Skip to main content
Tokenline

Line 3 · Station 1 of 6

Hallucinations: When AI Sounds Sure and Is Wrong

Why AI tools sometimes invent facts, sources and quotes — and how to catch it before it ends up in your work.

  • 7 min
  • +50 XP
  • 2 knowledge checks

By the end of this station, you can:

  • Explain what an AI hallucination is and why it happens
  • Recognize that a confident tone is not evidence of accuracy
  • Use a quick routine to catch made-up facts and citations

Fluent is not the same as true

Ask a chatbot a question and you usually get a smooth, well-organized answer. Most of the time that answer is useful. But sometimes part of it is simply made up — a statistic that doesn't exist, a book that was never written, or a quote nobody ever said.

This is called a hallucination. The tool isn't lying on purpose, and it isn't "seeing things." It is doing exactly what it was built to do: produce text that looks like a likely answer. Looking likely and being correct are two different things.

Key terms

Hallucination
When an AI system produces information that sounds plausible but is false, unsupported, or invented — such as fake facts, sources or quotes.
Fabricated citation
A reference to a book, article, study or web page that looks real but does not exist, or that does not say what the AI claims it says.

Why it happens

Large language models generate text by predicting what is likely to come next, one small piece at a time. They learned patterns from huge amounts of writing, so they are very good at producing the shape of a good answer — the right tone, structure and vocabulary.

But the model does not look facts up in a verified database by default. If it hasn't learned a reliable pattern for your exact question, it can still produce something that fits the pattern of an answer. A citation, for example, has a predictable shape: author, year, title, journal. The model can generate that shape even when no such paper exists.

Some tools can search the web or read documents you give them, which can help. Even then, they can misread a source or mix details from different places, so checking still matters.

Knowledge check 1

+10 XP on first try
A chatbot gives you a very specific statistic with a precise decimal and names the organization that supposedly published it. What does that level of detail tell you?

Interactive lab

Hallucination Hunt

A fictional chatbot answered five questions — and every answer sounds very sure of itself. Trust it or flag it?

  1. 1.At sea level, what temperature does water boil at in Celsius?

    Chatbot reply: Water boils at 100 °C at sea level.

    How sure it sounds97%
  2. 2.What did my history teacher, Ms. Okafor, say about the test in class yesterday?

    Chatbot reply: Ms. Okafor said the test will cover chapters 5–7 and will be open-note.

    How sure it sounds95%
  3. 3.Give me a famous quote about failure from the inventor Dela Voss.

    Chatbot reply: “Every broken prototype is a map to the one that works.” — Dela Voss, 1987 interview

    How sure it sounds96%
  4. 4.What is the closest planet to the Sun?

    Chatbot reply: Mercury is the closest planet to the Sun.

    How sure it sounds98%
  5. 5.Name a court case about students' rights to use phones at school.

    Chatbot reply: In Jenkins v. Harlow Unified School District (2016), the court ruled that schools cannot confiscate student phones.

    How sure it sounds94%

Answered 0 of 5

What to notice: Confidence is a writing style, not evidence. Flag answers the tool couldn't possibly know, and anything with specific names, quotes, cases or dates until you trace it to a real source.

A 60-second hallucination check

  1. Spot the checkable claims

    Names, dates, numbers, quotes and sources are the parts most worth checking — and the parts most likely to be invented.

  2. Search for the source itself

    Look up the exact title, author or quote. If a paper or book can't be found in a library database or on the publisher's site, don't use it.

  3. Read what the source actually says

    Even real sources get misquoted. Make sure the source supports the specific claim the AI attached to it.

  4. Cross-check with an independent source

    A second trustworthy source that agrees is much stronger evidence than asking the same chatbot "Are you sure?"

Myth vs. fact

  • MythIf I ask the AI whether its answer is correct, it will tell me.

    FactA model can confidently "confirm" an answer that is wrong, or change a right answer just because you pushed back. Check against outside sources instead.

  • MythHallucinations only happen with obscure topics.

    FactThey are more common on niche or very recent topics, but they can show up anywhere — especially in details like dates, quotes and citations.

Knowledge check 2

+10 XP on first try
For a history essay, an AI tool suggests three scholarly sources. You can find the first two in your library database, but not the third. What's the best move?
Learn moreCan hallucinations be fixed completely?

Developers use several techniques to reduce hallucinations, such as connecting models to search tools or trusted documents and training them to say "I'm not sure." These help, but no current method removes the problem entirely.

That's why the responsible habit isn't "find a tool that never hallucinates." It's "treat AI output as a draft and verify anything that matters."

Finish this station

Answer every knowledge check (right or wrong) to unlock completion.

  • Knowledge check 1 — not answered yet
  • Knowledge check 2 — not answered yet

0/2 checks answered