AI essentials: useful, responsible workflows
Use generative AI more thoughtfully: understand its limits, write clear prompts, check outputs, and protect private information with hands-on examples.
Learn it here, one small step at a time.
4 guided lessons · About 2h 30m including practice · No subscription
- Explain why fluent AI output still needs checking
- Write a specific prompt with constraints
- Review outputs for accuracy, privacy, and bias
Understand what an AI answer can and cannot tell you
Generative AI predicts likely next words from patterns in data. It can draft, summarize, and suggest ideas, but a fluent answer is not proof that a statement is true. It may invent facts, omit context, or reflect bias in its examples.
Choose a low-risk task first and decide what a correct answer should include. Use an approved tool, avoid sharing private records, and keep a person responsible for important decisions.
| Task | Useful approach | Why review matters |
|---|---|---|
| Brainstorm titles | Ask for several ideas | Check they suit the audience |
| Summarize public notes | Ask for a source-grounded draft | Compare with the original |
| Decide who gets a loan | Do not delegate the decision | Consequences require accountable review |
Ask an AI tool to suggest three headings for a public beginner guide. Check whether each heading accurately describes the material you plan to publish.
Reveal a suggested answer
No. Confidence in the wording is not evidence. Check important claims against reliable sources.
Optional further reading
Core lesson and practice are on this page. For more detail, visit the official free reference ↗.
Write a prompt with a clear job
A useful prompt says what to do, who the answer is for, what information may be used, and how the response should look. Put constraints in plain language. If the task needs facts, name the source material and tell the tool not to invent missing details.
Weak: “Explain this.” Better: “Explain the three public bullet points below to a beginner in four short sentences. Keep the numbers unchanged and say when something is not specified.”
| Part | Example |
|---|---|
| Task | Explain these three steps |
| Audience | Someone new to spreadsheets |
| Limits | Use only the text provided |
| Format | Four short sentences and one example |
Take the weak prompt above and add a named audience, a length limit, and one instruction for unknown information. Compare the two responses.
Reveal a suggested answer
It makes uncertainty visible and reduces the chance the model fills gaps with made-up detail.
Optional further reading
Core lesson and practice are on this page. For more detail, visit the official free reference ↗.
Check the output before you reuse it
Treat generated text as a draft. Verify names, dates, calculations, quotations, and links against primary sources. Run code only in a safe environment and inspect what it would change. If a claim matters, find a source that supports that exact claim, not just a source about the general topic.
A simple review log records the task, the sources checked, the changes made, and who approved the final version. Human review is part of the work, not a final decoration.
| Check | Question |
|---|---|
| Accuracy | Can I verify every important claim? |
| Completeness | What did the draft leave out? |
| Fairness | Could the wording treat groups differently? |
| Ownership | Who reviewed and approved the result? |
Ask for a short summary of a public page. Compare every sentence with the page and mark each as supported, unclear, or incorrect.
Reveal a suggested answer
Do not publish it as fact. Remove it, qualify it clearly, or find a reliable source that supports it.
Optional further reading
Core lesson and practice are on this page. For more detail, visit the official free reference ↗.
Protect private information and choose a safe workflow
Only use tools your organization has approved for the data involved. Do not paste passwords, personal learner details, customer records, confidential documents, or unpublished work into an unapproved service. Replacing a name with “Person A” may still leave enough context to identify someone.
For consequential uses, define who may use the output, who reviews it, how errors are reported, and when the workflow must stop. If the task affects a person, keep a human accountable for the decision.
| Information | Safe next step |
|---|---|
| Public generic example | Remove irrelevant details |
| Personal or customer data | Use only an approved protected tool |
| Password or secret key | Never include it; rotate it if exposed |
| Decision affecting a person | Require accountable human review |
Rewrite a real example using made-up details. List the risks that would remain if names were removed but the dates, role, and situation stayed specific.
Reveal a suggested answer
No. Other details can identify a person when combined, so sensitive information still needs protection.
Optional further reading
Core lesson and practice are on this page. For more detail, visit the official free reference ↗.
Your final knowledge check.
Score 80% or more to pass. Sign in to save your best result.