Quick answer: To tell if an AI answer is wrong, check the specific claim against a trusted source, confirm that any citation really exists and says what the AI claims, and compare with a second tool for important questions. AI can sound confident and still be wrong, so never rely on a chatbot alone for health, legal, money or study facts.
AI chatbots can produce answers that look credible but are fabricated or inaccurate. The industry term is hallucination. It affects every major assistant, including free and paid versions.
This checklist works for students, workers and travellers in the US, UK, Canada, Australia and Europe. It is general information, not professional advice.
Why do AI chatbots get things wrong?
VCU Libraries explains that generative AI mimics its training data, so flaws in that data can carry over, and it can produce information that looks correct but is not. Answers can sound “just as plausible when they are correct as when they are incorrect,” and citations may be misattributed, fabricated or point to sources that do not contain the claimed ideas.
Codecademy adds that AI cannot reason critically, so the checking is up to you.
How do I check an AI answer step by step?
- Decide how much it matters. The University of Nevada, Reno advises a light check for low-stakes information and a thorough one for anything with real-world impact.
- Check the claim, not the chatbot. Look for the same claim in another trusted source such as a government site, a trusted news outlet or a research database.
- Verify every citation. Search for the cited article or book directly. If you cannot find it, it may be fabricated.
- Open the link. Campbellsville University’s library guide says to read every source yourself and confirm it matches the claim. Its example: a confident answer about one university linked to a different school.
- Compare tools. Codecademy suggests asking another AI tool the same question; disagreement can flag a likely hallucination.
- Interrogate the answer. Ask for sources, more detail or how confident it is, then verify what it gives you.
What are the warning signs?
- Precise figures, dates or quotes with no source.
- Links or references you cannot find when you search for them.
- A real source that does not actually say what the AI claims. The Campbellsville guide describes Copilot citing a real library page but wrongly implying the library housed the writing center.
- A topic you know little about. Codecademy advises most care exactly there.
Images and audio need the same skepticism; see our guide to how to spot AI-generated images.
Does a better prompt help?
Somewhat. Codecademy recommends stating your intent precisely and narrowing the prompt when a reply misses the mark. Campbellsville notes that asking for peer-reviewed or reliable sources in the prompt is “still not enough” on its own. Our guide to writing better AI prompts covers the basics. Search-style tools are checkable too: see how Google AI Overviews work and how to check their answers.
Do AI search tools make this unnecessary?
No. VCU notes that retrieval-based tools summarize real human-written sources, but the summary can still be inaccurate if sources are off-topic or the summary distorts them or adds unsupported content. Compare the summary with the linked original.
Codecademy is clear that these methods reduce hallucinations but cannot eliminate them. For academic work, find the specific sources through research databases and search engines rather than relying on AI-generated references (University of Nevada, Reno).
What is a quick routine for everyday questions?
For low-stakes questions, such as a recipe swap or a film trivia point, a fast search to confirm the main claim is usually enough. For travel, study or money questions, go further:
- Confirm dates, prices, opening hours and rules on the official site of the airline, venue, school or government body.
- Ask yourself who can verify the claim, then go to that source.
- Keep a note of where each key fact was confirmed, so you can show your work.
The point is not to distrust every answer. It is to match the effort to the stakes, as the Nevada guide suggests, and to treat the chatbot as a starting point rather than the final word.
Which mistakes do people make most?
One common mistake is trusting a confident tone. VCU notes wrong answers can sound as plausible as right ones, so tone tells you nothing. Another is checking only that a link exists, not that it supports the claim. A third is asking the same chatbot to confirm itself. A second tool, or better a human-written reference, gives a more independent check.
More practical guides are in our more AI coverage section.
Frequently asked questions
Can AI make up sources?
Yes. Citations can be fabricated or misattributed. Search for the title directly in Google Scholar or a library database to confirm it exists.
Is asking the chatbot “are you sure?” enough?
No. You can ask for sources or confidence, but you still need to verify the facts yourself against a trusted source.
Are paid AI plans more accurate?
The sources reviewed here do not compare plans, so do not assume so. Check important answers whichever plan you use.
When should I never rely on AI alone?
For anything with real-world impact such as health, legal, financial or academic work, confirm with a qualified professional or an authoritative source.