Skip to content
EN

Lesson 2 of 8 · 6 min · +100 XP

Is AI always right, and how do you check its answers?

No, AI is not always right: it can sound confident and still be wrong. To check an answer, ask for its sources, open them yourself to see that they exist and say what the AI claims, and cross-check key facts, numbers and dates with a trusted independent source. Be extra careful with quotes and named studies, which chatbots can invent.

▶ Try it

Why it matters

Chatbots write smooth, confident text, even when a detail is wrong or made up. You'll practise spotting the claims that need checking and how to check them fast.

In 6 minutes you'll be able to

  • Check exact numbers, named studies, quotes and links first.
  • Open every source and make sure it really says that.
  • A chatbot's confidence is not evidence. Use an independent source.

What you'll practise

  1. Four quick checks

    You don't need to check every word. Focus on these.

    🔗 Open the source
    Ask for sources, then click them. Does the page exist? Does it really say that?
    🔢 Exact numbers and dates
    Very precise figures are easy to invent. Find the same number in an independent source.
    💬 Quotes and names
    Did that person really say it? Search the exact quote. Made-up quotes and studies sound very real.
    🧭 Cross-check
    Look the claim up somewhere else you trust, like an official site, an encyclopedia or a textbook.

  2. What needs checking?

    Here's Lumi's answer to Sam. Tap every claim Sam should check before using it.

    Lumi
    Bees are vital pollinators. A 2019 study by the Global Bee Institute found that 87.4% of crops need bees. As a famous scientist once said: "Without bees, we have four years left." Bees communicate with a "waggle dance". Source: bee-research-journal.example/2019/crops I hope this helps with your talk! 🐝

  3. What would you do?

    Short scenes. Pick the best next move.

    1. You open the article Lumi cited. It exists, but it doesn't mention the statistic at all.

      • Drop the statistic or find a real source for it (Answer)
      • Use it, the link works after all

    2. You ask Lumi: "Are you sure?" It says: "Yes, I'm completely sure."

      • Good, that settles it
      • Check it in an independent source anyway (Answer)

    3. Lumi gives you a quick idea for a poem title. Do you need to fact-check it?

      • No, there's no fact in it to check (Answer)
      • Yes, every AI word needs a source

  4. The fact-check routine

    Put the steps in a sensible order.

    1. Pick out the claims that matter: numbers, names, quotes
    2. Ask the AI for its sources
    3. Open each source and see if it says the same
    4. Cross-check with a trusted, independent source
    5. Use only what you could confirm

Remember

Fluent isn't true. Open the source, cross-check, and leave out what you can't confirm.

Open the source. If you can't find it, don't use it.

Key facts

  • In 2023 a US federal court fined two lawyers and their firm $5,000 after they submitted a legal brief citing non-existent court cases produced by a chatbot. · Source: US District Court, Southern District of New York, Mata v. Avianca, sanctions opinion, June 2023
  • The US standards agency NIST lists "confabulation", confidently stated false or made-up content, as one of 12 key risks of generative AI. · Source: NIST, AI 600-1 Generative AI Profile, 2024
  • Chatbots can produce references, links and quotes that look real but don't exist, so a citation is only useful once you've opened it and checked what it says.
  • Looking a claim up in a separate tab, often called lateral reading, is one of the quickest ways to see whether trusted sources agree.

Updated:

Questions people ask

Can AI make up sources?

Yes. Chatbots can produce article titles, links, studies and quotes that sound real but don't exist or don't say what is claimed. Always open the source yourself.

Why does AI give wrong answers?

A language model predicts likely words from patterns in its training data. That makes text fluent, but it doesn't guarantee each detail is true or up to date.

Is asking the AI "are you sure?" a good check?

Not on its own. A chatbot may confirm a wrong answer or change a right one. Use an independent source to confirm facts.

Which AI answers need checking most?

Anything you'll rely on or share: numbers, dates, quotes, names, and health, legal or money advice. Creative ideas and wording need less checking.

Related terms