What Is a Prompt?

Start with an example

Two people ask AI for help with the same thing.

Person A types: “Write a post.”

Person B types: “Write a 3-sentence Instagram post for my Chicago bakery announcing pumpkin bread this Saturday. Upbeat tone. End by inviting people to stop in before noon, because it sells out.”

Person A gets something generic about “exciting news” and “delicious treats.” Person B gets a post they can use almost as-is. Same tool. Same moment. Different prompt.

The difference isn’t luck or a secret trick. Person B simply gave the AI what it needed to know.

What a prompt is

A prompt is everything you give the AI to work with:

• Your instruction: “Write…”, “Summarize…”, “Rewrite…”, “Compare…”
• The details it needs: who, what, when, where, and why
• Examples of what good looks like
• Files or text you paste in: notes, a document, an old email, a photo

Remember from the last lesson: the model predicts its answer from what you give it. It knows general information from training, but it only knows your situation if you put it in the prompt. If a detail matters and it’s missing, the AI will guess, and guesses tend to be generic.

The four ingredients of a strong prompt

You’ll learn the full R.I.C.K. Model in the Prompting course. For now, most strong prompts include these four ingredients:

Ingredient	Question it answers	Example
Task	What do you want done?	“Write a thank-you email…”
Audience	Who is it for?	“…to volunteers who helped at our food drive…”
Key details	What facts must it include?	“…we collected 1,200 pounds of food, and the next drive is March 14…”
Format	What should it look like?	“…under 150 words, warm tone, with a P.S. about the next date.”

Put together: “Write a thank-you email to volunteers who helped at our food drive. We collected 1,200 pounds of food, and the next drive is March 14. Keep it under 150 words, warm, with a P.S. about the next date.”

That took about 30 seconds to write and will save several minutes of editing.

Weak vs. strong prompts

Weak prompt	What’s missing	Stronger version
“Summarize this.”	Audience, length, focus	“Summarize these meeting notes in 5 bullets for my team. Put action items first, with who owns each one.”
“Write a bio.”	Who, for where, how long	“Write a 60-word bio for a licensed counselor’s website. Warm, professional, first person. Mention 10 years of experience with teens.”
“Help with my resume.”	Goal and target	“Rewrite these 3 resume bullets to show results with numbers. Target role: operations manager at a hospital.”
“Explain inflation.”	Audience and level	“Explain inflation to a high school junior in under 200 words, using a grocery store example.”

Notice a pattern: the strong versions are longer, but not complicated. They just answer the questions a smart assistant would ask before starting.

Prompting is a conversation

You don’t have to get it perfect on the first try. After the first answer, keep going:

• “Shorter. Cut it to 3 sentences.”
• “More casual, like you’re talking to a neighbor.”
• “Turn this into a checklist.”
• “Give me three different versions with different tones.”
• “What questions should I answer to make this better?”

That last one is powerful. When you’re not sure what to include, ask the AI what it needs from you, then answer its questions.

Good follow-ups are specific. “Make it better” gives the model nothing to aim at. “Make the opening line more urgent and remove the second paragraph” tells it exactly what to change.

A worked example: building a prompt step by step

DeShawn coaches a youth basketball team and needs to email parents about a schedule change.

Version 1: “Write an email about practice.” The result is vague: something about “upcoming practice updates.”

Version 2 (adds task and audience): “Write an email to parents of my 12U basketball team about a practice change.” Better, but it invents a date and reason.

Version 3 (adds key details): “…Practice moves from Tuesday to Wednesday next week, same time, 6 to 7:30 pm, because the gym is being used for a school event.” Now the facts are right.

Version 4 (adds format): “…Keep it under 100 words, friendly, and end by asking parents to reply if their child can’t make it.” Ready to send after a quick read.

Follow-up: “Add a one-line reminder to bring water bottles.” Done.

Each step took a few seconds. Together, they turned a useless draft into a usable one.

Keep private things private

Before you paste anything into a prompt, ask yourself: “Would I be OK if someone outside my organization saw this?”

Don’t paste these into a tool unless your organization has approved it for that kind of information:

• Passwords, account numbers, or Social Security numbers
• Customer or patient health details
• Anything covered by a confidentiality agreement

Often you can simply remove the private part. Instead of a client’s name, write “the client.” Instead of a real account number, leave it out. The AI rarely needs those details to do its job. Lesson 5 covers privacy in more depth.

Think deeper: your prompt is never alone

When you type a message, the AI doesn’t see just your words. It sees a stack:

1. A hidden system prompt written by the company, which sets the assistant’s rules and tone
2. Any custom instructions you’ve saved
3. The whole conversation so far
4. Your newest message

All of it is fed in together as one long document. That explains three things.

Follow-ups work because your earlier corrections are still in the stack. But a long, messy chat can also carry old mistakes forward, which is why a fresh chat sometimes works better.

AI can be tricked by pasted text. To the model, everything is just text. If you paste an email that says “ignore your instructions,” a poorly built tool might obey it. This is called prompt injection, and it matters for AI tools that read emails, documents, or websites for you. Marking clearly where pasted material starts and ends helps.

A few saved sentences go a long way. Custom instructions about your role, audience, and preferred format get added to every chat automatically. Write them once and they improve hundreds of answers.

Questions to sit with:

• If a company’s hidden system prompt shapes every answer you get, how much of the AI’s “opinion” is really the company’s?
• What three sentences about yourself would make every AI answer more useful to you?

Practice: the four-ingredient rewrite

1. Start with this weak prompt: “Write a post about our event.”
2. Choose a real or made-up event: a car wash fundraiser, a store opening, a school play.
3. Add each ingredient: the task, the audience, two or three key details, and the format (platform, length, tone).
4. Run both the weak and strong versions in the same AI tool.
5. Compare them. Which one needs less editing?
6. Follow up once with a specific change and see how the AI responds.

Next, you’ll look closely at the biggest risk with AI writing: answers that sound right but aren’t, and how to catch them before they cause trouble.