Checking AI Output: Hallucinations, Bias and Verification
You recognize unreliable AI output and know how to verify.
Why this lesson matters
AI lies. The result is the same: you get information that looks reliable but is factually wrong. If you don't spot it, you make decisions based on nonsense.
This lesson makes you more critical. You'll learn to recognise when AI is unreliable, how to verify output, and when you're better off leaving AI out of it altogether.
What are hallucinations?
An AI hallucination is when the model presents made-up information as if it were fact.
Why does this happen?
AI models predict the most likely next word. They "know" nothing. They generate text that sounds statistically plausible. That means:
- They invent sources that don't exist
- They give confident answers to questions that have no answer
- They mix facts from different contexts into something that's wrong everywhere
Examples of hallucinations
| Situation | What AI does | Why it's dangerous |
|---|---|---|
| Legal question | Invents a statute with a realistic article number | You cite a law that doesn't exist |
| Statistic | Generates a percentage that sounds plausible | You build a business case on a made-up number |
| Looking up a person | Combines info from different people | You send an email with the wrong background |
| Citation | Invents a paper with real author names | You refer to research that doesn't exist |
Hallucinations are common. They happen with every AI model, at every provider. This is how the technology works, by design. The only question is: how do you deal with it?
Keep reading
Leave your name and email address and you can read the rest
You get the whole lesson right away, and every other lesson stays open after that. No password, no confirmation email.
Bias in AI output
AI is trained on human text. And people have prejudices. Those now sit in the model.
Types of bias
- Confirmation bias: AI confirms what you ask. If you ask "why is X good?", you only get arguments for X.
- Cultural bias: AI thinks mostly from an American/English perspective. It often misses Dutch or European nuances.
- Recency bias: Information from after the training cutoff is missing. The model doesn't know what happened yesterday.
- Selection bias: Topics with a lot written about them online get more attention. Niche topics get shallow answers.
How do you recognise bias?
Ask yourself:
- Is AI only giving me the side I wanted to hear?
- Is it missing a relevant perspective?
- Is this based on the Dutch/European context?
Test for bias by asking the opposite question. First ask "why is remote work better?" and then "why is office work better?" Compare the quality of both answers.
Verification checklist
Use this checklist for every AI output you share with others:
- 1Check the facts, Is every specific fact correct? Every number? Every name?
- 2Verify sources, If AI names sources, do they really exist? Check the URL.
- 3Check the logic, Does the conclusion follow from the arguments?
- 4Check the perspective, Is a relevant counterargument missing?
- 5Check the date, Is the information current enough for your purpose?
- 6Verify the context, Does this apply to the Netherlands/Europe or only to the US?
When NOT to use AI
There are situations where AI is simply the wrong tool:
| Situation | Why not |
|---|---|
| Legally binding texts | AI lacks legal context and case law |
| Medical diagnoses | Life-threatening if it's wrong |
| Financial reports to regulators | Every number has to be 100% verifiable |
| Current news | Training data is out of date |
| Personal references | AI invents details about real people |
Test your knowledge
What is an AI hallucination?
What is the best way to detect AI bias?
In which situation is AI output the riskiest to use without verification?
Want to keep your progress, get the practice files and sign up for the free evening? Create a free account. All you do is click a link in your email.
Create a free account