The same AI model can produce brilliant or useless output depending entirely on how you prompt it. Prompt engineering is not magic words; it is clear communication with a system that takes you literally. This guide gives you a repeatable framework that works across ChatGPT, Claude and other models.
1. Assign a role
Opening your prompt by assigning a role primes the model to respond with the right expertise and tone. "You are a senior financial analyst" produces different output than no role at all. The model adopts the vocabulary, depth and perspective of that persona.
- Choose a role that matches the expertise you need.
- Be specific: "veteran copywriter for SaaS brands" beats "writer."
- The role sets default assumptions, so pick it deliberately.
2. Provide context
Models have no idea about your situation unless you tell them. Provide the background that changes the answer: your audience, your goal, relevant constraints and any prior decisions. Without context, the model guesses, and its guesses are generic.
For example, asking for "a marketing email" yields filler, while asking for "a re-engagement email to lapsed users of a budgeting app, friendly tone, goal is to drive logins" yields something usable. Context is the difference between generic and tailored.
3. Define the task
State exactly what you want the model to do, using a clear action verb. "Summarize," "rewrite," "compare," "generate" and "critique" each set a precise expectation. Avoid stacking multiple unrelated tasks in one request, which dilutes focus.
Phrase instructions positively. Tell the model what to do rather than what to avoid, because positive instructions are easier to follow. If something must be excluded, state it clearly and put it where it will not be lost.
4. Specify the format
The output format is where many prompts fall short. If you want a table, say so. If you want five bullet points under 20 words each, say so. Specify length, structure, tone and any required sections. The model will match a clear specification closely.
Format instructions also make output easier to use downstream. Requesting JSON, markdown or a specific template means you can drop the result straight into your workflow without reformatting.
5. Add examples
When style or structure matters, examples outperform description. Show the model one or two samples of exactly what good output looks like, and it will pattern-match. This technique, sometimes called few-shot prompting, dramatically improves consistency.
Examples are especially valuable for tone, formatting and edge cases that are hard to describe in words. A single well-chosen example can replace a paragraph of instructions and produce more reliable results.
When examples help most
- Matching a specific writing voice or style.
- Producing a precise data structure or template.
- Handling tricky cases the model tends to get wrong.
6. Iterate and refine
Your first prompt rarely produces perfect output, and that is normal. Treat prompting as iteration. When the result misses, diagnose why: was the context thin, the format unclear, the task ambiguous? Change one element and test again.
Keep prompts that work in a personal library so you never solve the same problem twice. Over time you build a toolkit of proven prompts for your recurring tasks, which is the real productivity payoff of prompt engineering.
Tips for reliable prompts
- Front-load constraints so they are not buried.
- State instructions positively for easier compliance.
- Cut filler that does not change the answer.
- Show examples when format or style matters.
- Save winning prompts as reusable templates.
Conclusion
Effective prompting is a learnable skill built on a simple framework: assign a role, provide context, define the task, specify the format, add examples and iterate. Apply these six elements consistently and the model's output becomes predictable and useful. The words you choose are the steering wheel; learn to use them and the AI does exactly what you need.
