AI hallucinations — when AI generates plausible but incorrect information — are a well-documented limitation of all current language models. According to research from multiple institutions, even the best models occasionally fabricate facts, statistics, citations, and even URLs. Here's how to protect yourself.
Red Flags to Watch For
- Specific statistics without sources — "Studies show that 73% of..." Where's the study?
- Named citations — AI sometimes invents paper titles, author names, and journal references
- URLs — AI-generated URLs often lead to 404 pages
- Historical claims — Dates, names, and events can be slightly (or completely) wrong
- Technical specifications — Product features, pricing, and compatibility details may be outdated or invented
- Overly confident language — AI rarely hedges when it should
Verification Techniques
- Cross-reference with Perplexity — It provides cited sources, making verification easy
- Click every link — If AI provides URLs, open them. If they 404, the information may be fabricated
- Ask AI to cite its sources — Then verify those sources exist. If it can't provide sources, treat the claim as unverified
- Use Google Scholar — For academic claims, search Google Scholar to verify the paper exists
- Ask a second AI — If Claude and ChatGPT give different answers, dig deeper
Types of Content to Always Verify
- Medical information — Cross-reference with Mayo Clinic or NIH
- Legal information — Laws vary by jurisdiction; always consult a professional
- Financial data — Verify numbers, prices, and rates from official sources
- Historical facts — Check dates, names, and events against encyclopedias
- Company information — Verify from the company's own website or official filings
Safe Content Categories
Some types of AI output require less verification:
- Writing drafts — The factual content you provided is correct; AI just helps with structure and style
- Brainstorming ideas — You're using AI for creativity, not facts
- Code generation — You can test code directly to verify it works
- Format conversions — Transforming content from one format to another is verifiable by comparison
Building a Verification Habit
Make fact-checking routine, not exceptional. Before publishing any AI-generated content: 1) Identify all factual claims, 2) Verify statistics and dates, 3) Test all links, 4) Check named sources exist, 5) Have a human expert review domain-specific content. This takes minutes and saves your credibility. For more on using AI responsibly, browse our blog.