The Complete ChatGPT Prompt Guide
There are dozens of prompting strategies, but most of them reduce to a handful of core techniques that cover nearly every situation you will encounter. This guide covers those techniques with concrete examples you can adapt immediately.
Start with the strategy that fits your task. Master it before moving to the next.
The three foundational prompt types
Every advanced prompting technique builds on one of these three approaches. Get comfortable with all three before exploring more complex methods.
Zero-shot
Give the model a task directly, with no examples. Works for clear, well-defined requests where the expected output format is obvious.
"Translate the following text into Spanish: 'The meeting has been rescheduled to Thursday at 3pm.'"
Few-shot
Provide two or three examples of the input-output pattern you want before giving the model the actual task. Use this when the format or style is non-obvious.
Product: Wireless headphones - Description: Crisp audio, 30-hour battery, folds flat for travel.
Product: Standing desk - Description: Adjustable height from 60-120cm, switches in 3 seconds.
Product: Mechanical keyboard - Description:
Chain-of-thought
Ask the model to reason through a problem step by step before giving its answer. Use this for multi-step problems, analysis tasks, or decisions where the reasoning process matters as much as the conclusion.
"A company wants to reduce customer churn. Think through the possible causes step by step, then recommend the three most likely interventions in order of impact."
Role assignment: giving the model a perspective
Asking the model to take on a specific role or perspective consistently improves the relevance and tone of its responses. The role anchors the model's vocabulary, assumptions, and level of technical depth.
"As a university professor giving a talk to a general audience, explain quantum computing without using technical jargon. Use one analogy."
"Act as a senior software engineer reviewing this code for security vulnerabilities. List issues by severity, starting with critical."
Role assignment is especially useful when you need the model to match a specific expertise level: a response written for a medical professional reads very differently from one written for a patient, even if the underlying information is the same.
Output format specification
One of the most reliable improvements you can make to any prompt is specifying exactly what format you want the output in. When you leave format unspecified, the model picks one that seems plausible, which is often not what you need.
Bullet list
List the five main causes of deforestation as bullet points, one sentence each.
Numbered steps
Give me step-by-step instructions to set up a Python virtual environment on Windows. Number each step.
Table
Compare the three main cloud providers (AWS, GCP, Azure) on: pricing model, free tier limits, and strongest use case. Format as a table.
Word limit
Summarize the following article in exactly 150 words. Do not go over.
Template fill
Write a cold outreach email using this structure: [Problem statement] / [What we do about it] / [One proof point] / [Call to action]. Keep it under 120 words.
Troubleshooting common output problems
The same five problems account for most unsatisfying ChatGPT responses, and each has a direct fix.
Problem: Response is too vague or generic
Fix: Add specificity: name the domain, the audience, the exact task, and the constraints. Replace open questions with closed, specific ones.
Problem: Response is too long and unfocused
Fix: Specify a word limit and ask for the key points only. Add: “Be concise. No preamble, no summary at the end.”
Problem: Response misunderstands what you wanted
Fix: Add two or three examples of what the correct output looks like. Switch from zero-shot to few-shot.
Problem: Response cuts off midway
Fix: Add “Complete the full response without stopping” to the prompt, or follow up with “Continue from where you left off.”
Problem: Response repeats the same information
Fix: Explicitly tell the model: “Do not repeat points already made. Each point should add new information.”
Negative constraints: telling the model what not to do
Most prompt advice focuses on what to ask for. Negative constraints, instructions about what to exclude or avoid, are equally useful and often faster at fixing recurring problems.
The reason they work: the model has learned from enormous amounts of text that certain response patterns are common. Disclaimer-heavy legal-sounding language, hedging phrases like "it's worth noting", summaries that restate the question before answering. These patterns show up by default because they appear frequently in its training data. Negative constraints rule them out explicitly, which is faster than trying to describe the alternative.
Do not include any disclaimers or caveats.
Removes legal-style hedging from responses about health, finance, or legal topics.
Do not begin with a restatement of the question.
Cuts the filler opening that wastes the first sentence of many responses.
Do not repeat information already stated in a previous point.
Prevents the padding that inflates long responses.
Do not use phrases like 'it is important to' or 'it is worth noting'.
Removes hedging language that makes confident information sound uncertain.
Do not suggest consulting a professional for this type of question.
Useful when you want the model to engage directly with a topic rather than deflect.
The most effective prompts often combine positive instructions (what to produce) with negative constraints (what to avoid). This gives the model a precise target rather than just a direction.
Prompt chaining: connecting outputs into workflows
A single prompt handles a single task. For complex work that involves multiple steps, prompt chaining lets you build a sequence where the output of one prompt becomes the input of the next.
For example, a content research workflow might chain three prompts: first, generate an outline for an article on a given topic; second, expand each section of that outline into a full paragraph; third, rewrite the full draft in a specified tone and word count. Each step produces a focused output that feeds cleanly into the next step.
SIMPLE CHAIN EXAMPLE
Step 1
"List the five most common objections customers have when buying project management software."
Step 2
"Take objection #3 from the list above. Write a two-paragraph response that addresses it directly, using a specific case study format."
Step 3
"Rewrite the response above as a FAQ entry: a one-sentence question followed by a three-sentence answer. No marketing language."
Chaining works because it breaks a complex task into steps that the model can handle well individually. A single prompt that tries to do all three steps at once produces a weaker result than three focused prompts in sequence. The tradeoff is time: chained prompts take more turns. For high-quality outputs on important tasks, the extra turns are worth it.
Using prompt templates for repeatable tasks
For tasks you do regularly, a prompt template saves time and produces more consistent results. A template is a reusable prompt with bracketed placeholders you fill in each time.
When I first started using ChatGPT for data analysis summaries, I rewrote the prompt from scratch each time. The outputs were inconsistent and I lost time editing them to fit the same format. Once I wrote a single template with placeholders for the dataset name, audience, and key metrics, I got consistent, usable outputs every time. The template took 20 minutes to write and has saved hours since.
ANALYSIS TEMPLATE
Analyze the following [type of data]: [insert data or description]
Provide:
1. A 2-sentence overview
2. Three key findings
3. One recommendation based on the findings
Audience: [target audience]. Length: [short/medium/long].
What to do now
Pick one recurring task where you use ChatGPT today, and write a template for it using the format: task description, context, output format, constraints, and word limit. Use it for the next five times you do that task and refine it as you go.
Practical Prompt Engineering by Vajo Lukic covers all of these techniques in depth, plus prompt chains, recursive prompts, prompt tuning, and ethical AI practices, with 250+ ready-to-use templates across 12 categories including content creation, coding, data analysis, and creative writing. Get the book here.
Related guides
- Zero-Shot vs Few-Shot Prompting Explained- Deep dive on the two foundational strategies
- What Is Prompt Engineering?- The broader skill this guide belongs to
- ChatGPT for Beginners: Getting Better Results- Start here if the techniques above feel advanced