The plain-English answer, with real examples. Understand prompts in 5 minutes — and stop wondering why ChatGPT keeps giving you mediocre answers.
If you've ever asked ChatGPT a question and gotten a bland, surface-level reply, the problem usually isn't the AI — it's the prompt. Most people learn to use AI by guessing, and most guesses produce average answers. The good news is that writing a good prompt is easier than people make it sound. You don't need a course. You just need a friendly walkthrough, which is what the next 1,800 words give you.
An AI prompt is the message you type to an AI chatbot like ChatGPT, Claude, or Gemini. That's the whole definition. If you've ever asked an AI anything — "what's a good recipe for pasta," "rewrite this email," "explain machine learning" — you've already written a prompt. There is no special syntax to learn and no programming involved.
What separates a great prompt from a mediocre one is how clearly it tells the AI what you want. Think of the AI as an extremely well-read assistant who has never met you before. They can do almost anything you ask, but only if you tell them exactly what "good" looks like. Skip that part, and you'll get something generic. Spell it out, and you'll get something useful.
The fastest way to feel why prompts matter is to see the same task asked two ways. Pretend you're trying to write a quick LinkedIn post about a new role you just started.
Vague prompt: "Write me a LinkedIn post about my new job."
What you'll get: a chirpy four-paragraph post with a generic "thrilled to announce" opener, three corporate-sounding bullet points, and the obligatory "open to connecting" line at the bottom. It sounds like every LinkedIn post ever written. You won't post it.
Specific prompt: "Write a 6-sentence LinkedIn post announcing that I just joined Acme Co. as a senior product designer. Tone: warm but understated. No 'thrilled to announce.' Mention that I'll be focused on the onboarding experience for new users. End with one short, low-pressure sentence inviting product designers in fintech to message me. Avoid clichés."
What you'll get: a six-sentence post with a calm opener, your real beat (onboarding for new users), and a friendly invitation that doesn't sound like a marketing email. Same AI. Same model. Wildly different output. The only thing that changed was the prompt.
This is the entire point of learning to prompt: you stop blaming the AI and start writing better briefs.
Every strong prompt has up to five ingredients. You don't need all five every time, but knowing the full set means you can always diagnose why an answer came back weak.
Tell the AI who to pretend to be. "Act as a friendly career coach." "You're a senior copywriter who specializes in tech newsletters." "You are a no-nonsense fitness trainer." Roles are powerful because the model has read millions of examples of how those people write — assigning a role unlocks a more specific voice and better assumptions.
Tell the AI about your situation. Who's the audience? What industry? What constraint are you under? What have you already tried? Two or three sentences is usually enough. The more relevant detail you give, the less the AI has to guess. (Don't pour in everything you know — only what changes the answer.)
State what you want, plainly. "Write," "summarize," "compare," "list," "rewrite," "outline," "brainstorm." Use a single clear verb when you can. If your task has steps, number them. AI does very well with numbered instructions.
Tell the AI what shape the answer should take. A bullet list of 7? A 200-word paragraph? A table with three columns? A numbered plan? Pinning down the format is the single fastest way to stop the AI from rambling. It also makes the output easier to use afterward.
Tell the AI what not to do. "Avoid jargon." "Under 100 words." "Don't use the word 'leverage.'" "No emojis." "Sound human, not corporate." Constraints are the secret weapon of good prompting — they're how you stop the AI from defaulting to its blandest, safest voice.
Almost every weak prompt is making one of these five errors. If your AI answer feels off, walk down this list:
To make all of this concrete, here are three full prompts you can copy and try. Each one shows how the five parts come together.
Prompt: "Act as a calm executive assistant. Summarize the email below in 3 bullet points: what's being asked, who needs to act, and the deadline. End with one suggested reply I could send back in 2 sentences. Keep tone polite but efficient. EMAIL: [paste email here]."
Why it works: Role (executive assistant), context (an inbound email), task (summarize and suggest a reply), format (3 bullets + 2 sentences), constraints (polite but efficient). All five parts in 60 words.
Prompt: "You are a creative strategist who balances safe and bold ideas. Generate 8 names for a new dog-walking subscription service for busy urban professionals. Tone: warm, slightly playful, never cutesy. For each name, give a one-sentence reasoning. Rank them from 'safest' to 'boldest.'"
Why it works: The role steers tone, the audience pins down style, the format ("8 names with one-sentence reasoning, ranked") prevents a wall of text, and "never cutesy" is the constraint that stops AI from defaulting to puns.
Prompt: "Edit the paragraph below for clarity and flow. Keep my voice — don't make it more formal or more casual. Show me the edited version, then list the 3 most useful changes you made and why. PARAGRAPH: [paste paragraph here]."
Why it works: The "keep my voice" constraint stops the AI from rewriting your style entirely, and asking it to explain its 3 best changes turns one prompt into a quiet writing lesson.
You now know more about AI prompts than 90% of the people using ChatGPT. The next step is to actually try a few. Two natural follow-ups from here:
And if you'd rather just be walked through your first 10 minutes, head back to the 10-minute starting kit — it covers picking your AI tool and includes 5 starter prompts you can run right now.
Join readers learning to use AI better — one practical prompt and a short setup explaining why it works.
Unsubscribe anytime. We never sell your email.