What Is a Prompt? A Plain-English Guide (With 25 Examples)

A prompt is the text you give an AI to tell it what you want — a question, an instruction, or a description of a task. It's the single most important variable in whether you get a brilliant answer or a useless one. Same model, same question, different prompt: the gap between the two outputs is enormous, and it's almost entirely under your control.

Most people type a prompt the way they'd type a Google search — a few keywords, vague intent, fingers crossed. That works for search because Google ranks pages someone else already wrote. It fails for AI, because the model writes the answer fresh every time, and it can only aim as precisely as your words let it. Learn to write a real prompt and you stop gambling on results.

This guide defines a prompt in plain English, breaks down what separates a strong one from a weak one, walks through the five main types with tiny examples, and gives you 25 short before/after rewrites you can lift today.

What a prompt actually is

Strip away the jargon and a prompt is just the input that conditions the output. The model reads everything you wrote, predicts what a great response would look like given those exact words, and produces it. It has no idea what's in your head — only what's on the screen. If your prompt is ambiguous, the model fills the gaps with the most generic, average-case guess. That's why vague prompts produce bland answers: you asked an average question, so you got an average answer.

A prompt can be one line ("Summarize this email") or a structured brief with a role, context, constraints, and an example. Both are prompts. The difference is how much of the model's guessing you've replaced with your specification.

Weak: Write about marketing.

Strong: Write a 150-word LinkedIn post for B2B SaaS founders explaining why most cold emails fail, with one concrete fix. Confident, no hashtags, no emoji.

The first prompt could produce a textbook chapter, a tweet, or a poem — the model picks. The second leaves almost nothing to chance: format, length, audience, angle, and tone are all locked in. You're not hoping for a good answer anymore; you're describing one.

The anatomy of a good prompt

A strong prompt usually answers five questions before the model has to guess at any of them. You don't need all five every time, but the more important the task, the more of them you should include.

• Who the AI should act as (the role): "You are a senior copy editor."
• What you want done (the task): "Tighten this paragraph."
• Context it needs: who it's for, what's already been tried, the source material.
• Format of the output: length, structure, tone, what to leave out.
• Examples of what good looks like (optional but powerful).

Here's the same request with each layer added:

Weak: Fix my paragraph.

Strong: You are a senior copy editor. Tighten the paragraph below for a landing page aimed at busy founders. Keep it under 60 words, cut every hedge word, and end on a concrete benefit. Return only the edited version. [paragraph]

Notice the strong version names a role, states the task, gives context (landing page, busy founders), constrains the format (under 60 words, end on a benefit), and even controls the output ("return only the edited version" stops the model from padding its reply with commentary). That structure isn't an accident — it's a repeatable pattern. We break it down fully in the RCTCO prompt structure guide, and you can grab fill-in-the-blank versions in our prompt template.

A useful gut-check: read your prompt back and ask, "Could a smart freelancer do this task well with only these instructions?" If they'd have to email you three clarifying questions first, your prompt has three holes in it.

The five main types of prompt

Almost every prompt you'll ever write is one of these five shapes, or a combination. Knowing the names helps you reach for the right tool.

1. Instruction prompts

You tell the model to do something. This is the workhorse.

Rewrite this sentence in plain English for a 12-year-old: "Our solution leverages synergistic methodologies."

2. Question prompts

You ask, the model answers. Best for facts, explanations, and reasoning.

What are the trade-offs between renting and buying a home in your 30s? Give me three points for each.

3. Role prompts

You assign the model an identity so it answers from an expert's vantage point. This shifts its vocabulary, depth, and priorities before it even starts.

You are a pediatric nurse. A parent asks how to bring down a toddler's fever safely at home. Answer reassuringly in plain language.

Role prompting is one of the highest-leverage moves a beginner can learn — it's the difference between a generic answer and an expert one. See the full role-prompting guide for how far you can push it.

4. Few-shot prompts

You show the model two or three examples of the input-output pattern you want, then give it a new input. It copies the pattern.

Turn product features into benefits. Feature: 256-bit encryption → Benefit: Your data stays private, even if your laptop is stolen. Feature: Offline mode → Benefit: Keep working on the train, no signal needed. Feature: One-click export → Benefit:

The model finishes the last line in the same style. This is few-shot prompting, and it's the fastest way to enforce a format without describing it in words.

5. Chain-of-thought prompts

You ask the model to reason step by step before answering. This dramatically improves accuracy on math, logic, and multi-step problems.

A shirt costs €40 after a 20% discount. What was the original price? Think through it step by step, then give the final number.

By forcing the working out, you stop the model from blurting a confident wrong answer. More on when and why in chain-of-thought prompting.

25 before / after example prompts

Theory sticks faster when you see it. Below are 25 quick rewrites, grouped by what you're trying to do. In every pair, the after version simply removes ambiguity — it doesn't get longer for the sake of it.

Writing and editing

1. Weak: Write an email. → Strong: Write a 4-sentence email declining a meeting invite politely, suggesting async notes instead.
2. Weak: Make this shorter. → Strong: Cut this to 100 words, keep the first and last sentences, remove every adjective that isn't load-bearing.
3. Weak: Write a bio. → Strong: Write a 50-word third-person bio for a freelance illustrator, warm but professional, for an About page.
4. Weak: Improve my writing. → Strong: Rewrite this paragraph at a 9th-grade reading level, active voice, no sentence over 20 words.
5. Weak: Give me a headline. → Strong: Give me 5 headline options for a blog post on remote-work burnout, each under 60 characters, curiosity-driven, no clickbait.

These work because they swap an open-ended verb for a measurable target — word count, reading level, character limit, quantity.

Learning and explaining

6. Weak: Explain quantum computing. → Strong: Explain quantum computing to me like I'm a smart 15-year-old, using one everyday analogy, in under 200 words.
7. Weak: Teach me Spanish. → Strong: Give me 10 Spanish phrases for ordering food in a restaurant, with phonetic spelling and the literal English translation.
8. Weak: What is inflation? → Strong: Explain inflation in 3 short paragraphs: what it is, why it happens, and one thing it means for my savings.
9. Weak: Help me study. → Strong: Create 8 flashcard-style Q&A pairs covering the causes of World War I, ordered easiest to hardest.
10. Weak: Summarize this. → Strong: Summarize this article in 5 bullet points a busy manager could read in 30 seconds, then add one "so what" takeaway.

Each one tells the model the audience and the exact shape of the answer, so it stops defaulting to a wall of text.

Work and productivity

11. Weak: Write a job description. → Strong: Write a job description for a remote junior data analyst: 4 responsibilities, 4 must-have skills, friendly tone, no corporate clichés.
12. Weak: Plan my week. → Strong: Turn this to-do list into a Mon-Fri schedule, max 3 deep-work blocks per day, with buffer time between meetings. [list]
13. Weak: Make meeting notes. → Strong: Turn this transcript into notes with three sections: Decisions, Action items (with owners), and Open questions. [transcript]
14. Weak: Help me negotiate. → Strong: Draft 3 polite scripts to ask for a 10% raise, each framed around a different argument: tenure, results, market rate.
15. Weak: Write a proposal. → Strong: Write a one-page project proposal for redesigning a small bakery's website, with goals, timeline, and a price range.

The pattern here is structure: naming the sections you want forces a usable, scannable answer instead of prose you have to re-organize yourself.

Ideas and brainstorming

16. Weak: Give me business ideas. → Strong: Give me 7 low-startup-cost business ideas for someone with graphic-design skills and 5 spare hours a week.
17. Weak: Name my product. → Strong: Suggest 10 names for a meal-prep app aimed at gym-goers — short, easy to spell, .com likely available. Avoid "fit" and "meal."
18. Weak: Help me with content. → Strong: Give me 12 short-form video hooks for a personal-finance creator targeting Gen Z, each under 12 words.
19. Weak: Brainstorm gifts. → Strong: Suggest 8 gift ideas under €40 for a friend who loves hiking and hates clutter, ranked by how memorable they'd be.
20. Weak: Plan an event. → Strong: Plan a 2-hour birthday picnic for 10 adults: a loose timeline, a no-cook food list, and one backup plan for rain.

Constraints (budget, count, audience, what to avoid) make brainstorms sharper, not narrower — the model spends its effort inside the box you drew instead of wandering.

Code and analysis

21. Weak: Write some code. → Strong: Write a Python function that takes a list of emails and returns only the valid ones. Add comments and one example call.
22. Weak: Fix this bug. → Strong: This function returns None instead of a list. Explain the likely cause in one sentence, then show the corrected code. [code]
23. Weak: Explain this code. → Strong: Explain what this SQL query does in plain English, line by line, and flag anything that could be slow on a big table. [query]
24. Weak: Analyze this data. → Strong: Here's 12 months of sales numbers. Identify the top trend, one anomaly, and one question I should investigate next. [data]
25. Weak: Help me decide. → Strong: I'm choosing between two job offers below. Compare them on pay, growth, and risk in a table, then give your pick with one reason. [details]

Developers will recognize the same discipline in our prompts-for-developers pack: state the input, the expected output, and what to check for.

The 5 most common beginner mistakes

If your prompts keep underperforming, you're probably making one of these.

1. Being too vague. "Write something good about our app" gives the model nothing to aim at. Name the format, length, and audience.
2. Asking for everything at once. A single prompt that wants research, an outline, a draft, and SEO will do all four badly. Break it into steps.
3. Skipping context. The model doesn't know your industry, your customer, or what you already tried. If it matters, paste it in.
4. Not saying what you don't want. "No jargon," "no emoji," "don't summarize, just edit" — negative constraints save you a second round.
5. Accepting the first draft. The first answer is a starting point. Reply with "tighter," "more specific," "give me 3 versions" and you'll watch quality climb.

Every one of these traces back to the same root: a prompt that under-specifies the task. We go deeper on the why behind weak outputs in why most AI results are mediocre.

A simple 3-step method to improve any prompt

You don't need to memorize frameworks to write better prompts. Run any prompt through these three passes.

Step 1 — Add a role and an audience. Decide who the AI should be and who the output is for. "You are a [role]. Write for [audience]." This alone lifts most prompts a full grade.

Step 2 — Specify the output. State the format, length, and tone, and name anything to avoid. "In 5 bullets, plain language, no jargon." Now the model can't default to a generic shape.

Step 3 — Show or constrain. Either paste one example of what good looks like, or add the constraints that matter (a number, a deadline, a budget, a "must include"). This is where a decent prompt becomes a reliable one.

Run a weak prompt through those three steps and watch it transform:

Weak: Help me write a cover letter.

Strong (after 3 steps): You are a hiring manager who reviews 50 cover letters a week. Write a 200-word cover letter for a marketing-coordinator role at a small nonprofit, for a candidate switching from retail. Warm, specific, no clichés like "team player." Open with a hook, not "I am writing to apply."

If you'd rather not run the steps by hand, that's exactly what our free brain does — type a rough request and it returns an expert-grade version you can use as-is. To go further, Studio lets you build prompts visually as connected blocks (Goal, Role, Context, Instruction) so you can see and reuse the structure instead of cramming it into one paragraph.

Frequently asked questions

What is a prompt in AI, in simple terms?

A prompt is the text instruction or question you give an AI to tell it what you want. It's the input — the model reads your words and generates a response based entirely on them. A clear, specific prompt produces a clear, specific answer; a vague one produces a vague, average answer.

What makes a good prompt?

A good prompt gives the AI enough to stop guessing: who it should act as, what task to do, any context it needs, and the format you want back. The most reliable upgrade is to add a role, name your audience, and specify the output length and tone. You can see dozens of strong, ready-to-use examples in our prompt examples library.

What are the different types of prompts?

The five main types are instruction prompts (do this), question prompts (answer this), role prompts (act as an expert), few-shot prompts (copy these examples), and chain-of-thought prompts (reason step by step). Most real-world prompts combine two or three of these.

How long should a prompt be?

As long as it needs to be to remove ambiguity — and no longer. A one-line prompt is fine for a simple task; a complex task may need a short paragraph with a role, context, and constraints. Length isn't the goal; specificity is. Don't pad a prompt with words that don't change the output.

A prompt is the steering wheel for every AI tool you'll ever use, and getting good at it is the highest-leverage skill in the AI era. The fastest way to learn is by example: try a rough request in our free brain and watch it rewrite itself into something expert-grade, browse the Library of 2,750+ prompts you can copy or open in Studio, or read how to prompt AI correctly to put the fundamentals together. Start with one weak prompt, run it through the three steps, and you'll never type a lazy one-liner again.