AI Can Sound Certain and Still Be Wrong
AI hallucinations are false or invented answers presented with confidence. Once you know what raises the risk, checking AI output becomes a normal part of using the tool.
What is an AI hallucination?
A hallucination happens when an AI system generates information that is false, invented or unsupported. The UK's National Cyber Security Centre warns that large language models can hallucinate incorrect facts while still sounding convincing.
AI does not reliably know when it is hallucinating. It can present an invented claim with the same confidence it uses for a correct one, which is why the user must check important output.
Why it happens in plain English
Large language models generate text by predicting what is likely to come next based on patterns in training data. They are not automatically checking every claim against a trusted source.
- Missing information. The model may fill a gap with something that sounds plausible.
- Old information. The model may not know about a recent event or change.
- Vague instructions. The model may invent details to complete an unclear request.
- Rare topics. There may be fewer reliable patterns for the model to draw from.
What hallucinations can look like
- Made-up sources. In 2025, legal representatives submitted citations for cases that did not exist to the English High Court. The court's published judgment described citing non-existent authorities as improper and emphasised the duty to check AI-assisted legal work.
- Outdated facts. The answer describes an old company leader, policy or product as if it were current.
- Confident nonsense. A detailed explanation sounds authoritative but falls apart when checked by an expert.
How to spot a possible hallucination
- Check whether it is verifiable. Look for an official source, trusted publication or original document.
- Inspect specific details. Verify every citation, date, quotation and statistic.
- Compare it with what you know. A contradiction is a reason to pause and investigate.
- Consider the topic. Recent, niche and specialised questions carry a higher risk.
- Notice overly neat answers. A perfect structure or very confident tone does not prove accuracy.
A fact-checking habit you can start today
- Check key facts. Use official websites, original documents and reliable publications.
- Verify citations and quotations. Confirm that they exist and say what the AI claims.
- Use common sense. Ask whether the answer is realistic and internally consistent.
- Ask a follow-up question. See whether the AI changes or contradicts its answer.
- Get a human second opinion. Do this for high-stakes or specialist decisions.
You do not need to fact-check every casual idea. You do need to verify anything important before you rely on it.
Prepared by AI Did This
Last reviewed: 28 July 2026
Sources
- UK National Cyber Security Centre: ChatGPT and large language models: what's the risk? (opens in a new tab) (14 March 2023)
Explains hallucinated facts, convincing output and the risks of niche queries.
- Courts and Tribunals Judiciary: Ayinde v London Borough of Haringey and Al-Haroun v Qatar National Bank (opens in a new tab) (6 June 2025)
The official High Court judgment concerning non-existent case citations.
- Courts and Tribunals Judiciary: Artificial Intelligence (AI): Judicial Guidance (opens in a new tab) (31 October 2025)
Official guidance on hallucinations, confidentiality and personal responsibility.