Think carefully
before you use AI

8 min read

The core

AI makes creation easier. That’s why one skill is more important than ever: analytical thinking.

The problem

Output feels like the truth because it’s written convincingly. But convincing isn’t the same as correct.

The solution

You’re the director. AI is an assistant. Directing means iterating, verifying, comparing, and adjusting.

5 questions before you start

1

What exactly is the question, and what is the context?

2

What information do I need to support this?

3

How will I gather, evaluate, and select that information?

4

What am I deliberately omitting because it’s not relevant?

5

What assumptions am I making, and are they accurate?

De Bono’s Six Thinking Hats

A thinking model that helps you evaluate AI output critically and systematically. Wear one hat at a time.

White hat

Facts and data. What do we really know?

Red hat

Feelings and intuition. What does it evoke?

Black hat

Risks and weaknesses. What could go wrong?

Yellow hat

Value and benefits. What are the benefits?

Green hat

Creativity and alternatives. What else is possible?

Blue Hat

Process and management. What’s the next step?

The powerful thing is that this also helps you use AI more effectively. You create different prompts for each hat. You force yourself not to try to cram everything into a single answer.

The hardest step: validation

AI writes convincingly, even when it’s wrong. That’s why validation should be an explicit part of the process.

1

Apply a source hierarchy

Give priority to primary sources such as scientific publications, official statistics, and policy documents. Use blogs only as a guide.

2

Check each claim individually

Don’t let AI generate a single narrative that you then have to unravel. Break down the claims and verify them.

3

Triangulation

Verify key claims using at least two independent sources, not two articles that simply copy the same press release.

4

Context check

Always ask whether this applies to the Netherlands, to this sector, to this year, or to this target audience. Much of the content is generic or biased toward the U.S.

5

Counterargument required

For each conclusion, present a serious counterargument, including why you’re not following it after all.

6

Transparency in your work

Document assumptions, uncertainties, and data points you were unable to verify. That’s not a weakness, it’s professional.

From copy-paste to taking control

A practical method that both students and professionals can apply right away.

Step 1

Conduct in-depth research with AI

Start with targeted research using your favorite AI tool. Gather initial insights and structure.

Step 2

Repeat with a different model

Use a second AI tool or model to explore the same question. Differences provide valuable insights.

Step 3

Overlay the results

Compare both outputs. Highlight similarities, contradictions, and gaps in the information.

Step 4

Formulate follow-up questions

Use the gaps as a basis for new, more specific questions. Keep iterating until you’re satisfied.

Core principle

This forces you to think in steps and to treat output as input, not as truth.

Bullshit in, bullshit out.

Banning things is just treating the symptoms. Continuing to think for yourself is the real skill. Learn to collaborate with AI without losing your own judgment.