Prompting like a proMiddleLesson 965 min read

Check the answer: sources, verification and hallucinations

AI can invent facts, numbers and even sources that look real. Learn how to ask better, spot warning signs, and verify what matters.

✦ Ask the tutor🧪 Lab · 25 min📝 Lesson test
Lesson in motion

In 60 seconds

Check the answer: sources, verification and hallucinations

AI can invent facts, numbers and even sources that look real. Learn how to ask better, spot warning signs, and verify what matters.

1/5
In simple words
Sometimes AI makes things up and says them very confidently, like a student guessing in an exam. So when it matters, check its answer in a real book or an official website.
A hallucination is when a model gives false information as if it were true. It can invent dates, numbers, quotes, laws, book titles, website links and even research papers. They often look perfectly normal.

Warning signs

SignWhy it matters
Very exact numbers with no sourceNumbers are easy to invent and hard to notice
Quotes from famous peopleQuotes are often made up or wrongly credited
Links, book titles or paper namesThese can be invented; open and check them
Recent events or current pricesThe model's knowledge may be old unless the app searches the web
Local rules, fees, datesThese vary by place and change often
Answer changes when you ask againA sign the model is not sure

Prompts that help

Ask for uncertainty and sourcestext
Answer the question below. For each key fact, tell me
how confident you are (high / medium / low).
If you don't know, say "I don't know".
List any sources you used so I can check them myself.

Question: What documents are usually needed to open
a savings bank account in India?
Asking for sources helps only if you open and read them. Some chatbots can search the web and show links; others cannot and may invent links. A source that doesn't exist, or doesn't say what the AI claims, is a red flag.

The verification habit

  1. 1

    Pick what matters

    Which facts would cause harm or embarrassment if wrong?
  2. 2

    Find an independent source

    Official website, textbook, trusted news, or an expert.
  3. 3

    Compare

    Does the source say the same thing, in the same way?
  4. 4

    Fix or drop

    Correct the fact, or remove it if you can't confirm it.
Weak checking
  • Asking the same chatbot "Are you sure?" and accepting "Yes"
  • Trusting a link without opening it
  • Checking only the facts you already doubted
Strong checking
  • Checking with a source outside the chatbot
  • Opening each link and finding the exact sentence
  • Checking all numbers, names and dates you will use
Watch out
Asking "Are you sure?" can make a model change a correct answer, or defend a wrong one. It is not a reliable check. Use a source outside the chatbot.
Try this. Ask a chatbot for 3 facts about your town or district with sources. Open each source. Does it exist? Does it say the same thing? Note what you find.

Watch and read more

Lab

A fact-check table for one chatbot answer, with a verdict for each claim.

~25 min

The problem

Ask a chatbot a factual question you care about (for example a historical event, a government process, or a science topic). Ask it to list its key facts and sources.

  1. Make a table with columns: Claim, Source the AI gave, Source I checked, Verdict (correct / wrong / couldn't confirm).
  2. Check at least 5 claims using sources outside the chatbot.
  3. Open every link it gave and note whether it exists and matches.
  4. Write one sentence on how much you would trust this answer, and why.

You are done when

Hard questions

Try to answer before you reveal. If you can answer these, you understood the lesson.

Q1Why might a model be more likely to invent facts about a small village than about a big city?Reveal

Models learn from text, and there is far more text about big cities. With little information, the model has weak patterns to draw on, but it still produces fluent text, filling gaps with plausible-sounding details. Rare topics are where hallucinations are most common and hardest to notice.

Q2An answer is 90% correct. Why can the 10% still be dangerous?Reveal

Because the correct parts build trust, so readers stop checking. A single wrong dosage, date, deadline or amount can cause real harm, and it hides among correct facts. This is why you check the facts you will act on, even when the rest looks right.

Please sign in to continue.

Questions people ask

Why does AI make things up?

It predicts likely-sounding text. When it lacks the right information, the most "likely" text can still be a confident, invented answer. It has no built-in way to know that it doesn't know.

Are chatbots that search the web always right?

They are often better for recent facts, but they can still misread a page, choose a poor source, or mix up details. Open the links and check the key parts.

Which facts should I always check?

Anything you will act on or share: numbers, dates, names, laws, prices, medical or financial information, and quotes.

Can I use AI to check another AI?

A second chatbot might catch some errors, but both can be wrong in the same way. Use it as an extra hint, not as proof. Official and trusted human sources come first.

Lesson test

5 questions. Get 3 right (60%) to pass and complete this lesson.

Sign in to take the test and save your progress