Asking ChatGPT the Right Questions
On this page
The Art of Prompting: Get Exactly What You Want
The decisive skill with ChatGPT isn’t technical: it’s formulation. A good prompt—the established term for your message—separates the generic response you’ll toss from the one that saves you half a day. This lesson gives you a reproducible method applicable to any task.
The CRAFT Method
CRAFT is a five-letter mnemonic: Context, Role, Action, Format, Tone. It’s not an OpenAI standard, but a convenient way to remember the advice from their help center: be clear and precise, provide enough context, specify the tone, then refine. Each letter answers a question ChatGPT can’t solve alone.
Context provides background information: who you are, the situation, who the result is for. A prompt starting with “I’m an HR manager at a 200-person company…” already shapes the vocabulary and detail level of the response.
Role specifies the posture to adopt. “Act as a marketing strategy consultant…” doesn’t change the model’s knowledge, but it changes how it organizes it: priorities, blind spots it will seek, the register.
Action is the core of the prompt and uses a precise verb—write, analyze, compare, summarize, translate, list, explain, simplify. “Write a welcome email for new employees…” leaves no room for interpretation, unlike “help me with onboarding”.
Format describes the expected form: “…as bullet points, max 200 words, with a professional but warm tone.” Without this, you get three screens of text to edit yourself.
Tone, finally, sets the register and vibe: “Formal but accessible tone. No technical jargon.” It’s the instruction most often forgotten, and the one that decides if the text is ready to send.
Four Techniques That Change Results
Showing beats describing. Rather than explaining at length what you mean by “catchy,” give an example:
“Rewrite these headlines to make them catchier. Here’s an example of what I expect:
- Before: ‘Q1 2026 Quarterly Report’
- After: ‘Q1 2026: The 3 Numbers That Change Everything’
Now, rewrite these headlines: [your headlines]”
For complex tasks, impose a process rather than a result. Asking for a four-step analysis produces traceable reasoning, where you can challenge one step without restarting everything:
“Analyze this business plan in 4 steps: 1) identify strengths, 2) identify weaknesses, 3) evaluate financial feasibility, 4) give your final verdict with a score out of 10.”
Constraints narrow the target. The more numerical limits you set, the less the response digresses:
“Give me exactly 5 arguments, each in one sentence of 20 words max.”
Asking for multiple versions, finally, lets you choose instead of correct—often faster:
“Propose 3 versions of this email subject line: one formal, one casual, one punchy.”
Weak vs. Effective: Two Comparisons
Take the most mundane task: summarizing an article. The weak prompt is three words: “Summarize this article.” The effective prompt sets the scene: “Summarize this article in 5 bullet points. Each point must be one sentence max. Highlight numerical data. Target audience: my director who has 2 minutes to spend on it.” The first gives you an arbitrary-length summary to rework; the second, a deliverable.
Same gap with an email. “Write an email to my client” can’t produce anything usable. The effective version gives everything ChatGPT doesn’t know: “I need to inform my client Pierre (CTO at Nexus) that the project delivery will be delayed by 2 weeks due to an API integration issue. Write a professional email that acknowledges the delay, briefly explains the cause, proposes a new timeline, and reassures on final quality. 150 words max.”
Two pitfalls lie in wait. Vagueness, first—“help me with my project” produces nothing usable, because ChatGPT only knows what you’ve told it: in the conversation, in your custom instructions, or in what its memory retained. Overload next: a five-hundred-word prompt stacking fifteen instructions will inevitably see some ignored. When your request becomes a specification sheet, break it into successive steps.
Your Turn
CRAFT on a real task. Choose a task you actually need to accomplish today—an email, a summary, a presentation, an analysis. Write the full prompt reviewing the five letters of CRAFT, send it, then evaluate coldly: is the response usable as is, or does it need a follow-up? Refine if necessary.
Then redo the same exercise with a simple “do me a [task]” and compare the two results. The gap you see is exactly what you’ll gain every day by polishing your prompts—a skill that keeps improving with practice. When part of the context returns in all your requests (your job, your audience, your style), place it in your custom instructions rather than repeating it.
Official sources: Prompt engineering best practices for ChatGPT (OpenAI help center: clarity, context, tone, iteration) · Prompt engineering (OpenAI guide: role, instructions, examples, context) · ChatGPT Custom Instructions (permanent context)
Key Takeaways
- The CRAFT method (Context, Role, Action, Format, Tone) structures your prompts effectively
- Giving examples is often more effective than long descriptions
- Constraining the format (word count, number of points, structure) improves precision
- A good prompt is built and refined—iteration is normal
- Prompting is a skill that improves with daily practice