Most people write AI prompts the way they’d type a Google search. That’s why they get mediocre results.
AI models aren’t search engines. They’re collaborative thinking partners. And like any good collaboration, the quality of what you get back depends entirely on what you bring to the table.
Here are the 7 rules that separate forgettable AI outputs from genuinely useful ones.
Rule 1: Assign a Role Before You Ask Anything
The single highest-leverage change you can make to any prompt is to open with a role assignment.
Without role: “Give me feedback on my business plan.”
With role: “Act as a venture capitalist who has funded 50+ startups. Your job is to give brutal, honest feedback on business plans — the kind you’d give in a real partner meeting. Review my business plan and tell me the 3 things most likely to kill this company.”
The role shapes the AI’s entire frame of reference — its vocabulary, the depth of its expertise, even its tone. Always start with who you need.
Rule 2: Specificity Beats Brevity Every Time
More words in your prompt almost always means better output. Precision isn’t verbose — it’s efficient.
Vague: “Write a marketing email.”
Specific: “Write a 200-word email promoting a 48-hour flash sale on noise-cancelling headphones. Audience: remote workers aged 28-45 who complain about home office distractions. Tone: friendly and slightly urgent. Include: a subject line, a 3-sentence opening that acknowledges their pain point, the 20% discount offer, one social proof element, and a CTA button text. No exclamation marks.”
The second prompt takes 30 more seconds to write. It saves 20 minutes of editing.
Rule 3: Specify the Output Format Explicitly
If you don’t tell an AI how to structure its response, it will make a choice — and it might not be the one you need.
Always include one of these:
- “Give me your answer as a numbered list”
- “Format this as a table with columns for [X, Y, Z]”
- “Write this as a 3-paragraph essay with a clear thesis”
- “Give me only the bullet points, no intro or explanation”
- “Structure this with H2 headers for each main section”
Format instructions eliminate the most common frustration in AI use: getting the right information in the wrong shape.
Rule 4: Tell It What to Avoid
Constraints are as powerful as instructions. Telling an AI what NOT to do is often the fastest path to what you want.
Examples that work:
- “No corporate jargon”
- “Don’t recommend [competitor products]”
- “Avoid clichés like ‘at the end of the day’ or ‘think outside the box'”
- “Don’t start any sentence with ‘I'”
- “No bullet points — use full paragraphs”
- “Don’t give me platitudes. Only give me specific, actionable advice.”
Rule 5: Give It the Right Context
An AI can only work with what you give it. The more relevant background you provide, the more tailored the response.
Ask yourself before every prompt: “What would a human expert need to know before they could help me with this?”
For business tasks, that usually means: your industry, company size, target audience, budget constraints, timeline, and what you’ve already tried.
For writing tasks: the audience, publication, tone, word count, goal of the piece, and examples you like.
Rule 6: Ask It to Think Step by Step
For any prompt involving reasoning, analysis, or complex decisions, explicitly asking the AI to “think step by step” dramatically improves output quality.
This is called Chain-of-Thought prompting, and the research behind it is solid: models that show their reasoning make fewer errors.
Instead of: “What’s the best pricing strategy for my SaaS product?”
Try: “Think step by step about the best pricing strategy for my SaaS product. First, analyze my situation [details]. Then consider 3 different pricing approaches and their tradeoffs. Then give me a recommendation with your reasoning.”
Rule 7: Always Iterate — The First Response Is a Draft
The biggest mistake people make with AI: treating the first response as the final answer.
The best AI users treat every first response as a draft. They follow up with:
- “Make this 30% shorter without losing the key points”
- “This is too formal — rewrite in a more conversational tone”
- “The third point is weak. Expand it with a concrete example”
- “Now do the same thing but for an audience that’s skeptical of AI”
- “What are the 3 biggest weaknesses in the analysis you just gave me?”
Iteration is where the real value lives. Most people quit one step too early.
Putting It All Together: A Template
Here’s a universal prompt structure that incorporates all 7 rules:
Act as [specific role with relevant experience].
Your task is to [specific action] for [specific context/audience].
Background information: [give relevant details — the more specific, the better].
Format your response as [specify: list, table, essay, outline, etc.].
Constraints:
- [What to avoid #1]
- [What to avoid #2]
- [Length/tone requirement]
Think through this step by step before giving your final answer.This template won’t work perfectly for every situation — but it’s a strong starting point. Customize, remove, and add sections based on what your specific task needs.
One More Thing
These rules aren’t complicated. The challenge is making them habits — using them automatically, on every prompt, without thinking about it.
The fastest way to build that habit? Practice on real tasks. The next time you open ChatGPT, Claude, or Gemini, run your prompt through these 7 rules before you hit send. Do that for a week and it becomes automatic.
More examples, templates, and deep dives — every week at PromptRefinery.ai.