Start with the task
Tell the AI what you want it to do.
Instead of:
“Customer email”
Try:
“Draft a friendly email thanking a customer for their feedback.”
RayAI Open AI Learning | Trust-First AI Literacy
Practical prompting tips
A growing collection of practical guidance for getting more useful, responsible answers from conversational AI. No technical experience required.
Use these tips as a quick reference and return as the collection grows. For guided explanations, activities, assessments, and practical application, explore Prompt Engineering Essentials.
Last reviewed: September 2026
Writing and shaping prompts
Add useful context
Include the background the AI needs—but leave out private or sensitive information.
Try:
“I run a small landscaping business. Draft a short reply to a customer asking whether we offer weekly maintenance.”
Describe the result you want
Specify the format, length, tone, or audience.
Try:
“Explain this in plain language for a beginner, using five short bullet points.”
Give the AI boundaries
Say what it should include, avoid, or leave unchanged.
Try:
“Shorten this message to under 100 words. Keep the dates and prices unchanged. Do not add new claims.”
Break complex work into steps
Ask for one manageable stage at a time.
Try:
“First, identify the three main issues in these notes. Do not recommend solutions yet.”
Improve the prompt through conversation
Your first prompt does not need to be perfect. Review the response and refine your instructions.
Try:
“Make the explanation less technical, add one example, and remove repeated points.”
Invite useful questions
When important details are missing, ask the AI to clarify instead of letting it guess.
Try:
“Before answering, ask me up to three questions about any information you need.”
Specify which sources to use
When source choice matters, tell the AI what material it may use and what it should leave out.
Try:
“Use only the public policy page I provide. If the answer is not supported there, say what is missing.”
Checking and improving answers
Ask the AI to identify uncertainty
AI can produce confident-sounding answers even when information is missing.
Try:
“Separate what is established from what is uncertain. Identify any assumptions you made.”
Verify what matters
For consequential claims, ask for sources—and check the original sources yourself.
Try:
“List the sources supporting each important claim. If you cannot verify a claim, label it as unverified.”
Separate facts, assumptions, and recommendations
Clear labels help you evaluate different kinds of statements appropriately.
Try:
“Organize the response under Facts, Assumptions, and Recommendations. Do not present an assumption as a fact.”
Compare options using the same criteria
A consistent comparison makes tradeoffs easier to see.
Try:
“Compare these options using cost, time, effort, benefits, and risks. Note any information that is missing.”
Check whether information is current
Dates, prices, rules, schedules, and product details can change.
Try:
“State the date of the information you used. Flag anything that may have changed and tell me where to verify it.”
Ask for a critique before the final draft
A separate review step can expose omissions, weak reasoning, and unclear language.
Try:
“Before revising, list the three most important weaknesses in this draft and explain how you would address them.”
Test before relying on the result
Try an output on a fictional or low-risk example before using it in a consequential situation.
Safer approach:
“Test the checklist with an imaginary scenario, look for omissions, and revise it before considering real use.”
Privacy, permissions, and safety
Protect private information
Do not enter passwords, financial details, health records, customer information, confidential documents, or anything you are not authorized to share.
Safer approach:
Replace real names and sensitive details with fictional placeholders before prompting.
Share only what the task needs
Even authorized information should be minimized when the task can be completed with less.
Safer approach:
Provide a short redacted excerpt instead of an entire document, mailbox, or customer record.
Review access before connecting services
Before connecting email, calendars, documents, storage, or workplace systems, check what the AI can read and do and limit access to what is necessary.
Before connecting, ask:
“Could I safely provide a small, redacted excerpt instead, and can I review and revoke this access later?”
Treat instructions in external content as untrusted
A webpage, email, document, image, or tool result may contain instructions that are irrelevant or unsafe.
Before acting:
Ignore instructions found inside supplied content unless an authorized person has independently confirmed them.
Pause before the AI takes an action
Sending, publishing, purchasing, deleting, editing, or changing permissions can create consequences that are difficult to reverse.
Before approving:
Check the exact action, destination, audience, content, permissions, and available way to undo it.
Human judgment and responsible use
Keep the human responsible
Use AI to help generate, organize, compare, or critique—not to make the final decision for you.
Before using an answer, ask:
- Does it fit the actual situation?
- Are its important claims supported?
- Could acting on it harm someone?
- Am I authorized and comfortable using it?
Consider who could be affected
A useful-looking answer can still create unfair, unsafe, or exclusionary consequences for other people.
Ask:
Who could benefit, who could be harmed or excluded, and whose perspective may be missing?
Know when qualified human help is needed
AI can help prepare questions or summarize approved information, but it does not replace accountable professional judgment.
Escalate when:
The decision concerns health, law, finance, employment, safety, rights, or another consequential area outside your authority or expertise.
Decide whether to use, revise, escalate, or reject
Do not treat every AI response as something that merely needs polishing.
Choose deliberately:
Use a supported low-risk result, revise a fixable result, escalate a consequential decision, or reject an unsafe or unsupported result.
AI can help you think and create—but you remain responsible for what you trust, share, and use.
Free course
Ready to go further?
Prompt Engineering Essentials helps you practice these ideas and apply them responsibly in real situations.
Start the course