Can I trust what AI tells me?

AI can be incredibly helpful, but it doesn't always get things right.

You might ask an AI assistant a question and receive a detailed, confident answer within seconds. It can feel as though you're talking to someone who knows everything. But AI doesn't know everything, and sometimes it makes mistakes.

That doesn't mean you shouldn't use it. It just means it's worth knowing when to trust an answer and when to check it.

Why does AI sometimes get things wrong?

AI assistants learn patterns from enormous amounts of information and use those patterns to produce answers. They're very good at explaining things, suggesting ideas and helping with everyday tasks.

But they can also:

• Give information that's incorrect or out of date.
• Confuse people, places, dates or events.
• Misunderstand what you're asking.
• Make up details, quotations or references that sound genuine.
• Give an answer without explaining that there's uncertainty.

Sometimes this is called an AI hallucination. That's a technical term for when AI produces information that isn't true or isn't supported by reliable evidence.

A confident answer isn't necessarily a correct one.

What happens if you challenge an AI answer?

Sometimes AI will tell you something as though it's an established fact, even when it's wrong. If you question the answer, it might then admit that it made a mistake, misunderstood something or provided information it couldn't support.

That doesn't sound right to me. Are you sure?

The AI might apologise and give you a completely different answer. But just because AI changes its answer doesn't mean the new answer is correct either. AI can sometimes agree with you simply because you've challenged it.

Think about it the same way you would when someone tells you something in everyday life. You don't have to believe it simply because they say it confidently.

Ask questions. Think about what you already know. Consider whether the answer makes sense, and check reliable sources when it matters. Your experience, common sense and judgement are valuable. Use them.

Does that mean I can't trust AI at all?

Not at all. Think of AI as a helpful assistant rather than an expert who's always right. It can help with recipes, birthday messages, unfamiliar words, gardening ideas or history. For these sorts of questions, it can be a wonderful starting point.

What sort of information should I check?

Some mistakes matter more than others. If AI suggests a recipe you don't like, you can try something else. But incorrect information about your health, finances or safety could be serious.

• Health: Medicines, symptoms or treatments.
• Money: Banking, investments, pensions or taxes.
• Legal matters: Rights, obligations or documents.
• Online safety: Whether a message is genuine or what to do after a scam.
• Current events: Breaking news or recent developments.
• Travel: Flight schedules, entry requirements, prices or opening hours.

For important decisions, use an official source or seek appropriate professional advice.

How can I check an AI answer?

1. Ask where the information came from

Where did you get that information? Can you show me a reliable source?

Some AI assistants can search the internet and provide links. AI can sometimes provide incorrect or made-up references, so open the links and check what they actually say.

2. Check with a trusted source

For health information, ask a doctor or pharmacist or consult Health New Zealand. For banking, use your bank's official website. For government services, use official government websites. Museums, libraries and recognised archives can help with history.

3. Check the date

News, sports results, travel arrangements and other details can change. Ask when the information was last updated and check it against a current source.

4. Ask AI to explain its answer

Is there anything about this answer that might be uncertain?

This can help you understand an answer, but AI isn't guaranteed to spot its own mistakes.

An example: Exploring aviation history

Imagine asking about de Havilland Vampire aircraft used by the Royal New Zealand Air Force. AI may give interesting details about the aircraft and pilots, but a particular serial number or event date could be wrong.

Can you help me find official RNZAF records or museum information to check those details?

That way AI helps you explore your interest while pointing towards sources you can check.

What if two AI assistants give different answers?

Different assistants may use different information or interpret questions differently. Rather than asking more assistants until one agrees with you, check a reliable source.

One last thing

You don't need to distrust everything AI tells you, but you shouldn't believe everything automatically either. Enjoy exploring, learning and creating. When something matters, take a moment to check.

AI can help you find answers. You still get to decide what to believe and what to do next.