top of page
AI-Header.jpg
A method to use AI effectively

Three moves. Before you open it, while you are in it, after you close it.

​

Before AI, think.

During AI, question.

After AI, verify and take responsibility.

before_ai.png
Before: Pause

  • Ask whether you need it at all. Some tasks are the thinking. Hand those off and you lose the exact thing you were trying to build.
     

  • Form your own rough take first. It gives you something to compare against.
    Without it you will adopt the first fluent answer you see and call it your opinion.
     

  • Decide what a good result looks like.
    If you cannot describe it, you cannot judge what comes back.
     

  • Take out anything private. Names, addresses, health details, other people's information, work you are not allowed to share. Assume anything you type may be stored and read.
     

  • Ask whether this needs a qualified human. If the answer is yes, stop here.

during_ai.png
During: Probe

  • Say precisely what you actually want. Purpose, audience, constraints, format.
    "Write about climate change" gets you an encyclopedia impression.
    "Write 200 words for a ninth grade science class on why the ocean absorbing carbon dioxide is a problem, no jargon" gets you something you can use.
     

  • Ask for three approaches, not one. The first answer is the most average answer.
    That is literally how it was produced.
     

  • Ask what it assumed. Most bad answers start with a hidden assumption you never agreed to.
     

  • Ask what it might be wrong about. Then ask what a smart person who disagrees
    would say. Both answers are useful.
     

  • Keep the consequential decisions in your hands. Use it to widen the options,
    not to pick one.

after_ai.png
After: Proof

  • Verify anything that matters against a source you found yourself.
    Not a source the AI named. A source you located.
     

  • Check what is easiest to fake: quotes, numbers, citations, dates, links, images.
     

  • Ask who is missing. Whose experience never showed up in the answer?
     

  • Rewrite it in your own voice. If you cannot explain it without the document open,
    you do not have it yet.
     

  • Say when AI helped, in any setting where someone would assume you did it alone: schoolwork, published writing, anything you put your name on.
     

  • Then take a position and own it. The output is a draft. The judgment is yours,
    and so is the responsibility.

Do not let AI decide this​

There is no version of the method that makes these safe. Get a person.

  • Medical, legal, financial, or safety-critical decisions

  • Anything happening in an emergency

  • Whether a person is honest, guilty, dangerous, or worth hiring

  • A hard emotional stretch where you need someone who knows you

  • Anything you cannot undo

What it is good at, what it is bad at

Reach for it when:

  • Medical, legal, financial, or safety-critical decisions

  • Anything happening in an emergency

  • Whether a person is honest, guilty, dangerous, or worth hiring

  • A hard emotional stretch where you need someone who knows you

  • Anything you cannot undo

The thread running through all of those: AI is strongest when you can check the result, and when roughly right is enough to move you forward.

bottom of page