Topic 5: Structured prompting framework
Structured prompting framework
Once you are comfortable writing basic prompts, the next step is to give AI more structured and relevant context.
A simple prompt can work perfectly well for a simple task. But when you are asking AI to help with something more important or complex, such as developing a roadmap, analysing requirements or preparing a product recommendation, a few lines of instruction may not be enough.
A structured prompt helps AI understand the situation, the problem, the people involved, the outcome you want and what a good response should look like.
Start simple
You do not need a detailed framework for every AI conversation.
For many everyday tasks, a standard prompt might be enough:
“Act as a Product Manager. Create a summary of the customer research for our B2B SaaS product, targeting senior stakeholders, so that they can understand the three most important customer problems. Use a concise executive format.”
This already gives AI several useful pieces of information:
- the role it should adopt
- the output you want
- the product context
- the audience
- the goal
- the format
For quick or low-risk work, this may be all you need.
For more complex work, you can go further.
The structured prompting framework
The framework gives you 11 areas of context you can use when building a more complete prompt.
You do not need to use all 11 every time. Think of them as building blocks. Choose the ones that matter for the task you are doing.
1. Role
Tell AI what role or perspective you want it to take.
For example:
“Act as an experienced Business Analyst.”
Or:
“Act as a Product Manager responsible for a B2B SaaS product.”
A role can help guide how AI approaches the task.
2. Business context
Explain the environment in which the work is happening.
Useful context might include:
- project
- industry
- business area
- product or platform
- current situation
This helps AI understand the bigger picture rather than treating your request as a generic exercise.
3. Users
Tell AI who the relevant users or customers are.
For example, a product may have primary users, secondary users and internal stakeholders.
Knowing who the product serves helps AI make its response more relevant to the people involved.
4. Problem
Clearly describe the problem you are trying to solve.
For example:
“New customers are abandoning the onboarding journey before completing account setup.”
This gives AI something specific to analyse rather than asking it to solve a broad product problem.
5. Goal / Outcome
Explain what you ultimately want to achieve.
Your problem describes what is wrong today. Your goal describes what better looks like.
For example:
“We want to improve onboarding completion while reducing the amount of support customers need.”
This keeps AI focused on the outcome rather than simply producing activity.
6. Constraints
Real product work always has boundaries.
You might have:
- budget limitations
- technology constraints
- regulatory requirements
- deadlines
- existing platform limitations
- agreed business rules
Tell AI about the constraints it needs to work within.
Give AI something to work with
One of the most powerful ways to improve AI output is to provide relevant reference material.
7. Reference material
Instead of expecting AI to know everything about your project, give it the information you already have.
This might include:
- templates
- research
- examples
- roadmaps
- product backlogs
- user personas
- requirements
- other project documents
This connects directly to contextual prompting. The more relevant information AI has, the less it needs to guess.
8. Output requirements
Be clear about what you actually need AI to produce.
For example:
“Identify the five most important customer problems and provide evidence from the research for each.”
Or:
“Create six user stories and highlight any missing requirements that need clarification.”
Defining the output makes the response much more usable.
9. Format
Tell AI how you want the response presented.
For example:
- table
- report
- presentation
- user stories
- OKRs
- roadmap
- checklist
The right format can make the output much easier to take into your next activity.
10. Tone
Think about how the response needs to sound.
A document for senior leadership may need to be concise and business-focused. A customer communication may need to be simple and reassuring. A working document for your product team may need more detail.
Tell AI who will read the output and what tone is appropriate.
11. Success Criteria
Finally, explain what a good answer needs to achieve.
For example:
“The recommendations should be supported by customer evidence, achievable within the existing platform and clearly linked to our onboarding objective.”
Success criteria give AI something to shape and evaluate its response against.
You don’t need all 11 every time
This is important.
Structured prompting should not become a checklist that you mechanically complete before every AI conversation.
If you simply want AI to summarise a meeting, you probably don’t need 11 layers of information.
But imagine you want AI to help you:
- analyse a complex business problem
- develop a product strategy
- review a large set of requirements
- create an important product roadmap
- prepare recommendations for senior stakeholders
For these tasks, richer context becomes much more valuable.
Start with what the AI needs to know. Add more structure when the task’s complexity and importance justify it.
From better prompts to better outcomes
The real value of structured prompting is not creating longer prompts.
It is reducing the amount AI has to guess.
When AI understands the business context, users, problem, goal, constraints and expected output, its response has a much better chance of being relevant to your actual work.
The principle is simple:
Better context → Better outputs → Better outcomes
DSM Tip
Structure should help you think, not complicate prompting.
Start with a simple prompt. For more complex work, add the context that matters. Define the problem clearly, give AI relevant information and tell it what success looks like.
