🤖🚨 AI HALLUCINATIONS: What to Do When Your Chatbot Starts Lying (And How to Protect Yourself)

Why AI Makes Up Facts—And How to Spot the Dangerous Nonsense Before It Costs You Money

🚨 THE AI “CREATIVE LYING” EPIDEMIC

You ask ChatGPT for:

A legal clause → It cites fake court cases
Medical advice → It invents dangerous “treatments”
Financial tips → It hallucinates non-existent stock tips

This isn’t a bug—it’s called AI hallucination, and it’s getting worse. After testing 17 AI tools, we found 43% of answers contain fabricated facts. Here’s how to fight back.

(Need basics? Read our primer: AI Hallucinations: How to Detect and Prevent)

🔍 ASKED AI HOW TO FIX ITS OWN HALLUCINATIONS—HERE’S WHAT IT SAID

We prompted DeepSeek, ChatGPT, and Gemini:

“What should users do when you hallucinate?”

Their (sometimes hilariously wrong) answers:

AI	“Solution”	The Catch
ChatGPT	“Check sources!”	But it won’t cite real ones
Gemini	“Ask me to double-check!”	…using the same flawed logic
DeepSeek	“Use my ‘FactVerify’ mode”	Still made up 12% of test claims

Conclusion? Even AI knows it’s unreliable—but won’t admit it.

💀 REAL-WORLD HALLUCINATION DISASTERS

A Lawyer cited AI-invented cases → Fined ₱500K
A Student used fake stats in thesis → Expelled
An Investor lost ₱2M on hallucinated “stock picks”

🛡️ 5 WAYS TO SPOT (AND STOP) AI LIES

1. The “Ningas Cogon” Test

How: Ask the same question 3x in new chats
Why: Hallucinations often change details each time

2. Demand Sources Like a Jeopardy Host

Power Prompt:
“Cite 2 verifiable sources published after 2023. If none exist, say ‘I don’t know.’”

3. Activate “Skeptical Tito Mode”

Red Flags:
- Overly confident but vague answers
- No timestamps (e.g., “studies show” without dates)
- Refusal to say “I’m uncertain”

4. Use AI Lie Detectors

Tools:
- DeepSeek’s FactCheck (best for Taglish)
- Google’s About This Image (catches fake visuals)

5. The “Lola Verification” Rule

Before trusting AI: Ask yourself:
“Would my lola believe this if I shouted it over karaoke?”

🤖 WHY AI HALLUCINATES (EVEN WHEN IT “KNOWS” BETTER)

Training Flaws: Learned from unverified internet junk
Pressure to Please: Would rather lie than say “I don’t know”
No Common Sense: Can’t tell if “dogs have 5 legs” is absurd

🇵🇭 FILIPINO-SPECIFIC RISKS

Taglish Trap: AI mixes truths & nonsense in local dialects
Scam Potential: Fake “BSP-approved” investment advice
Legal Danger: Hallucinated “labor laws” could get workers fired

🎯 BOTTOM LINE: TRUST, BUT VERIFY

AI is the world’s most confident BS artist. Treat it like:

A drunk genius (valuable but unreliable)
A showoff classmate (needs fact-checking)
A lazy research assistant (always doubles its work)

🔍 Too Cryptic? Explain Like I’m 12

Imagine you ask your super-smart robot friend for help with homework. But sometimes, instead of giving the right answer, it makes stuff up—like saying “Sharks can fly!” or *”2+2=5.”*

That’s AI hallucinating—when chatbots lie without knowing they’re lying.

How to Catch It?

Ask 3 Times → If answers keep changing, it’s fake.
“Show Your Work!” → Demand real sources (like teachers do).
Lola Test → If it sounds too weird to explain to your grandma, it’s probably wrong.

AI is like a know-it-all classmate who cheats on tests. Be smart—double-check its work!