Embedded Files
You don't need a perfect prompt. You need a good conversation.
You don't need a perfect prompt. You need a good conversation.
Think of AI like a capable new coworker. If you walk up and say, "Make me a report," you shouldn't expect exactly what you had in mind. They need context, a clear assignment, an idea of what success looks like, and some boundaries. They may also need to ask a few questions. AI works the same way.
Give it a good brief.
Give it a good brief.
A good brief usually includes five things:
Give context
Give context
What does the AI need to know about you and the situation?
Define the job
Define the job
What exactly do you want it to do?
Define success
Define success
What would a useful answer look like?
Add boundaries
Add boundaries
What should it keep in mind, or avoid?
Have a conversation
Have a conversation
Respond, correct, and keep going.
You won't need all five for every question. But when the task is important or complicated, or the AI isn't giving you what you need, these five can make a big difference.
Give context
Give context
AI only knows what you tell it, what you've shared with it, and what's available in the tool you're using. Give it enough background to understand the situation: your role and industry, who the audience is, how the process works today, why it matters, and any systems or terminology involved.
Instead of:
“Help me improve our production meeting.”
Try:
“I work for a custom plastic injection molding company. Our production supervisors meet every morning to review the previous day's performance and today's priorities. The meeting frequently runs too long and we sometimes spend more time reviewing information than identifying actions. Help me think through ways to make the meeting shorter and more action-oriented.”
The AI now has something to work with.
Define the job
Define the job
Be clear about the assignment. Do you want it to explain, analyze, compare, summarize, brainstorm, troubleshoot, draft, rewrite, organize, challenge an idea, recommend options, or create a step-by-step plan?
Instead of:
“Look at this report.”
Try:
“Review this report and identify the three trends that deserve the most management attention. Explain why each one matters and what additional information you would want before recommending action.”
The second prompt gives the AI a job.
Define success
Define success
Don't make the AI guess. Tell it who the final audience is, how long and detailed the answer should be, the tone and reading level, and the format you want, such as a table, checklist, email, report, outline, or step-by-step instructions.
Instead of:
“Explain this SQL query.”
Try:
“Explain this SQL query to someone who understands our ERP data but is not an experienced SQL developer. Walk through each section in plain language and tell me what records the query is ultimately trying to return.”
Same question. Much more useful answer.
Add boundaries
Add boundaries
Constraints are instructions too. Sometimes the easiest way to improve a response is to tell the AI what you don't want. For example:
"Keep this under 300 words."
"Don't use technical terminology unless you explain it."
"Don't make assumptions about information I haven't provided."
"Don't recommend new software until we've explored whether our existing tools can solve the problem."
"If you aren't confident about something, tell me rather than guessing."
"Separate facts from assumptions."
Boundaries help keep the AI focused on the problem you're actually trying to solve.
Have a conversation
Have a conversation
This may be the most important one. You don't have to put everything into one giant prompt. Start the conversation, then respond:
"This part is right, but this assumption is wrong."
"Let's work through this one step at a time."
"Before we continue, ask me what you need to know."
"I don't like any of those options. Let's try a different direction."
Needing a follow-up isn't failing at prompting.
It's having a conversation.
Copy-and-paste prompt: Give it a good brief
Copy-and-paste prompt: Give it a good brief
Need somewhere to begin? Copy, paste, and customize this:
Need somewhere to begin? Copy, paste, and customize this:
CONTEXT
“I work in [ROLE / INDUSTRY].
Here is the situation I'm working with: [RELEVANT CONTEXT]”
THE JOB
“I need your help to: [DESCRIBE WHAT YOU WANT THE AI TO DO]”
SUCCESS
“A useful result would: [DESCRIBE THE DESIRED OUTCOME]
Please provide the answer as: [TABLE / CHECKLIST / EMAIL / ANALYSIS / STEPS / REPORT / OUTLINE / ETC.]”
BOUNDARIES
“Please keep these requirements in mind: [IMPORTANT REQUIREMENTS, LIMITATIONS OR THINGS TO AVOID]”
CONVERSATION
“Before completing the task, identify any important information that is missing.
If you need more information, ask me questions one at a time before proceeding.
When you provide your answer, tell me about any important assumptions you've made, risks I should consider, or information I should verify.”
You don't have to use the whole template.
You don't have to use the whole template.
Sometimes "Can you explain what this error means?" is all you need.
Sometimes "Can you explain what this error means?" is all you need.
Add context when you want better results, boundaries when the AI wanders off track, and success criteria when the answer isn't coming back in a useful form.
Add context when you want better results, boundaries when the AI wanders off track, and success criteria when the answer isn't coming back in a useful form.
Show it what "good" looks like
Show it what "good" looks like
Sometimes an example communicates more than a long explanation. Depending on your company's policies and the tool you're using, you might share a sample report, email, work instruction, checklist, procedure, or meeting summary. Then try:
“This is an example of the structure and level of detail I'm looking for. Analyze how it is organized. Then help me create a new version for [NEW PURPOSE]. Do not copy information from the example that isn't relevant to the new task.”
Remember to remove or anonymize anything that shouldn't be shared.
Separate thinking from doing
Separate thinking from doing
You don't always need the answer right away. Sometimes it's more useful to work through the problem together. Instead of "Fix our inventory process," try working in stages:
First: "Don't recommend a solution yet. Help me identify the possible causes of this problem."
Then: "Ask me questions to help determine which causes are most likely."
Then: "Give me three possible approaches."
Then: "Compare the effort, risk, cost, and likely impact of each approach."
Finally: "Recommend which option we should test first and explain why."
This turns AI from an answer machine into a thinking partner
Ask AI to challenge you
Ask AI to challenge you
AI doesn't only have to help you develop an idea. It can help you test one. Try:
"What am I overlooking?"
"What assumptions am I making?"
"What could cause this plan to fail?"
"Argue against this recommendation."
"What unintended consequences should I consider?"
"What question haven't I asked that I probably should?"
Sometimes the most useful AI response isn't an answer.
It's a question you hadn't thought to ask.
💡 Pro Tip: Ask AI to Write the Next Prompt
💡 Pro Tip: Ask AI to Write the Next Prompt
Once you've worked through a problem and finally reached the result you wanted, don't lose what you learned.
“We've worked through this successfully. Based on our conversation, create a reusable prompt I can save and use the next time I need to complete this task. Include the important context, instructions, boundaries, and questions that helped us reach the final result.”
AI doesn't need you to sound like a computer.
AI doesn't need you to sound like a computer.
Talk to it naturally. Explain the problem, give it context, correct it when it's wrong, ask why, and challenge its assumptions.
Give it a good brief. Then have a good conversation.
Page updated
Google Sites
Report abuse