Prompt Engineering Basics

Updated January 2026 · 7 min read

Prompt engineering sounds technical, but it's mostly common sense applied consistently. The five techniques below cover 90% of what actually improves AI output. You can use them with ChatGPT, Claude, Gemini, or any modern model.

What is prompt engineering?

It's the practice of writing instructions that get better, more reliable results from a language model. It's not a programming language — just a set of patterns that consistently work.

1. Role prompting

Tell the model who it should be. This shifts vocabulary, priorities, and the level of assumed context.

You are a senior copywriter at a B2B SaaS company. Rewrite this landing page headline to be clearer and more benefit-driven.

Without the role, you'd get a generic rewrite. With it, you get someone thinking about positioning and audience.

2. Few-shot examples

Show the model one or two examples of what you want. It'll copy the pattern.

Classify each sentence as positive, negative, or neutral. "The food was amazing." → positive "The service was slow but the food was okay." → neutral "The waiter forgot our order twice." →

This works far better than describing the categories in prose. It's also the fastest way to enforce a specific style.

3. Chain-of-thought

Ask the model to think step by step. For anything involving reasoning, math, or multi-step logic, this measurably improves accuracy.

A team of 3 designers finishes a project in 12 days. How long would 5 designers take? Think step by step before answering.

Modern models sometimes do this automatically, but asking explicitly still helps on harder problems.

4. Format specification

Tell the model exactly what shape the output should take. Tables, bullet lists, JSON, single paragraphs — all valid.

Summarize this article in exactly three bullet points, each under 15 words. Use no bullet emoji.

If you're piping output into another tool (or a human who's skimming), format matters more than content.

5. Constraints and negative instructions

Say what you don't want, not just what you do. Models are trained to be helpful, which often means verbose and hedgy unless you push back.

Explain what a reverse proxy is. Do not mention Nginx, Apache, or any specific product. Do not use analogies. Assume the reader is a junior developer.

Constraints are the difference between "informative" and "actually useful for me".

When to combine them

A strong prompt usually combines two or three. For example, role + format:

You are an editor at a tech publication. Review the paragraph below. Output a table with three columns: issue, severity (1-3), suggested fix. Do not rewrite the paragraph itself.

That's four techniques in 40 words. Powerful and simple.

The easy way

You don't need to remember all five. Epromptly takes a rough prompt and applies all of these automatically — role, structure, format, constraints, everything — based on what you're targeting (ChatGPT, Claude, coding, writing, etc.).

Paste your rough idea, pick a target and style, and you'll get a rewritten prompt you can send straight into any AI.

Skip the learning curve

Let Epromptly apply these techniques for you. Free, no signup, 10 uses per day.

✨ Enhance a prompt now