
A Practical Guide to the Business Change Lifecycle
22 November 2025AI is rapidly becoming part of everyday business analysis, whether we planned for it or not. Many professionals use Large Language Models (LLMs) like ChatGPT, Claude, Perplexity, Gemini, or Copilot as if they were advanced search engines: we ask a question and expect a clear, reliable answer. Sometimes that works—other times the output feels shallow, inconsistent, or simply wrong. The reason is simple: LLMs rely heavily on the clarity and structure of the instructions they receive.
Weak prompts force the model to guess. Strong prompts guide its reasoning. When your instructions are structured, the model delivers more accurate, reliable, and usable results. This is where the CRAFT framework comes in. It offers a simple, repeatable, and highly effective way to design prompts that work across all major models while allowing small adjustments for tool-specific behavior.
Why Prompt Structure Matters
LLMs interpret instructions based on patterns. If a prompt lacks focus or detail, the model fills in the gaps with assumptions. Structured prompts reduce that guesswork and help the AI understand:
- What background and context it needs,
- Which perspective or role to take,
- What tasks to perform,
- What format to produce,
- Who the content is intended for,
The clearer these elements are, the better the results.
The CRAFT Framework Explained
The CRAFT method organizes a prompt into five essential components that guide the model more effectively.
C: Context
Provides background and purpose, preventing invented details.
R: Role
Assigns the expert persona the model should adopt, shaping tone and depth.
A: Action
Describes the exact task, avoiding vague summaries.
F: Format
Defines how the output should be structured, improving clarity and usability.
T: Target Audience
Aligns the tone, complexity, and language with the intended readers.

All together, these elements form a blueprint that significantly improves prompt consistency and quality.
CRAFT in Practice: From Weak Prompt to Strong Output
A weak prompt such as “Explain systems thinking.” leaves too many decisions to the model. Tone, depth, structure, and audience all become guesses.
Using the CRAFT structure provides clarity:
Context: You are creating material for a business analysis workshop.
Role: Act as an experienced BA trainer.
Action: Define systems thinking and provide one simple example.
Format: Two short paragraphs followed by three bullet points.
Target Audience: Beginners.
This simple structure already produces more consistent, high-quality results.
If you want more, watch this video from Lawton Learns that explains in more details how this prompt works:
Why Different LLMs Need Slightly Different Prompts
Not all models respond to prompts the same way. Differences in training, architecture, and instruction tuning affect how they interpret your request.
For example:
- ChatGPT follows roles and structure very well.
- Claude tends to expand unless given tight boundaries.
- Gemini performs best with step-by-step instructions.
- Copilot prefers short, direct prompts.
- Perplexity excels at retrieval-style answers and concise summaries, but benefits from very clear task framing and explicit output instructions.
The strength of CRAFT is that it keeps your structure consistent while allowing you to adjust only the parts each model is most sensitive to.
Looking Ahead
As LLMs continue to grow in capability and variety, prompt engineering will shift from a “nice to have” to a core professional skill for business analysts. Models will become more specialized, organizations will demand more from AI-assisted work, and expectations for clarity, accuracy, and reliability will rise.
A structured method like CRAFT helps you stay ahead of that curve. It gives you a scalable, repeatable way to get better results from any model — today and in the future.
If you want to start applying CRAFT immediately, download the free CRAFT Prompt Template and use it in your next AI-assisted task. The more intentionally you prompt, the more value you unlock.




