Learn to identify when AI makes things up. Build critical thinking skills for verifying AI outputs.
AI can sound confident while being completely wrong. It doesn't "know" truth—it predicts likely text. This practice tool helps you build intuition for catching errors before they cause real problems.
Read each AI response carefully. Does it contain fabricated information?
Loading...
These examples show common patterns to watch for in AI outputs.
What AI Said: "According to a 2024 study by Dr. Sarah Mitchell at Harvard published in the Journal of Cognitive Science, users who practice prompt refinement show 47% improvement..."
Reality: No such person, study, or journal article exists. AI generated a convincing-sounding citation from patterns in its training data.
What AI Said: "In 1847, inventor Thomas Edison demonstrated his first successful telephone prototype at the Philadelphia Science Exhibition."
Reality: Edison didn't invent the telephone (Alexander Graham Bell did), the dates are wrong, and the exhibition is fictional. Plausible details combined incorrectly.
What AI Said: "Clinical trials show that taking vitamin D supplements reduces COVID-19 severity by 62% in patients over 50."
Reality: While vitamin D research exists, this specific statistic and trial doesn't exist. AI combined general knowledge into a specific false claim.
When AI generates information that sounds plausible but is factually incorrect, made up, or unsupported. This happens because AI predicts likely text patterns, not truth.
AI often invents specific percentages, dates, and numbers. The more precise the statistic, the more likely it needs verification.
Non-existent papers, authors, journals, and studies are common. AI can generate convincing academic-sounding references that don't exist.
AI often combines real information incorrectly—attributing achievements to wrong people, mixing up dates, or merging separate events.
AI expresses things it doesn't know with the same confidence as facts. It can't signal when it's uncertain or guessing.
AI's training has a cutoff date. Claims about recent events, current prices, or living people may be incorrect or outdated.
A systematic approach to checking AI outputs for accuracy.
If AI cites a study, paper, or statistic, look it up directly. AI often invents convincing-sounding citations that don't exist.
Be extra skeptical of precise numbers, dates, percentages. The more specific the claim, the more likely it needs verification.
Compare AI's claims with multiple reliable sources. If you can't find corroboration, be skeptical of the information.
Ask the AI to explain its reasoning or show its work. Hallucinations often fall apart under follow-up questions.
For legal, medical, or financial information, always verify with qualified professionals before taking action.
Domain experts can quickly spot errors that non-experts miss. Consult specialists for specialized topics.
When AI provides suspiciously precise statistics (e.g., "87.3% of users" or "founded in 1847"), these often signal fabrication. Real data is rarely this convenient.
If the information perfectly supports the point being made, verify it. AI tends to generate content that sounds right rather than content that is right.
AI's knowledge has a cutoff date. Any claims about current events, recent research, or living people's current status may be outdated or invented.
If you can't find an AI-provided fact, quote, or citation with a quick search, assume it's fabricated until proven otherwise. Real information is usually findable.