Writing AI Prompts: Step-by-Step

Step 7: Explore More (Optional)

When ready, try personas, structured reasoning, and style controls to push your results further.

Goal: Add advanced options that improve relevance and presentation without rewriting your whole prompt.
Levers: 🎭 Personas · 🧭 Reasoning structure · 📰 Style & tone · 🧱 Output formats · ✅ Quality checks

🎭 Personas (Role Prompting)

Ask the AI to adopt a role to shape assumptions, vocabulary, and depth.

• Career Coach: “Act as a career coach. For a first-gen student applying to marketing internships, outline 3 resume tweaks and 2 interview talking points.”
• Librarian/Info Pro: “You are a university business librarian. Recommend 5 databases for apparel e-commerce research and explain why each fits.”
• Skeptical Reviewer: “Act as a skeptical reviewer. Identify weak claims and suggest stronger evidence or sources.”

Tip: Add audience, constraints, and success criteria: who it’s for, limits (time/length), and what “good” looks like.

🧭 Structured Reasoning (High-Level)

Encourage transparent, high-level reasoning to improve reliability.

• Outline then answer: “Briefly list 3–5 steps you’ll take, then provide the final answer.”
• Assumptions first: “List key assumptions (bullets), then proceed with the solution.”
• Plan–Execute: “Propose a plan (bullets), wait for my OK, then execute.”

Note: Some models don’t reveal detailed internal chain-of-thought. Ask for a concise rationale or high-level outline instead of step-by-step inner reasoning.

📰 Style & Tone Controls

Guide the voice, length, and structure.

• Tone: “Write this like a news brief for campus readers.”
• Length: “Limit to 120–150 words.”
• Structure: “Use H3 headings and bullet lists.”
• Formatting: “Include a 3-column HTML table (Name, Why it’s useful, Link).”
• Audience fit: “Plain language for first-year students; define jargon.”

Examples: “Write this like a news brief…”, “Summarize for a dean…”, “Rewrite in active voice…”

🧱 Output Formats (Make it Paste-Ready)

• JSON: “Return JSON with fields: title, audience, bullets[], sources[].”
• CSV: “Return CSV with columns: Database, Coverage, Why_Useful, Link.”
• Table/HTML: “Produce an HTML table suitable for LibGuides.”
• Code Block: “Wrap the snippet in a single HTML code block.”

Example JSON prompt → “Summarize 3 credible sources. Return:
{
  "title": "string",
  "audience": "string",
  "bullets": ["string", "string", "string"],
  "sources": [{"title":"", "publisher":"", "year": "", "url": ""}]
}
”

✅ Quality Levers & Guardrails

• Criteria: “Use peer-reviewed sources from 2020–present; include DOIs where available.”
• Self-check: “Provide a brief accuracy check and note weak points.”
• Citations: “Include APA citations and links.”
• Clarify first: “Ask up to 3 clarifying questions before answering.”
• Boundaries: “Exclude personal data; don’t infer identities.”

🧰 Copy-Ready Mini-Prompts

• “Act as a career coach. Draft 5 resume bullets for an entry-level data analyst with retail experience. Limit to 12 words each.”
• “Briefly list your approach (3 bullets), then produce a 150-word overview of 2025 e-commerce trends for apparel.”
• “Write this like a news brief. Headline (≤12 words) + 3 bullets + one-sentence kicker.”
• “Return JSON with fields: ‘key_point’, ‘example’, ‘link’. Include 3 items.”
• “Ask 2 clarifying questions. After I respond, generate a table (Tool, What it’s for, Link).”

🧪 Try It

Pick a recent topic you’ve used (e.g., “college students’ AI use”). Apply one persona, one reasoning instruction, and one style control. Compare the result to your Step-6 version and note differences in clarity, structure, and usefulness.