Why AI makes things up (and how to catch it)

What “hallucinations” are, why confident answers can be wrong, and a 30-second checking habit.

⏱ 5 min read · ▶ 1:12 video
The short version

AI chatbots predict words that sound right. Usually that matches the truth, but sometimes they invent a date, a quote, a law, a book or a web link and say it confidently. This is called a hallucination. The fix: for anything that matters, check a trusted source.

📅 Checked as of October 2026. Apps and rules change, so check the sources below if something looks different.

Why it happens

A chatbot’s language model learned patterns from huge amounts of text. When it doesn’t really “know” something, it still produces the most likely-sounding answer instead of saying “I don’t know.” OpenAI has explained that the way models are trained and scored tends to reward a confident guess over admitting uncertainty, and companies are working to reduce this.

Think of it like this: it’s like a very well-read friend who hates saying “I’m not sure,” so they fill gaps with something plausible.

Where it goes wrong most

  • Exact facts: dates, numbers, prices, opening hours, phone numbers.
  • Sources: quotes, book titles, studies, court cases and web links that look real but don’t exist.
  • Recent events: things that happened after the model was trained, unless it searched the web.
  • Local or niche topics: your town’s rules, a small business, a rare medical condition.

A 30-second checking habit

  1. Ask: “Where did you get that? Give me the source.”
  2. Click the link and make sure the page exists and says what the AI claimed.
  3. For health, money, legal or travel questions, confirm on an official site (your doctor, bank, government site or the airline).
  4. If it got something wrong, say so. It will often correct itself, but check the correction too.

Prompts that reduce mistakes

  • “If you’re not sure, say so.”
  • “Only use the text I pasted below.”
  • “List anything I should double-check.”

The Prompt Creator can add lines like these for you.

Quick tips

  • Treat AI answers as a first draft, not the final word.
  • Never rely on AI alone for medicine doses, legal deadlines or money decisions.
  • Search results with an AI summary at the top can be wrong too. See Google’s AI answers, explained.
  • Play AI or Not? to practice spotting mistakes.

Sources

Test yourself: play AI or Not? or Agent or Chatbot? AI Lingo Quiz, quick free games.

🛠️ Try it step by step

More in our How-To guides.

📖 Words to know

Hallucination (AI) Large language model (LLM) Training data AI Overview