How to write prompts that produce useful results

Why prompt quality is decisive

Prompt engineering is the ability to structure instructions for a generative model in a way that produces high-quality, highly relevant outputs. Gartner includes it among the most impactful AI skills an organization can develop in its workforce, and in the Hype Cycle for Generative AI 2025, it classifies it as an early mainstream technology with penetration already exceeding 20% of the target audience.

Beginner users tend to treat GenAI as if they were conversing with a person. In reality, generative models require precise instructions: what to do, how to do it, in what format, for which audience, in what tone. The more context provided, the more focused and useful the output.

The ReFLECT framework

Gartner developed the ReFLECT framework to standardize the structure of an effective prompt. The acronym stands for Role, Format, Language, Example, Context, and Task. Each component helps steer the model toward the desired output.

Role defines who the model should be in generating the response, both as a persona and as the type of audience the output is intended for. An example: "You are an expert data analyst. The audience is a manager with no technical background." Format specifies the output structure: list, table, paragraphs, step-by-step guide. Language indicates the register and tone: formal, direct, persuasive, conversational. Example provides a concrete reference of the expected result. Context includes all necessary background material: documents, prior data, constraints. Task defines the specific action required with a precise verb.

The role of action verbs

Gartner identifies a second high-impact lever: action verbs in prompts. Using precise verbs significantly improves the quality and relevance of outputs because they communicate to the model exactly what cognitive operation is required. The main categories are: recommend, propose, optimize, improve for recommendation tasks; prioritize, organize, categorize, sort for organizational tasks; create, generate, design, develop for creative tasks; explain, clarify, simplify, summarize for explanatory tasks; analyze, evaluate, compare, assess for analytical tasks.

A prompt that says "tell me something about Paris" produces a very different output from one that says "describe the culture of Paris comparing it to other French cities in a 150-word paragraph for an audience of experienced travelers."

The prompt as an iterative dialogue

It's not necessary to include every detail in the initial prompt. Generative models support progressive refinement through successive prompts. If the first output has the right content but the wrong format, a second prompt like "convert the bullet points into a two-column table" is enough to reorient the model. This iterative approach reduces frustration for beginner users and accelerates the path to effective use.

A useful test for evaluating prompt quality is to ask yourself: if I gave this same instruction to a new hire on their first day, with no further explanation, what would they produce? If the answer is "something random," the prompt needs more context.

From individual skills to organizational capabilities

Prompt engineering becomes even more powerful when it stops being an individual skill and becomes an organizational capability. Companies that build structured training programs, distribute shared frameworks, and collect best prompts in repositories accessible to all teams are accelerating their AI adoption systematically, not randomly.