AI is becoming part of everyday life and work. Learn what AI hallucinations are, how to spot warning signs and what to check before trusting an AI-generated answer.

Artificial intelligence can write, explain, summarise and research information in seconds. But there is an important limitation every user should understand.

AI can give you an answer that sounds completely convincing and still be wrong.

This is commonly called an AI hallucination. NIST uses the term "confabulation" for AI systems producing false or erroneous information that can appear convincing. Fabricated citations and incorrect supporting information are among the recognised risks.

How Big Is the Problem?

The answer depends on the AI model, the question and how accuracy is tested.

Stanford's 2026 AI Index tested 26 leading models on a specific factual reliability benchmark. Reported hallucination rates ranged from 22% to 94% in that test.

That does not mean AI gets 22% to 94% of everyday questions wrong. It shows that reliability can vary considerably between models and tasks.

South Africa Has a Real Example

In April 2026, South Africa's Draft National AI Policy was withdrawn after the Department of Communications and Digital Technologies confirmed that fictitious sources appeared in its reference list.

The minister said AI-generated citations were the most plausible explanation and acknowledged that proper human verification had not taken place.

It is a useful reminder that an AI-generated source can look legitimate without actually existing.

What Should You Look Out For?

Be extra careful when AI gives you:

• Very specific statistics, dates or names
• Quotes or studies you cannot find
• Legal, tax or government information
• Medical or financial advice
• Current prices, regulations or news
• Links or references that do not open or do not support the claim

A polished answer is not proof.

What Should You Do?

You don't need to stop using AI. Instead, match your checking to the importance of the information.

For anything important:

1. Find the original source.

2. Check that it actually supports what the AI said.

3. Check the date to make sure the information is current.

4. Compare important claims with another reliable source when appropriate.

5. For high-stakes decisions, use the relevant professional, authority or original documentation rather than AI alone.

And think carefully before putting personal or confidential information into an AI service.

The Bottom Line

AI is a powerful tool, but it is not an automatic fact-checker.

Use it to help you work faster, explore ideas and understand information.

Just remember one simple rule:

When the answer matters, don't just ask AI. Check it.