4 in 5
Kiwis used AI in the past year
1 in 5
Chatbot answers with major errors
6+
NZ court cases with AI-made-up info
Sources: InternetNZ, 2025 · EBU/BBC study of AI answers to news questions, 2025 · NZ court decisions, 2024–2026
What’s an AI hallucination?
It’s when an AI outputs something made up or wrong, as if it were true.
Chatbots work a bit like the autofill on your phone: they predict the most likely answer to your question. Most likely isn’t always true.
The tricky part is that a made-up answer can look exactly like a real one.
If it matters, check the answer yourself
Check the sources actually exist
AI tools can invent articles, people and events that look and sound completely real. Make sure you check the sources yourself.
- Ask where the information came from
- Click every link and make sure it opens
- Use a search engine to look up titles and names
Tip
Laws: legislation.govt.nz. Court decisions: courtsofnz.govt.nz. Official figures: stats.govt.nz.
Check they say what the AI claims
Even real sources can get twisted, mixed up, or credited with things they never said.
- Open the source and find the exact fact
- Use Ctrl+F or Find on Page to jump to it
- Check the date, in case it’s out of date
Example
In 2026 the NZ Supreme Court found real cases cited for things they never said, and warned it could amount to contempt of court.
Verify it yourself
Don’t let one AI answer be the final word, especially before you share it or act on it.
- Find a second, independent source
- Don’t ask the same AI to check itself
- For health, legal or money, ask an expert
Free NZ help
Healthline 0800 611 116 · Citizens Advice Bureau 0800 367 222 · your local Community Law Centre
Be extra careful when…
- It gives exact numbersPrecise stats with no source are a classic sign of guessing.
- It quotes a real personQuotes are easy to invent and hard to spot.
- It’s about recent eventsAI tools don’t always have access to recent news and information, and may make something up to fill the gap.
- It’s legal, medical or moneyIf you’re using AI for something important like your health and wellbeing, check with an expert before you act.
- It agrees with everything you sayEven if it changes its answer when you push back, that doesn’t mean the new answer is right either.
Can you tell which answer is hallucinated?
Two chatbot answers to “When did women in New Zealand get the right to vote?” Can you tell which one is hallucinated?
Answer A
NZ women won the vote in 1893, the first self-governing country to do so.
Answer B
NZ women won the vote in 1919, shortly after Britain did following World War I.
Show the answer
A is right. B was made up to show how believable a wrong answer can be. It even borrows a real detail, Britain’s 1918 law, to sound convincing. One search of nzhistory.govt.nz settles it.
You don’t have to stop using AI
It’s genuinely useful. Just match how much you check to how much is riding on it.
Usually fine
Brainstorming, rewording an email, explaining an idea in simpler words.
Check first
Facts for an assignment, figures for work, anything you’ll share with others.
Know someone who uses AI for school or work?
Send them this guide so they know what to look out for.