Back to Learn
The AI Foundation
Getting Started Guide

How to Know When AI is Wrong: A Practical Guide to Fact-Checking AI Output

Learn to spot AI hallucinations, verify claims, and build habits that prevent costly mistakes from artificial intelligence responses.

Apr 21, 20267 min read

# How to Know When AI is Wrong: A Practical Guide to Fact-Checking AI Output

Artificial intelligence tools like ChatGPT, Google Bard, and Claude have become incredibly helpful for everything from writing emails to researching topics. But here's the catch: AI can be confidently wrong. It might tell you that penguins live in the Arctic (they don't—they're in Antarctica) or give you a recipe that's missing crucial steps.

This phenomenon, called "hallucination," happens when AI generates information that sounds plausible but isn't accurate. The good news? You can learn to spot these mistakes and verify AI responses before they cause problems.

Understanding AI Hallucinations: Why AI Gets Things Wrong

AI doesn't actually "know" things the way humans do. Instead, it predicts what words should come next based on patterns in its training data. Think of it like an incredibly sophisticated autocomplete system that sometimes fills in gaps with educated guesses rather than facts.

Common types of AI mistakes include:

  • Outdated information — AI training data has cutoff dates, so recent events might be missing or incorrect
  • Fabricated details — AI might invent specific dates, names, or statistics to complete a response
  • Misunderstood context — Complex topics might get oversimplified or mixed up
  • Source confusion — AI might combine information from multiple sources incorrectly
  • Red Flags: When to Be Extra Cautious

    Certain situations require heightened skepticism when using AI:

    High-Stakes Decisions

    Never rely solely on AI for medical advice, legal guidance, financial decisions, or safety-critical information. Always consult qualified professionals for these topics.

    Specific Facts and Figures

    Be particularly wary when AI provides:

  • Exact dates, especially recent ones
  • Specific statistics or percentages
  • Phone numbers, addresses, or contact information
  • Direct quotes from people
  • Technical specifications or measurements
  • Current Events and Recent Information

    Most AI models have training data cutoffs, meaning they might not know about recent developments. If you're asking about anything that happened in the past year, double-check the information.

    Niche or Specialised Topics

    AI might struggle with highly technical subjects, local information, or topics with limited online coverage.

    Quick Verification Techniques You Can Use Right Now

    The Cross-Reference Method

    Don't stop at one source. If AI tells you something important:

  • Search for the same information using Google, Bing, or DuckDuckGo
  • Look for multiple sources that confirm the same details
  • Check official websites when possible (government sites, company pages, academic institutions)
  • Use fact-checking sites like Snopes, FactCheck.org, or PolitiFact for controversial claims
  • The Reverse Question Technique

    Ask AI follow-up questions that test its initial response:

  • "What sources support this claim?"
  • "Are there any counterarguments to this position?"
  • "How recent is this information?"
  • "What are the limitations of this approach?"
  • If AI can't provide satisfactory answers to these follow-ups, be more skeptical of the original response.

    The Common Sense Check

    Does the AI's response pass basic logic tests?

  • Are the numbers reasonable? (A recipe serving 4 people shouldn't need 20 cups of flour)
  • Do the timelines make sense? (Events should happen in logical order)
  • Are the claims too good to be true? ("This one simple trick" solutions are often oversimplified)
  • Reliable Verification Tools and Resources

    Free Fact-Checking Resources

  • Google Scholar — (scholar.google.com): Find academic papers and research
  • Wikipedia — Good starting point, but check the references at the bottom
  • Government websites — End in .gov or .gc.ca for Canadian sites
  • Wayback Machine — (archive.org): See how websites looked in the past
  • News and Current Events

  • AllSides — (allsides.com): Shows news from different political perspectives
  • Ground News — (ground.news): Compares coverage across multiple news sources
  • Reuters — and **Associated Press**: Generally reliable wire services
  • Specialised Verification

  • Wolfram Alpha — (wolframalpha.com): Computational knowledge engine for math and science
  • PubMed — (pubmed.ncbi.nlm.nih.gov): Medical and scientific research database
  • Statistics Canada — (statcan.gc.ca): Official Canadian statistics
  • Building Better AI Interaction Habits

    Start with Better Prompts

    Instead of asking "What's the best way to invest money?", try:

    "What are some commonly recommended investment strategies for beginners, and what are the potential risks of each?"

    This approach encourages more balanced, nuanced responses and signals that you want multiple perspectives.

    Use AI as a Starting Point, Not an Endpoint

    Treat AI responses like a rough draft or brainstorming session. Use them to:

  • Identify topics to research further
  • Generate questions you should ask experts
  • Find keywords for your own searches
  • Understand different perspectives on an issue
  • Document Your Verification Process

    For important decisions, keep notes about:

  • What AI told you initially
  • Which sources you checked
  • Any contradictions you found
  • Your final conclusion and reasoning
  • This creates a paper trail and helps you improve your fact-checking skills over time.

    When You Can (Cautiously) Trust AI More

    Some AI responses are generally more reliable:

    Well-Established Information

  • Historical facts from decades ago
  • Basic scientific principles
  • Common mathematical calculations
  • General explanations of how things work
  • Creative and Brainstorming Tasks

  • Writing inspiration and idea generation
  • Brainstorming session facilitation
  • Creative writing assistance
  • Learning new concepts (with verification)
  • Process and Structure

  • Email templates and formatting
  • Project planning frameworks
  • Step-by-step process outlines
  • Organisational systems
  • Even in these "safer" categories, spot-check important details before acting on them.

    Your First Steps: Getting Started Checklist

    Ready to become a smarter AI user? Here's your action plan:

    This Week:

  • ☐ Choose one AI tool you use regularly (ChatGPT, Claude, Google Bard, etc.)
  • ☐ Practice the "reverse question technique" on three different AI responses
  • ☐ Bookmark three fact-checking sites that are relevant to your interests
  • ☐ Try cross-referencing one piece of information AI gives you
  • This Month:

  • ☐ Develop a personal verification routine for high-stakes AI responses
  • ☐ Join or follow one reliable news source and one fact-checking organisation
  • ☐ Practice writing more specific, nuanced prompts instead of simple questions
  • ☐ Keep a simple log of when AI got something wrong and how you caught it
  • Ongoing Habits:

  • Always verify before sharing AI-generated information with others
  • Treat AI responses about recent events with extra skepticism
  • When in doubt, consult human experts for important decisions
  • Stay curious and keep learning about AI's limitations and capabilities
  • Where to Go Next

    Want to deepen your AI literacy? Here are some next steps:

    Learn More About AI:

  • Follow The AI Foundation on social media for regular tips and updates
  • Read "The AI Foundation's Guide to Responsible AI Use" (theaifoundation.ca)
  • Join online communities focused on AI literacy and critical thinking
  • Develop Critical Thinking Skills:

  • Take a basic course in logic or critical thinking
  • Learn about cognitive biases and how they affect decision-making
  • Practice evaluating sources and identifying reliable information
  • Stay Updated:

  • Subscribe to newsletters from reputable fact-checking organisations
  • Follow AI researchers and journalists who cover AI responsibly
  • Regularly review and update your verification techniques as AI technology evolves
  • Remember: The goal isn't to stop using AI, but to use it more wisely. With these skills, you can harness AI's incredible capabilities while protecting yourself from its occasional mistakes. Start small, be consistent, and gradually build your confidence in navigating our AI-enhanced world.

    AI is a powerful tool, but like any tool, it works best when you understand both its strengths and its limitations. By developing these fact-checking habits now, you're setting yourself up for success in an increasingly AI-integrated future.

    Stay in the Loop

    Get updates on upcoming AI workshops, resources, and insights for Canadian organizations.

    No spam, ever. Unsubscribe at any time.