Most of the frustration with AI comes from using it for the wrong pile of tasks. Ask it for twenty ideas and it is brilliant. Ask it for the one true fact and it is a gamble. The tool did not change between those two requests; the job did. Sorting jobs before you start is the single habit that removes most of the risk.
| AI is the right tool (generate, shape, explain) | AI is a starting point only (facts about the world) |
|---|---|
| Brainstorming ideas, names, angles | Prices, limits and plan details |
| Drafting from your own notes and facts | Laws, rules, deadlines and eligibility |
| Rewriting for tone, length or audience | Medical, financial or legal specifics |
| Explaining a settled concept | Anything about a real person |
| Structuring: outlines, tables, checklists | Quotes, citations and references |
| Practising: quizzes, role-play, language drills | Recent events and anything that changed this year |
Keep the brainstorm out of the fact pile
A brainstorm goes wrong when the model slips facts into it — a product that does not exist, a statistic to make an idea sound better. You can tell it not to.
Your turn
A brainstorm that stays a brainstorm
Give me [NUMBER] ideas for [TASK]. One line each, numbered. Do not include any statistics, prices, product names or claims about real companies or people — ideas only. If an idea needs a fact to work, write "(needs checking: …)" after it.
Hint: The last sentence is the useful one: it makes the model label the ideas that would need lesson 4's check.
The final authority for nothing
This is the sentence to keep: AI is the final authority for nothing. Not because it is usually wrong — it is usually right — but because it has no way of knowing when it is not, and neither do you from the text alone. For anything that matters, the last step is a source that is not a model: the vendor's page, the statute, the document, the person. AI gets you to that step faster. It does not replace it.
Knowledge check
Which of these tasks needs no fact-check afterwards?