Most people’s AI prompts are terrible. They type “write me a blog post about dogs” and wonder why the output is generic fluff. The difference between a mediocre AI user and a great one isn’t the tool — it’s the prompt.
This guide covers the prompt patterns that consistently produce high-quality results across ChatGPT, Claude, Midjourney, and every other AI tool.
The 5 Elements of a Great Prompt
Every great prompt has most of these elements. The more you include, the better the output:
- Role — Who should the AI be?
- Task — What exactly should it do?
- Context — What background does it need?
- Constraints — What are the rules?
- Format — How should the output look?
Example: Before and After
Bad prompt:
Write a blog post about productivity
Good prompt:
You are a productivity coach who writes for remote workers aged 25-35. Write a 1,200-word blog post about time-blocking for productivity. Include 3 specific techniques, real-world examples, and a 5-step morning routine. Use a conversational tone with short paragraphs. No bullet points — use numbered steps instead. End with a 1-paragraph summary.
The second prompt gives the AI role, task, context, constraints, and format. The output will be dramatically better.
Prompt Patterns That Work
1. The Expert Pattern
You are a [role] with [X years] of experience in [field].
[Task description].
Consider: [key considerations].
Avoid: [common mistakes].
Format: [output format].
Example:
You are a conversion rate optimization expert with 10 years of experience in e-commerce. Review this landing page copy and suggest 5 specific improvements. Consider mobile users and checkout friction. Avoid generic advice like “make it shorter.” Format as a numbered list with before/after examples.
2. The Step-by-Step Pattern
Think through this step by step:
1. [First step]
2. [Second step]
3. [Third step]
Then [final output].
Why it works: AI models produce better reasoning when asked to “think step by step.” It forces the model to show its work, which reduces errors.
3. The Few-Shot Pattern
Give 2-3 examples of the input → output you want, then ask for the same pattern:
Example 1: [input] → [output]
Example 2: [input] → [output]
Now do: [new input]
Example:
Write product descriptions matching this style:
“The Aero Pro running shoe. 7.2 oz of engineered mesh on a carbon-fiber plate. Built for 5K race day.” “The Studio headphones. 40mm drivers in a titanium frame that blocks 98% of ambient noise. Made for deep work.”
Now write: “The Summit backpack”
4. The Constraints Pattern
Write [task] with these rules:
- No more than [X] words
- No [forbidden words/phrases]
- Must include [required elements]
- Tone: [tone description]
- Audience: [audience description]
5. The Revision Pattern
Don’t accept the first output. Ask for revisions:
That's good, but:
- Make it more [concise/punchy/detailed]
- Remove [specific issue]
- Add [missing element]
- Change the tone to be more [professional/casual/authoritative]
Pro tip: The best outputs come from 2-3 rounds of revision, not from a single perfect prompt.
Platform-Specific Tips
ChatGPT Prompts
- Use system prompts for consistent behavior across a conversation
- Ask for specific lengths (“1,200 words” not “a long article”)
- Say “don’t use these words” instead of “don’t be generic” — concrete instructions work better
- Use “act as a [role]” to set expertise level
- For coding: specify the language, framework, and error messages
Claude Prompts
- Claude follows detailed instructions better than ChatGPT
- Give it documents to reference — it handles long context well
- Use XML tags to structure inputs:
<context>...</context> - Ask Claude to critique its own output — it’s surprisingly good at self-evaluation
Midjourney Prompts
- Put the subject first, then style, then parameters
- Use style references (–sref) for consistent aesthetics
- Use character references (–cref) for consistent characters
- Add –stylize (0-1000) to control artistic freedom
- Add –weird (0-3000) for more unusual results
- End with –ar for aspect ratio (–ar 16:9, --ar 9:16)
Example Midjourney prompt:
professional product photo of a ceramic coffee mug on a wooden desk, morning sunlight, shallow depth of field, warm tones, shot on 85mm lens --ar 4:5 --stylize 250 --v 6.1
Common Mistakes
1. Being too vague “Write something good” → “Write a 500-word email to existing customers announcing a 20% spring sale. Tone: friendly but not overly casual. Include a clear CTA to shop the sale.”
2. Not specifying format “Give me ideas” → “Give me 10 ideas, each with a title and 2-sentence description, formatted as a numbered list.”
3. Accepting the first output AI gets better with iteration. Revise 2-3 times.
4. Ignoring the audience “Explain machine learning” → “Explain machine learning to a marketing manager who has never written code. Use analogies from advertising and data analytics.”
5. Not giving examples Showing the AI exactly what you want (few-shot prompting) produces dramatically better results than describing it.
The Prompt Template Library
Save these and customize for your use case:
Blog Post:
You are a [niche] expert writing for [audience]. Write a [length]-word blog post about [topic]. Include [specific sections]. Use [tone] tone. Start with a hook that [specific hook type]. End with [CTA or summary]. No filler paragraphs.
Email:
You are a [role] writing to [audience]. Write a [length]-word email about [topic]. Subject line should be [under 50 characters / question / benefit-driven]. Body should [specific structure]. CTA: [specific action]. Tone: [tone]. Avoid: [specific words/phrases].
Product Description:
Write a [length]-word product description for [product name]. Target audience: [audience]. Key features: [list 3-5 features]. Tone: [tone]. Include [specific elements like dimensions, materials, warranty]. End with [CTA].
Social Media Post:
Write [number] [platform] posts about [topic]. Each post should be [length] characters max. Include relevant emojis. First line must be a hook. End each post with a CTA to [action]. Tone: [tone]. Hashtags: [number] relevant tags.
For more AI tool guides, visit aiverdict.co — the final word on AI tools.