How to Write Better AI Prompts: 7 Techniques Tested and Ranked
We ran 100 prompt comparison tests across Claude, ChatGPT, and Gemini to rank which prompt improvement techniques actually move the needle. Here are the results, ordered by impact.
Rank 1: Specificity Over Brevity (Impact: Very High)
The single biggest improvement comes from replacing vague language with specific details. "Write a blog post about productivity" produces generic content. "Write a 1,200-word blog post for mid-level marketing managers at B2B SaaS companies about protecting deep work time in an interrupt-heavy culture. Include 3 specific tactics, 1 counterintuitive insight, and a closing call-to-action to download a focus audit template" produces something publishable.
Rank 2: Define the Output Format (Impact: High)
Explicitly state what the output should look like. Instead of "Give me ideas," write "Give me 10 ideas formatted as: [Idea name] | [Why it works] | [Implementation difficulty: Easy/Medium/Hard]." Format specifications dramatically reduce editing time.
Rank 3: State the Audience (Impact: High)
"Explain how options pricing works" gets a generic explanation. "Explain how options pricing works to a 45-year-old real estate investor who understands risk but has no stock market experience and wants to learn about covered calls" gets an explanation that actually serves your reader.
Rank 4: Role Prompting (Impact: Medium-High)
Assigning expertise improves quality on specialized tasks. "As a senior UX researcher with 15 years of experience reviewing enterprise SaaS products, critique this user onboarding flow" produces more specific and actionable critique than "Critique this onboarding flow."
Rank 5: Chain-of-Thought Instruction (Impact: Medium-High)
"Think step by step" consistently improves accuracy on reasoning tasks, math, and multi-step decisions. Its effect on creative tasks is minimal, so apply it selectively.
Rank 6: Example-Based Prompting (Impact: Medium)
Providing 2–3 examples of desired outputs before your request aligns the model's output style. Most valuable for format-sensitive tasks (structured data, specific writing styles) where describing the format is harder than showing it.
Rank 7: Negative Constraints (Impact: Medium)
"Do not include generic advice," "avoid bullet lists," and "do not mention [topic]" help you avoid known failure modes. Useful as refinements but rarely a first-pass technique.
Bonus: The Iterative Approach
The highest-performing technique we tested is not a single technique at all — it is iteration. A first prompt rarely produces your best result. The workflow that consistently produced top outputs:
- Write a reasonable first prompt
- Get the output
- Identify the specific gap between output and ideal
- Write a refined follow-up: "The previous response was good but [specific issue]. Now [specific instruction to fix it]."
Three rounds of targeted refinement produces better results than trying to write a perfect first prompt. AI assistants are collaborative tools, not vending machines.
