Anyone who uses AI chat tools for long eventually sees it: a confident, well-written answer that turns out to be wrong. A book that doesn't exist. A quote nobody said. A statistic with no source. A deadline from a different year. This is often called "hallucination", and it's one of the most important things for beginners to understand.
The short explanation is that AI language tools generate text that sounds right based on patterns they learned. Usually that produces accurate information, but not always, and the tool doesn't reliably signal when it's on shaky ground. You can't fully eliminate the problem with a prompt. You can reduce it a lot, and you can make errors much easier to catch.
Habit 1: Give it permission to say "I don't know"
AI tools tend to try to answer whatever you ask. Explicitly telling the AI that uncertainty is acceptable, and that you'd rather hear it, can reduce the pressure to fill gaps with guesses.
Copy-ready prompt
The honesty instruction
Weaker prompt
Stronger prompt
Habit 2: Give it the source material
The most reliable way to get accurate answers about a specific document, policy, or article is to paste the text in and tell the AI to use only that. When the information is right there in the conversation, the AI doesn't have to rely on memory.
Copy-ready prompt
Answer only from what I give you
Asking for quotes is a useful check. If the quoted sentence isn't actually in your document, you've found a problem.
Habit 3: Don't ask for sources it can't actually check
Asking an AI "give me five studies that prove this" is one of the most common ways people end up with made-up references. Some AI tools can search the web and show links; others answer purely from what they learned in training. Even with search, always open the link and confirm it says what the AI claims.
Weaker prompt
Stronger prompt
Habit 4: Ask it to check its own answer
A second pass often catches errors the first pass made. It's not foolproof, because the AI can also miss its own mistakes, but it's quick and frequently useful.
Copy-ready prompt
The self-check follow-up
Habit 5: Be careful with numbers, dates, and recent events
Specific figures, prices, laws, schedules, and anything that changes often are where errors are most likely. Many AI tools also have a training cutoff, which means they may not know about recent changes unless they can search the web. When the answer depends on current information, say so and verify it with an official source.
Copy-ready prompt
When it needs to be current
For math, ask the AI to show its working step by step, and double-check any total that matters with a calculator. It's a small step that catches a lot.
Habit 6: Ask for reasoning, not just conclusions
When the AI explains how it reached an answer, it's easier for you to spot a weak step. "Explain your reasoning step by step" turns a black-box answer into something you can follow.
Habit 7: Narrow the question
Broad questions ("tell me everything about the tax rules for freelancers") give the AI lots of room to drift. Narrow ones ("what are the general categories of expenses freelancers often track for taxes, and who should I ask to confirm what applies to me?") keep it on firmer ground and lead to more checkable answers.
When you absolutely must verify
Treat AI answers as a starting point, not an authority, for anything involving:
- Health and medication. Confirm with a doctor or pharmacist.
- Legal questions. Laws vary by place and change; confirm with an official source or a qualified professional.
- Money and taxes. Check with the relevant agency or an adviser.
- Safety. Electrical work, chemicals, car repairs, anything that could hurt someone.
- Anything you'll quote or publish. Find the original source.
This isn't about distrusting AI completely. It's about using it the way you'd use a well-read friend: great for explaining, brainstorming, and getting oriented, but not the final word on facts that matter.
Does it depend on which AI tool you use?
Different tools and versions vary, and they keep changing. Some can search the web and show sources; some can read documents you upload. No current chat tool is error-free, so the habits above are worth using no matter which one you prefer. Our ChatGPT vs Claude guide covers how to think about choosing between them.
Frequently asked questions
Why does ChatGPT make things up?
AI language tools generate text based on patterns they learned, rather than looking up facts in a verified database. Most of the time that produces accurate information, but sometimes it produces something plausible and wrong, stated just as confidently.
Can a prompt completely stop AI hallucinations?
No. Good prompting reduces the risk and makes errors easier to spot, but it doesn't eliminate them. Verify anything important with a reliable source.
Are links from an AI tool always real?
Not always. Tools that search the web can provide real links, but you should still open them and check that they say what the AI claims. Tools without web access may produce references that look plausible but don't exist.
Is it safer to paste in my own documents?
Usually, yes, for questions about that document. Asking the AI to answer only from the text you provide, and to quote it, keeps the answer grounded and easy to check.