Beginner · 8 min

Prompt Engineering Basics

Five principles and the anatomy of a great prompt — with before/after examples.

5 Principles of Great Prompts

Vague prompts produce vague results. Specify format (bullet points, table, paragraph), length (under 200 words, 5 items), and what to include or exclude. The more constraints you add, the less the model has to guess.

Weak

Write me a marketing email.

Strong

Write a cold email to a freelance designer who hasn't replied in 2 weeks. Under 80 words. Casual tone. One clear CTA: schedule a 15-minute call. No subject line needed — just the body.

The Anatomy of a Prompt

Every great prompt contains some combination of these five elements. You don't need all five every time — but the more you include, the better the output.

Role— Sets the perspective and knowledge base.

You are a senior UX designer with 10 years of B2B SaaS experience.

Context— Gives the model your specific situation.

I'm redesigning the onboarding flow for a project management tool. Current drop-off rate is 68% at step 3.

Task— The clear, specific action you want taken.

Identify 3 possible reasons users are dropping off at step 3 and suggest one design change for each.

Format— Exact output structure you need.

Format as: Problem | Likely Cause | Design Fix. One row per issue.

Constraints— Boundaries that keep outputs practical.

Each fix must be implementable in one sprint. No redesigns requiring new user research.

See the Difference

A weak prompt vs the same request rebuilt with all five elements.

Before

Help me with my resume.

Platform-Specific Tips

ChatGPT, Claude, and Gemini are built differently and respond to prompts differently.

ChatGPT

Responds well to 'Think step by step' — add it before complex analytical tasks
Use 'Give me 3 options: conservative, moderate, and aggressive' to generate useful variation
Code Interpreter can run Python — ask it to analyze data or create charts, not just write code

Claude

Explicitly ask for direct recommendations — Claude hedges by default, so say 'Give me your actual recommendation in the first sentence'
Paste full documents rather than summaries — its 200K context window is a feature, use it
Ask it to 'think out loud' before answering complex questions to surface hidden assumptions

Gemini

Ask for 'search-grounded' answers when you need current information beyond its training data
Use Gemini within Google Docs and Gmail directly — the in-app integration is more powerful than the standalone interface
NotebookLM is separate and specialized: upload sources first, then interrogate them

Next: ChatGPT Mastery