Always question AI

This is the most important lesson of all. AI is a powerful tool, but it has a dangerous flaw: when it does not know, it does not say "I don't know". It makes up an answer and writes it with the same confidence it has when it is right. In the AI world this has a name: hallucination.

It is not rare, and the risk is greatest where it matters most

The best models have improved a lot on general questions, but on technical topics the error rate is still high. In legal queries, studies point to fabrication rates ranging from 6% to more than 50% depending on the tool. In document summaries, the newer reasoning models exceed 10% hallucination. And one figure that is alarming: nearly half of corporate AI users have already made at least one business decision based on information the AI made up.

Two reasons beyond making things up

The golden rule for construction

Never use a number, a contract clause or a code requirement that came from AI without human checking. For code and standards, confirm with the local jurisdiction or with the specialized tool. For contracts and taxes, confirm with a lawyer and an accountant. AI speeds up the draft; it does not take on the responsibility, you do.

How to question it in practice

• Ask for the source: "where did you get this from? give me the link or the reference".
• Be suspicious of numbers that are too round or too specific. Ask "are you sure?" and see if it backs down.
• For current topics, use an AI with internet access and ask it to look up up-to-date information.
• Check every statement that depends on a date, a figure or a law against the official source.

Using AI well is not about believing it. It is about using it like a fast, helpful intern who produces a lot, but whose work you always review before you sign off.