Blog

One sentence is enough to start an AI conversation. Four clear inputs can make that sentence far more useful. When I write AI prompts, I focus on context, role, task, and output instead of trying to sound technical.

A weak prompt often creates extra work. I ask for an email, get a generic draft, then spend three more turns fixing the tone, audience, and length. That does not mean the AI failed. It means I left out information the tool needed. The cost is small each time, but it adds up when I use AI all day. When I write AI prompts with enough context, I reduce that back-and-forth. It also makes the result easier to review because I know what I asked for. I want a prompt that a smart coworker could follow without guessing.

People make prompting harder than it needs to be because the term sounds technical. It is just an instruction. OpenAI’s own guidance says to be clear, specific, and detailed about the task, context, tone, and style. I write AI prompts with a simple four-part model called CRTO: Context, Role, Task, Output. I do not need every part for every request. I use more detail when the result matters more.

Write AI prompts with one clear task

A prompt should make the main job obvious. I start with a verb that tells the model what to do. One task per prompt is easier to control than five tasks mixed together.

Start with a direct verb such as write, compare, summarize, or plan.

Name the exact item you want produced.

Remove side requests that can wait for a second prompt.

Add a clear finish point.

Ask one follow-up after you review the first result.

What good looks like: The model can restate the task in one short sentence without guessing.

Add the context a coworker would need

AI does not know the details in my head. I give it the facts that shape the answer. Context can include the audience, goal, offer, background, and limits.

Name the audience for the output.

Explain why the task matters now.

Add facts the model should use.

Include examples when style matters.

Leave out private data that is not needed.

What good looks like: The draft reflects the real business situation, not a generic version of it.

Assign a role when expertise helps

A role tells the model which lens to use. I use roles when I want a certain kind of thinking, such as sales, operations, or editing. I avoid roles that imply a licensed expert should make a final high-risk decision.

Pick a role tied to the task.

Keep the role simple and specific.

Ask for reasoning that fits that role.

Keep legal, medical, and financial decisions with qualified people.

Change the role if the first lens is not useful.

What good looks like: The answer uses the right lens without pretending to replace professional judgment.

Describe the output before AI starts

The output section is where I set shape and limits. It saves editing because the model knows what finished should look like. This can include length, tone, structure, or fields.

State the format you want back.

Set a useful length limit.

Name the tone in plain words.

Ask for bullets, a table, or steps when helpful.

Define anything the answer must include or avoid.

What good looks like: The first answer arrives in a form I can use, edit, or paste with little cleanup.

Refine the answer with one follow-up

Good prompting is a conversation. I do not rewrite the whole prompt when one part is off. I tell the model what to keep and what to change.

Point to the part that works.

Name the part that misses the goal.

Give one new constraint or example.

Ask for a revised version only.

Save the final prompt if the task will repeat.

What good looks like: The second version is better because the feedback is specific, not because the prompt became much longer.

Run this CRTO prompt for a customer follow-up

Here is the same structure in a business task you can use today.

Context: A new customer bought [product or service] on [date]. I want to follow up after delivery. The customer is [type of customer], and the main value they wanted was [goal].
Role: Act as a helpful customer success writer.
Task: Write a friendly follow-up email that checks on the result, offers help, and makes it easy to reply.
Output: Keep it under 120 words. Use plain English, short paragraphs, and a warm professional tone. Give me one subject line. Do not sound pushy.

What good looks like: The prompt is clear enough that the first draft only needs facts and voice edits.

Avoid these common mistakes

Asking for several unrelated tasks in one prompt and getting a mixed result.

Leaving out the audience, goal, or output format, then fixing them later.

Treating a confident AI answer as verified fact without checking the source.

What good looks like: Prompts stay short enough to use and detailed enough to guide the result.

Answer these common questions

How do I write a good AI prompt?

Write a good AI prompt by stating the context, role, task, and output you want. Start with the job, then add only the facts that shape the answer. Be clear about the audience and format. Review the first result, then use a short follow-up to correct what missed the goal.

What should an AI prompt include?

An AI prompt should include the task and enough context to complete it well. Add a role when a certain lens helps. State the output format when you care about length, tone, structure, or fields. You do not need every detail, only the details that change the answer.

Do AI prompts need to be long?

No. AI prompts need to be clear, not long. A short prompt can work well for a simple task. More complex work needs more context, rules, or examples. I add detail when the cost of a weak answer is higher, not because long prompts are always better.

How do I get better answers from AI?

Give the model a clearer task, better context, and a more specific output. If the first answer is weak, tell it what worked and what needs to change. You can also break a large request into smaller steps. Better feedback often improves the result more than adding extra words up front.

What good looks like: A beginner can build a useful prompt from four parts and improve it with one focused follow-up.

I write AI prompts like I brief a capable coworker. I give the right context, set the job, and define what useful looks like. Take one task you already do today and rewrite the request with CRTO before you run it again.

It is a long established fact that a reader will be distracted by the readable content of a looking at its layout.It is a long established fact that a reader will be distracted by the readable content of a looking at its layout.

It is a long established fact that a reader will be distracted by the readable content of a looking at its layout.It is a long established fact that a reader will be distracted by the readable content of a looking at its layout.

It is a long established fact that a reader will be distracted by the readable content of a looking at its layout.It is a long established fact that a reader will be distracted by the readable content of a looking at its layout.

It is a long established fact that a reader will be distracted by the readable content of a looking at its layout.It is a long established fact that a reader will be distracted by the readable content of a looking at its layout.

It is a long established fact that a reader will be distracted by the readable content of a looking at its layout.It is a long established fact that a reader will be distracted by the readable content of a looking at its layout.