Everywhere you look, someone seems to be getting remarkable results from AI while you are left staring at a mediocre answer wondering what you did wrong. Most of the time, the gap is not the tool. It is the prompt. The same model can produce something generic or something genuinely useful depending entirely on how you ask.
This is a practical guide to asking well. You will get a clear definition of what a prompt is, the anatomy of one that works, the main types, a library of examples you can copy and adapt today, and the mistakes that quietly ruin most people's results. No jargon for its own sake, just what actually moves the output.
What is an AI prompt?
An AI prompt is the instruction or question you give a generative AI tool to tell it what you want. It can be a single sentence or a detailed brief, and its clarity largely determines the quality of what you get back.
When you type a prompt, the AI does not look up an answer. It predicts a response one piece at a time, using the patterns it learned during training and the context you provided. It reads not just your words but how they relate to each other, then generates something new that fits the request. This is why the same model can feel brilliant or useless: it is working from what you gave it, so a thin prompt produces a thin result.
Here is a simple example of a clear prompt:
Write a Python function to calculate the Fibonacci sequence up to the 10th term. Add comments explaining each step, and keep it efficient.
Notice what it does. It states the task, sets a parameter (the 10th term), and adds requirements (comments, efficiency). That specificity is the whole game, and everything below is about doing it deliberately.
The anatomy of a good prompt
The strongest prompts tend to include some combination of five ingredients. You do not need all of them every time, but knowing them gives you a checklist when an answer disappoints.
Role
Who it should be
Tell the AI the perspective to take. "You are a senior copywriter" sets a very different output than no role at all.
Task
What to do
State the action plainly and specifically. One clear task beats three vague ones bundled together.
Context
What it needs to know
Audience, background, constraints, and any source material. This is where most weak prompts fall short.
Format
How it should look
Length, structure, tone. "A bulleted list under 150 words" removes guesswork.
A fifth ingredient, examples, is optional but powerful: showing the AI one or two samples of what good looks like (called few-shot prompting) often beats describing it. Put the ingredients together and a vague request becomes a reliable one:
Write a product announcement email.
You are a B2B marketer. Write a product announcement email to existing customers introducing our new scheduling feature. Explain that it removes back-and-forth booking, keep it under 120 words, warm but professional, and end with a clear call to try it.
Same tool, same model, very different result. The second one will need far less editing, and the only thing that changed was telling it what you actually wanted.
Best practices that consistently help
- Be specific. Replace "write about marketing" with the topic, angle, audience, and length. Vagueness is the number one cause of disappointing output.
- Give context. Paste the document, name the audience, state the goal. The AI cannot read your mind, only your prompt.
- Use positive directives. Tell it what to do rather than what to avoid. "Use plain language" works better than "don't be too technical."
- Show, don't just tell. Include an example of the style or structure you want when it matters.
- Iterate. Treat the first answer as a draft. Say what to change and ask again. The second and third replies are usually much better.
- Let the AI help. Ask it to improve your prompt, or to ask you clarifying questions before answering. It is a surprisingly good prompt coach.
Types of prompts
People mean two different things by "types of prompts." Both are useful to know.
By what you want to create
The medium changes the tool and the wording, though the principles stay the same. Text prompts drive the chat assistants for writing, analysis, and code (ChatGPT, Claude, Gemini). Image prompts turn descriptions into visuals (Midjourney and the image tools built into the major assistants), and reward rich visual detail. Video and audio prompts are a growing category with tools like Google's Veo. Code prompts ask an assistant to write, explain, or fix code, and benefit from stating the language and constraints.
By technique
These are the moves that change how the model approaches your request, and they work across tools:
- Zero-shot. You simply ask, with no examples. Fine for straightforward tasks.
- Few-shot. You include a couple of examples of the input and the output you want. Powerful for matching a specific style or format.
- Role prompting. You assign a persona ("act as a financial analyst") to shape tone and depth.
- Step-by-step. You ask the model to reason through a problem before answering ("think it through step by step"). This improves results on anything with logic or multiple stages.
- Iterative. You refine across a short conversation rather than expecting one perfect prompt to do everything.
Prompt examples you can use today
The fastest way to improve is to start from a strong prompt and adapt it. Here is a library across common tasks. Copy them, swap in your specifics, and refine from there.
Everyday work and writing
Summarize the document below for a busy executive. Give me three key takeaways, any decisions required, and one risk to watch, in under 150 words. [paste document]
Why it works: it sets the audience, the exact output, and a length, so you get a usable brief instead of a wall of text.
You are a careful editor. Tighten the text below for clarity and flow without changing my meaning or voice. Show the edited version, then list the three biggest changes you made and why. [paste text]
Below are my raw meeting notes. Pull out every decision made and every action item, with an owner and a due date where one was mentioned. Flag anything that was left unresolved. [paste notes]
Marketing and content
Turn the blog post below into five LinkedIn posts for a B2B audience. Vary the angle (one data point, one story, one contrarian take, one how-to, one question). Keep each under 80 words and conversational. [paste post]
Write five Google Ads headlines (max 30 characters each) and two descriptions (max 90 characters) for a project management tool aimed at small agencies. Emphasize saving time. Make them distinct, not reworded versions of one idea.
Software development
Review the function below and rewrite it to reduce its time complexity. Explain what you changed and why, and note any tradeoffs. [paste code]
Create a RESTful API endpoint for a user registration system in Node.js with Express. Include input validation, error handling, and brief comments. Follow common security best practices.
Explain what the code below does in plain language, as if onboarding a new developer. Call out anything that looks risky or outdated, and suggest where tests would be most valuable. [paste code]
Data and analysis
Below is a set of customer feedback comments. Identify the top five recurring themes, rank them by how often they appear, give one representative quote for each, and suggest one action per theme. [paste comments]
Learning something new
Explain how a REST API works to someone who is comfortable with spreadsheets but not coding. Use one everyday analogy, keep it under 200 words, then ask me one question to check my understanding.
Customer support
You are a friendly support agent. Draft a reply to the customer message below. Acknowledge their frustration, explain the fix in plain steps, and offer a next step. Keep it warm and under 120 words. Flag anything you are unsure about for me to check. [paste message]
Design and images
Image prompts reward visual detail. Compare these two:
A picture of an office.
A bright modern open-plan office at golden hour, large windows, plants, a few people collaborating at a whiteboard, warm natural light, shot wide-angle, photorealistic.
Why it works: subject, setting, lighting, mood, and style all guide the result. The more visual specifics you give an image tool, the closer it lands to what you pictured.
Common mistakes to avoid
When an answer underwhelms, the cause is usually one of these.
- Under-specifying. "Create a marketing plan" gives the AI nothing to work with. Add the product, audience, channel, and goal.
- Over-complicating. Cramming five questions and heavy jargon into one prompt confuses the model. Break big requests into a sequence of smaller ones.
- Skipping context. Without the audience or purpose, you get content pitched at the wrong level. Always say who it is for.
- Treating it as one-shot. Giving up after the first try is the most common mistake of all. Refining is the method, not a failure.
- Trusting blindly. AI can be confidently wrong. Verify anything that matters rather than pasting it straight through. This is the habit most people skip.
Beyond basic prompting
Once individual prompts feel natural, a few next steps make AI dramatically more useful. They share a theme: the work shifts from clever wording to giving the AI the right context and the right reach.
Context engineering is the modern evolution of prompting. Instead of crafting the perfect sentence each time, you set up the AI with standing instructions, reference material, and examples so it has what it needs every time. Most major tools now support saved instructions and project files for exactly this.
Custom assistants let you package that context into a reusable helper tuned to a specific job, the natural successor to the early "custom GPT" idea. Many also use retrieval-augmented generation to ground answers in your own documents.
Agents are where prompting changes shape entirely. Instead of asking for an answer, you give an agent a goal and guardrails, and it carries out a multi-step task using tools while you supervise. If you want to see how this works in practice, our guide to building agents with Claude goes deeper, and if you are still finding your footing, the beginner's guide to generative AI is a good companion.
Turn prompting skill into team-wide fluency
AI Training for Teams builds the practical, hands-on skills that turn scattered experimentation into real, repeatable productivity across your organization.
Start Your AI TrainingAnd when prompting turns into building software with AI, a structured operating model is what keeps the speed reliable. That is the role of our Generative-Driven Development approach.
Frequently asked questions about AI prompts
What is a prompt in generative AI?
A prompt is the instruction or question you give a generative AI tool to tell it what to produce. It can be as short as a sentence or as detailed as a full brief with a role, context, and format. The clearer and more specific the prompt, the better the output tends to be.
What are the different types of prompts?
One way to group them is by output: text, image, video and audio, and code prompts. Another is by technique: zero-shot (just asking), few-shot (giving examples), role prompting (assigning a persona), step-by-step (asking the model to reason), and iterative (refining over a short conversation). Most good prompting combines several of these.
What makes a good AI prompt?
A good prompt is specific and gives the AI what it needs: a clear task, relevant context, the audience or purpose, and the desired format. Including a short example of what you want often helps. The biggest single improvement most people can make is simply being less vague.
Can you give an example of a good prompt?
Yes. Instead of "write a product email," a strong version is: "You are a B2B marketer. Write a product announcement email to existing customers about our new scheduling feature, keep it under 120 words, warm but professional, and end with a clear call to try it." The role, task, context, and format are all specified.
What is prompt engineering?
Prompt engineering is the practice of designing prompts to get reliable, high-quality results from AI, especially in repeatable or production settings. For most everyday use it is less about special syntax and more about clarity, context, and iteration. At scale, it increasingly blends into context engineering, where you set up the AI's standing instructions and reference material rather than perfecting one-off prompts.
How do I get better at prompting?
Practice on real tasks, start from strong examples and adapt them, and refine rather than abandoning a weak first result. A fast shortcut is to ask the AI to improve your prompt or to ask you clarifying questions before it answers.
Do prompts work the same across different AI tools?
The core principles (be specific, give context, show examples, iterate) transfer across ChatGPT, Claude, Gemini, and others. The details differ: image tools reward visual description, and the newest reasoning models need less step-by-step instruction. Expect to adjust slightly when you switch tools, but the fundamentals carry over.
