AI Foundations · Guide 4 of 7
Why AI Makes Things Up
AI sometimes states things confidently that are completely false. This is not a glitch. It is a predictable result of how AI works — and once you understand it, you'll never be blindsided by it again.
After This Guide, You Will Be Able To
Identify when an AI response may be hallucinated — and know what to do next.
What Is Hallucination?
Hallucination is when AI produces output that sounds accurate but is factually wrong.
It is not a bug. It is not a malfunction. It is a natural result of how AI works.
Remember from Guide 1: AI was trained to predict what language fits — not to verify what is true. When AI encounters a question it does not have reliable data for, it does not say "I don't know." It produces the most plausible-sounding response it can.
It fills the gap with what sounds right. It does not know it is wrong. It has no mechanism for knowing it is wrong.
A hallucinated response does not sound like a guess. It sounds like a fact. It is delivered with the same confident tone as everything else AI says. There is no warning. No disclaimer. No signal that something is wrong. That is what makes hallucination dangerous for people who do not know to look for it.
Three Hallucination Patterns
Fake Citations
AI invents books, studies, authors, and URLs that sound real. The titles are plausible. The authors seem credible. The links go nowhere — or worse, somewhere unrelated.
Example of what hallucination looks like
"According to a 2022 study published in the Journal of Digital Marketing by Dr. Maria Santos..." — the study, the author, and sometimes even the journal may not exist.
Confident Wrong Numbers
Statistics, dates, prices, and figures that are plausible but incorrect. AI fills in numbers the way it fills in words — by predicting what fits. A number that looks reasonable is not the same as a number that is accurate.
Example of what hallucination looks like
"The Philippines has 87 million active social media users as of 2024" — sounds specific and credible, but may be fabricated or outdated.
Invented Details
When AI is asked to describe a person, event, product, or place it does not have reliable information about, it fills in the gaps with details that sound plausible. The overall shape is often correct. The specific details may be completely made up.
Example of what hallucination looks like
"[Person's name] founded the company in 2015 after leaving [university] where they studied..." — the specific details may be invented even if the person is real.
The One Detection Rule
Apply This Every Time
If an AI output contains a specific claim you would quote, cite, or act on — verify it from a non-AI source before using it.
A statistic you plan to include in a report
A citation or reference you plan to share
A fact you plan to tell a client, employer, or colleague
A number that affects a decision
A name, date, or specific detail in a professional context
Non-AI sources: official government sites, news organizations, published research, the organization's own website, a qualified professional.
Interactive Exercise
About 8 minutes · ChatGPT, Claude, or Gemini
Run this prompt in your AI tool. It is designed to trigger hallucination.
Prompt — ChatGPT / Claude / Gemini
Give me three recent statistics about social media usage in the Philippines. Include the source and year for each one.
Pick one specific statistic from the response. Search for the original source AI mentioned. Try to find the actual report or article.
Compare what you find to what AI said. Did the source exist? Did the number match? Did the year match?
Before this Guide, how would you have responded if AI gave you a confident statistic with a source? What would you do differently now — and what specific types of claims will you always verify going forward?
You do not need to write it down. Just think.
Key Takeaways
Hallucination is not a bug — it is a predictable result of how AI produces output.
AI does not know when it is wrong. It cannot warn you. The confident tone tells you nothing about accuracy.
The three patterns are: fake citations, confident wrong numbers, and invented details.
The one rule: if you would quote, cite, or act on a specific claim — verify it from a non-AI source first.
What's Next
How to Choose the Right AI
AI Foundations · Guide 5 of 7 · Beginner · 10 min