Text Generation Best Practices

Learn prompting technique and best practices to make best use of text generation models

Quickstart

Write what you want, not what to do: Instead of “Write me an essay,” describe the output you want (“An essay in 3 paragraphs explaining the pros and cons of solar energy”).
Set context upfront: Who is the audience? What’s the tone? Formal, casual, persuasive, technical?
Be specific about format: Say if you want bullet points, numbered steps, story, email draft, or a Q&A style.
Start simple and refine: Begin with a clear, minimal prompt, then add details like tone, style, examples, or constraints.
Add constraints for clarity: Mention word count, sentence length, or specific sections (“Limit to 200 words,” “Use 3 bullet points”).
Iterate: If the response isn’t right, tweak details—shorten or restructure the prompt.

What Gives Best Results

Strategy	Why It Helps	Tips
Clarity & specificity	Reduces ambiguity, guides the AI to your intent	Instead of “Write about AI,” say “Write a 150-word blog intro about how AI speeds up drug discovery.”
Audience + tone	Ensures content matches purpose	“Explain blockchain to a 12-year-old” vs. “Explain blockchain to a CTO.”
Structured format	Models follow explicit structure better	“Give me: 1. Hook, 2. Problem, 3. Solution.”
Examples in prompt	Models mimic style if shown	“Write a product description like: [insert short example].”
Constraints (length, style)	Keeps response within usable bounds	“Keep under 100 words. Use persuasive language.”
Iterative refinement	Rarely perfect first time	Ask: “Make it more formal” / “Summarize shorter.”

What to Avoid

Vague or abstract requests: “Write something inspiring” → too broad.
Overloading multiple tasks: “Write an essay, a poem, and a LinkedIn post in one go.”
Conflicting instructions: “Be formal but casual, technical but simple.”
Overly long adjective lists: “Professional, inspiring, casual, modern, poetic, witty…”
Command-style or conversational phrasing: Avoid “Can you please write…”; instead state desired output.
Unclear audience: Without target reader info, results may miss the mark.

Examples

Example 1 – Blog Intro

Prompt:


Write a 120-word blog introduction about the role of generative AI in healthcare.
Use a professional but engaging tone, highlight efficiency in medical research, and end 
with a question that makes the reader curious.

Example 2 – Email Draft

Prompt:


Draft a short, polite follow-up email to a client named Priya who hasn’t responded 
for a week. Mention the proposal sent last Tuesday, ask if she had time to review it, 
and keep tone respectful but proactive

Example 3 – Creative Story

Prompt:


 
Write a 200-word short story in the style of magical realism. 
A boy discovers a talking crow in an old city library. 
Keep it descriptive and atmospheric, with a bittersweet ending

Technical Terms to Know (Useful for Better Quality Prompts)

Tone – Formal, casual, persuasive, academic, humorous, neutral, etc.
Voice – First-person (“I/we”), second-person (“you”), third-person (“they”).
Style – Narrative, descriptive, expository, persuasive.
Register – Level of formality (business-like, conversational, scholarly).
Format constraints – Word/character limit, bullets, numbered steps, dialogue, paragraphs.
Audience specification – Define who’s reading (experts, children, general public).
Temperature (if configurable) – Controls creativity: low = precise, high = imaginative.
Top-p (nucleus sampling) – Another knob for randomness/creativity.
Few-shot prompting – Giving short examples to steer output style/quality.
Chain-of-thought prompting– Asking the model to explain reasoning step by step.
Negative prompting (if supported) – Explicitly stating what not to include (“Do not mention prices”).

⚡Tip: Think like an editor giving instructions to a writer. The clearer your instructions, the better your text generation results.