01 · Know the kitchen
Understand the tool you’re using.
An AI model learns patterns from data. An AI tool gives you a way to use that model. A large language model, or LLM, can generate language, but a fluent sentence is not proof that it is true.
Before starting, ask: Can this tool search current sources? Can it see the document I’m asking about? Can it take actions, or only suggest them? These details shape what you can reasonably expect from its answer.
A chef can suggest a dish from experience. To tell you what is in today’s pantry, they need today’s inventory. AI also needs access to the information your question depends on.
02 · Place a clear order
Give the AI a goal and enough context.
Your prompt is the request you send. Include the task, the audience, the information it should use, and the format you want back. context engineering is the broader work of making the right information available when the system needs it.
“Write something about our event.”
“Use the event details below to write a 100-word invitation for high school students. Use a friendly tone. Include the date and location exactly as provided. If a detail is missing, ask me instead of inventing it.”
Read the first result, then refine your request. Tell the tool what needs to change. More words alone won’t fix a vague goal; relevant details are what help.
03 · Taste before serving
Check the answer against real evidence.
A hallucination is a confident response containing incorrect or invented information. Check names, dates, calculations, quotes, and links before relying on them. Open a cited source and confirm that it supports the specific claim.
ai grounding connects an answer to relevant sources. It can help, but you still need to check that the source is reliable and the answer represents it correctly. For decisions with serious consequences, use an appropriate qualified person and authoritative sources.
Ask the AI to separate facts from assumptions and point out missing information. Then verify the important facts yourself. The AI reviewing its own answer is useful feedback, not independent proof.
04 · Set the house rules
Choose what to share and what to approve.
Use made-up or non-sensitive information while learning. Before entering customer records, private documents, or passwords, check your organization’s rules and the tool’s data settings. Share only what the task actually needs.
An AI agent may be able to send messages, edit files, or make purchases. agent authorization defines which actions it is allowed to take. Keep a person involved for important decisions, especially before an action is sent, published, purchased, or otherwise difficult to undo.
human-in-the-loop means a person reviews or guides part of the work. Decide who checks the result and what “good enough” means before you start.
05 · Learn by doing
Three small AI literacy exercises.
Improve a request
Ask for a short explanation of a hobby. Then add the audience, a length limit, and one useful example. Compare the results and name the detail that helped most.
Check a short summary
Provide a short, non-sensitive text you know well. Ask for a summary using only that text. Mark anything that was added, missed, or changed in meaning.
Plan without taking action
Ask an AI to plan a pretend event using a fixed budget. Have it list assumptions and identify which steps need your approval. Check the totals and make the final choices yourself.
A quick check before you use an AI result
- I understand what the tool can and cannot access.
- My request includes the goal and relevant context.
- I checked the important facts against real sources.
- I protected private information and set action limits.
- I can explain why the result is useful and who is responsible for it.
If one of those steps is unclear, pause there and improve the request or review. AI literacy grows through that habit, not through memorizing every new product name.