The three things every prompt must cover (What, How, Who)

Here is something that might surprise you: most people, even regular AI users, only communicate about one-third of what the AI needs to do its best work. They describe what they want — "write me an email," "summarise this article," "create a lesson plan" — and then wonder why the result feels generic.

The reason is straightforward. Telling an AI what you want is like telling a taxi driver you want to go "somewhere nice." You will end up somewhere, but probably not where you had in mind.

The Prompt Trinity is a simple framework that ensures you communicate three essential dimensions every time you work with AI. Think of it as a three-legged stool — remove any leg and things get wobbly.

Leg 1: What (Output Description)

This is the dimension most people already use, but rarely with enough specificity. The "What" is your description of the desired output. To make it precise, consider these five elements:

• Format: Do you want a bullet list, a table, a narrative paragraph, a script, a JSON file, a slide outline?
• Length: Should this be 100 words or 1,000? Three bullet points or fifteen?
• Audience: Is this for a board of directors, a group of year 9 students, your Instagram followers, or a technical team?
• Style: Should it read like a news article, a casual blog post, an academic paper, or a conversational message?
• Purpose: What will you actually do with this output? Present it? Send it? Use it as a starting point for further work?

Weak "What":

Write me a summary of climate change.

Strong "What":

Write a 200-word summary of the key causes of climate change, formatted as three short paragraphs, suitable for a company intranet article aimed at non-specialist employees.

Notice how the strong version gives the AI five concrete anchors: length (200 words), structure (three paragraphs), topic focus (key causes), format (intranet article), and audience (non-specialists). The AI does not have to guess any of these, which means it spends its processing on quality rather than assumptions.

Leg 2: How (Process Guidance)

This is the dimension that most people miss entirely, and it is often the most transformative. The "How" tells the AI the approach you want it to take — the method, the reasoning process, or the steps to follow.

Think of it this way: if "What" is the destination, "How" is the route. Two different routes to the same destination can produce very different experiences.

Process guidance can take several forms:

• Step-by-step instructions: "First, identify the three main arguments. Then, evaluate each one for logical consistency. Finally, rank them by strength of evidence."
• Methodology: "Use the SWOT framework to analyse this business idea."
• Examples to follow: "Here is an example of the tone and structure I want: [example]. Now produce something similar for [new topic]."
• Reasoning requests: "Think through this step by step before giving your final answer."
• Sequencing: "Start with the conclusion, then work backwards to explain how you got there."

Without "How":

Analyse this business plan.

With "How":

Analyse this business plan using the following approach: First, summarise the core value proposition in one sentence. Then, identify three strengths and three risks. For each risk, suggest a specific mitigation strategy. Present this as a structured report with clear headings.

The difference is enormous. The first prompt will give you a generic, meandering response. The second gives the AI a clear process to follow, which means the output will be organised, thorough, and useful.

Leg 3: Who (Persona and Tone)

The "Who" dimension sets the character of the interaction. It covers two things: who the AI should be in this conversation, and what kind of interaction style you want.

This is not about making AI pretend to be a person. It is about calibrating the level of expertise, the communication style, and the relationship dynamic. Consider these spectrums:

• Expert ↔ Beginner: Should the AI respond as a specialist using technical terminology, or as someone explaining things in plain language?
• Formal ↔ Casual: Should the tone be professional and polished, or relaxed and conversational?
• Challenging ↔ Supportive: Should the AI push back on your ideas and play devil's advocate, or should it build upon and encourage your thinking?
• Concise ↔ Detailed: Should responses be tight and punchy, or thorough and expansive?

Without "Who":

Give me feedback on this marketing copy.

With "Who":

Act as a senior brand strategist with 15 years of experience in consumer goods marketing. Review this marketing copy and give me direct, candid feedback. Be specific about what works and what does not. Challenge any assumptions you think are weak. Keep your feedback concise — no more than five bullet points.

The persona instruction fundamentally changes the quality of the response. Without it, you get polite, surface-level feedback. With it, you get the kind of pointed, expert critique that actually helps you improve.

Putting the Trinity Together

When you combine all three dimensions, you create a prompt that leaves very little room for misinterpretation. Here is a complete example:

Who: You are an experienced curriculum designer who specialises in adult workplace learning.

What: Create a 90-minute workshop outline on giving effective feedback, designed for mid-level managers in a financial services company. The outline should include timing for each section, facilitator notes, and one interactive exercise.

How: Structure the workshop using the "Tell, Show, Do, Review" methodology. Start with a brief icebreaker, build to the core skill practice in the middle, and end with a personal action plan. For the interactive exercise, design a realistic role-play scenario relevant to financial services.

You do not need to label the three parts explicitly in every prompt — that was just for illustration. In practice, you weave them naturally into your request. The key habit is simply asking yourself before you hit send: Have I covered What, How, and Who?

Most of the time, if your result is disappointing, it is because you skipped one of these three legs. Add the missing dimension and try again — you will be surprised how much difference it makes.