Part 2 - Minimising AI Hallucinations

AI is good at sounding confident - even when it's completely wrong.

Whether it's making up quotes, faking links, or summarising something that was never said, hallucinations are common across tools like ChatGPT, Claude, and Perplexity.

The good news? You can spot (and reduce) most of them with a few simple habits.

What Is a Hallucination?

It's when the AI makes something up.

That might look like:

Invented stats or case studies
Incorrect summaries
Broken links or citations that lead nowhere
Descriptions of tools or laws that don't exist
Confident answers based on a wrong assumption

How To Minimise Hallucinations

1. Be Clear About What You Want

Vague prompts lead to vague answers.

Instead of:

Summarise this for me

Try:

Give me 3 key points from this document, without adding examples

Be specific about what you don't want too:

Don't make up stats or cite anything unless it's in the document.

2. Check What It Actually Says

If you're using AI to summarise, always glance back at the source. Especially for:

Policies or compliance documents
Anything going public
Internal comms

AI is helpful - but not trustworthy by default.

3. Don't Make Things Up

If AI makes something up, and you copy it into a document or deck, you become the one who's making things up.

So:

Read it as a draft, not a final answer
Ask it clearly:

Where did this come from?
Did you make that up?

If you're not sure, leave it out

Good rule of thumb: if you wouldn't say it in a meeting without checking, don't paste it in from a chatbot.

Adding don't make anything up or only use information from the context helps reduce this, but not completely.

4. Ask For Sources (And Follow Them)

Especially in tools like Perplexity or NotebookLM, you can say:

List your sources.
Which document is this based on?

But don't assume links work. Click them. Scan them. Check dates.

If a source doesn't exist - delete the claim.

5. Watch The Blur

Hallucinations often show up in summaries that blend ideas:

It's widely known that...
In general, experts agree that...
Most people feel that...

These phrases often mean the AI is stitching together a vague average of internet chatter. Ask it to be more precise - or just leave that part out.

Checklist: Don't Trust, Verify

Did I ask a clear question?
Does this match the original source?
Are any stats or quotes backed up?
Do the links actually work?
Does it sound too confident without evidence?

Minimising AI Hallucinations

Want More Resources Like This?