You know when…

You know when someone is telling a story and they can’t quite remember a detail, so they fill in the gap with something that sounds right? They do not pause and say “I am not sure.” They just keep going and you do not question it because the rest of the story checks out. That is exactly what AI does. Except AI does it with your client proposals, your compliance documents and your website content.

So what is an AI hallucination?

An AI hallucination is when an AI generates information that is factually incorrect, fabricated, or has no basis in its training data but presents it with the same confidence as everything else it writes. The AI is not guessing and it is not deliberately making things up. It is doing what it always does: predicting the most likely next word. Sometimes that prediction lands on something true. Sometimes it lands on something plausible but completely invented. The AI cannot tell the difference.

Does that mean the AI is just making everything up?

No. The vast majority of what a well-prompted AI produces is accurate or directionally correct. Hallucinations are not the norm. They are the failure mode. The problem is that you cannot tell from the output alone which parts are real and which parts are invented because the AI delivers both with identical confidence. That is what makes hallucinations dangerous. It is not that they happen often. It is that when they happen they are not obvious.

How can I catch hallucinations before they cause damage?

1. Never let AI be the last set of eyes.

Treat every AI output as a first draft, not a final product. If the content includes names, dates, regulations, statistics, or quotes, verify them against a primary source before publishing or sending. This is not optional. It is the single most important habit in responsible AI use.

2. Ask the AI to show its working.

Add “explain your reasoning” or “cite your sources” to your prompt. The AI may still hallucinate the sources, but the act of forcing it to justify its claims makes the output easier to fact-check. If it cannot explain where a claim came from, that is your red flag.

3. Keep your prompts specific and bounded.

Hallucinations spike when the AI is asked to operate outside its depth. Broad, vague questions with no constraints give the AI room to invent. The narrower and more specific your prompt, the less room the AI has to fill gaps with invention. “Summarise this document” hallucinates far less than “tell me everything about this topic.”