated, sustainable channels for an AI productivity product for sales representatives. For every channel, it requires tactics, validation, differentiation, risks, budget, hiring needs, and an execution guide.
→ Claude produces a detailed strategy with channel-specific analysis, budgets, skill sets, risks, and implementation guidance that can be refined through iteration.
Common mistakes
Omitting economic context
The model recommended events and conferences partly because the prompt did not specify the product's average contract value, an input that affects channel economics.
Stopping after the first output
The initial response can expose missing assumptions and sections that need greater depth. Iteration is part of the mechanism.
Using AI without domain judgment
The hosts explicitly note that strategic usefulness depends on the user's own strategy and prompting ability.
Is it for you?
Best for
It is best for experienced operators using an LLM to draft marketing strategies, channel plans, or similar multi-option analyses.
Not ideal for
It is not ideal when the user lacks enough domain knowledge to judge incorrect assumptions or weak strategic recommendations.
From the transcript
“I first of all start with role, right? I give it a role.”
“Then we give it a task.”
“And then I tell it the exact format, right?”
From the episode
The Marketing Mistake That Could End Starbucks