Write prompts that actually work
A practical, no-fluff guide to the six dimensions of a great prompt — the same framework AURE uses to score and improve your prompts automatically.
What is prompt engineering?
Prompt engineering is the practice of writing inputs that reliably steer an AI model toward the output you want. The same model can produce a vague, generic answer or a sharp, useful one depending entirely on how the request is framed.
A good prompt removes ambiguity. It tells the model who to be, what to do, what context matters, how to format the answer, and what to avoid — so the model spends its capacity solving your problem instead of guessing at it.
Treat the model like a brilliant new teammate with no context on your project. Everything obvious to you must be made explicit.
1. Clarity — say exactly what you want
Start with a specific, action-oriented instruction. Replace vague verbs like 'help with' or 'look at' with precise ones: 'summarize', 'rewrite', 'classify', 'generate five'. State the deliverable in the first sentence.
Ambiguity is the single biggest cause of disappointing output. If a human would need to ask a clarifying question to act on your request, the model does too — and it will guess instead.
Before: 'Help me with my email.' After: 'Rewrite the email below to be more concise and friendly, keeping it under 120 words.'
2. Context — give the background
Models don't know your domain, audience, or goal unless you tell them. Supply the relevant background: what the task is for, who will read it, any constraints from your business, and the problem you're actually trying to solve.
Context is what turns a generic answer into one tailored to your situation. Paste the source material, describe the scenario, and name the outcome you're optimizing for.
Add a purpose line: 'This is for a landing page aimed at non-technical small-business owners.'
3. Role definition — assign an expert persona
Opening with a role — 'You are a senior financial analyst' or 'Act as an experienced copy editor' — primes the model to draw on the right knowledge, tone, and standards. It measurably improves relevance and depth.
Be specific about the expertise. 'You are a marketer' is weaker than 'You are a B2B SaaS growth marketer who writes for technical buyers.'
Pair the role with a standard: 'You are an accessibility expert reviewing against WCAG 2.2 AA.'
4. Output format — specify the shape
Tell the model how to structure the response: a Markdown table, JSON with named keys, a numbered list, bullet points, or prose of a certain length. Explicit formatting makes output predictable and easy to use downstream.
If you need machine-readable output, describe the exact schema and ask for that and nothing else — no preamble, no explanation.
'Respond with a JSON object with keys "title", "summary", and "tags" (array). Return only the JSON.'
5. Constraints — set the boundaries
Define the limits: word or character count, tone, reading level, style rules, things to avoid, and the audience. Constraints keep the model from drifting and make results consistent across runs.
Negative constraints ('do not use jargon', 'avoid marketing clichés') are as useful as positive ones.
'Keep it under 150 words, use a warm but professional tone, and avoid buzzwords like synergy or leverage.'
6. Examples — show, don't just tell
Few-shot prompting — including one or more examples of the input/output pattern you want — is one of the most powerful techniques available. Examples communicate tone, structure, and edge-case handling that would take paragraphs to describe.
Two to three well-chosen examples usually beat a long written specification. Make sure they're representative and correctly formatted, because the model will imitate them faithfully — mistakes included.
Show a sample 'input → ideal output' pair, then give the real input and ask for the same treatment.
Iterate and measure
Your first prompt is a draft. Run it, inspect where the output falls short, and adjust the weakest dimension — usually clarity or context. Small, targeted edits beat rewriting from scratch.
This is exactly what AURE automates: it scores a prompt across all six dimensions, shows you which ones are weak, and suggests concrete fixes — so you improve deliberately instead of by trial and error.
Paste any prompt into the AURE optimizer to get a dimension-by-dimension score and rewrite suggestions.
Put it into practice
Paste a prompt into AURE and see it scored across all six dimensions in seconds.
Open the optimizer