When AI sounds sure—but is wrong

Fluent does not mean factual

A generative AI tool builds an answer by predicting useful-looking language. It can produce a sentence that sounds polished even when a name, date, quote, calculation, or source is wrong. People often call this an AI hallucination.

The tricky part is confidence: the answer may not contain words like “I’m guessing.” That means you need a checking habit.

Sort claims into three boxes

Imagine an AI gives you a paragraph about compound growth. Mark each sentence as:

• Definition: “Compound growth includes growth on earlier growth.”
• Checkable fact: “This government agency launched a program in 2019.”
• Assumption or opinion: “A 7% return is realistic for everyone.”

Definitions still deserve a trusted reference. Dates and numbers need direct verification. Assumptions should be labelled and challenged.

The SIFT mini-check

1. Stop. Do not copy, share, or act yet.
2. Investigate the source. Is the named source real and qualified?
3. Find better coverage. Look for an official agency, textbook, library database, or established educational institution.
4. Trace the claim. Open the original page—not just a search summary or AI citation.

For a school project, record the pages you opened. For a money decision, involve a trusted adult and, when appropriate, a qualified professional.

Try it

Ask an AI tool for two sources about a historical event. Before using them:

• search the exact title;
• check the author and publisher;
• confirm the page contains the claim;
• compare it with another independent source.

A made-up citation that looks official is still made up.

Better follow-up prompts

• “Which parts of your answer are uncertain?”
• “List only sources you can quote directly, with working links.”
• “What evidence would prove this wrong?”
• “Separate facts, assumptions, and examples.”

These prompts can improve an answer, but they do not replace your own verification.