AI chatbots don't know things, they predict plausible text. Most of the time it's right enough, and sometimes it invents a court case, a statistic or a book title with total confidence.
These five checks take two minutes and catch nearly every confident error before it lands in your work.
The fast version
- Ask for sources immediately. Follow up with 'cite your sources with links'. If the model can't point to real publications, or the links don't open, treat the answer as a rumor until proven.
- Interrogate the specific numbers. Statistics and dates are hallucination magnets. Search the exact figure quoted: if you find it only on AI generated pages, that's a red flag factory right there.
The thorough version
- Check the freshness. Ask 'what's your knowledge cutoff?' in any chat. Models don't know what happened after their training date (unless they search the web): prices, laws, versions and 'the latest update' are all suspect zones.
- Cross examine a second tool. Paste the claim into a different chatbot or plain search. Independent agreement across two systems is decent evidence; one model's word alone is not.
- Trust the boring answer more. Hedge words like 'generally' and 'typically' often mark safer ground than dramatic specifics ('studies show 73.2%'). Extraordinary precision from a chatbot deserves extraordinary suspicion.
What trips people up
- Believing an answer because it's beautifully written. Fluency and truth are unrelated skills.
- Forwarding AI output to others without checking is how invented facts travel. You're the editor of record for everything you pass on.
Small tweaks, big difference
- Enable web search mode when your tool offers it: grounded answers with real citations change the game for current events.
- For medical and legal topics, AI is a great first reader and a terrible final authority. Professionals exist for the decisions.
- Keep a mental list of its weak spots in YOUR area, everyone who uses AI seriously knows where theirs stumbles.
Before you ask
How often do chatbots hallucinate?
On general knowledge, rarely enough to lull you. On niche facts, citations, very recent events and exact figures, often enough that verifying is a habit worth building.
Is enabling web search foolproof?
Better, not foolproof. It can still misread sources, so important claims still deserve your two minute check.
Keep reading
- How to Write Emails With AI Without Sounding Like a Robot
- How to Get the Most Out of Gemini on Your Android Phone
- How to Use AI to Write a Cover Letter That Doesn't Sound Like AI
The short version
Ask for sources, check the numbers, mind the cutoff date, get a second opinion, suspect perfect precision. All the truth checking you'll ever need.
