Quick answer
AI tools go wrong in three main ways. Hallucination is confidently stating things that aren't true. Sycophancy is agreeing with you instead of challenging you. Bias is defaulting to common assumptions that don't fit your situation. You reduce all three by asking for sources and confidence levels, explicitly inviting disagreement, and giving specific context about your market.
Key takeaways
- Hallucination: ask for sources and a confidence level, and verify numbers.
- Sycophancy: tell the AI to argue against you before it agrees.
- Bias: give your specific market, customer and constraints.
- Never publish an AI-generated statistic you haven't checked.
Unlock the full guide (free)
Enter your email to read the step-by-step guide and copy every prompt. One sign-up unlocks the whole library.
- The full walkthrough and copy-paste prompts
- Every guide in the library
- The free 7-day AI Quick Start, one short email a day
Already signed up on this device? The guide opens automatically. No spam; unsubscribe anytime. On Instagram? Comment TRUTH on any @mahzyn.ai post.
AI is an excellent first-draft machine and a risky final-answer machine. Knowing how it fails is what separates people who get value from it from people who get burned.
1. Hallucination: confident and wrong
Ask for a statistic and you'll often get a precise-looking number with no real source. The model is predicting what a plausible answer looks like, not looking it up.
For every factual claim, statistic or quote in your answer: 1. Mark your confidence: High, Medium or Low. 2. Say where it would come from (study, organisation, year) or write "no reliable source". 3. Flag anything I should verify before using it publicly. If you don't know, say so. "I don't know" is a better answer than a guess.
Rule of thumb: if a number will appear in a proposal, post or report, open the original source yourself.
2. Sycophancy: telling you what you want to hear
Share your business idea and AI will often call it "a great idea". That's not analysis. It's politeness.
Before you give me any encouragement, argue against this plan as a sceptical investor would. My plan: [describe it] Give me: the 5 strongest reasons it could fail, the assumption I'm most likely wrong about, and what evidence would change your mind. Then, and only then, tell me what's genuinely strong about it.
3. Bias: the average answer for an average business
Ask "how should I price my service?" and you'll get the generic internet answer. Your market, city and customers might be completely different.
Answer this specifically for my situation, not the typical case. Question: [your question] My context: [industry, city or region, customer type, price point, team size, constraints] Before answering, list any assumptions you're making that might not fit my context, and ask me about the riskiest one.
Frequently asked questions
Why does AI make things up?
Language models generate text based on patterns, not a database of verified facts. Without a source to check, they can produce plausible but false details.
Do newer models still hallucinate?
Less often, and tools with web search or document access reduce it further. It still happens, so verify anything important.
Is it safe to use AI for legal or financial decisions?
Use it to prepare questions and understand concepts, then confirm with a qualified professional.
What's the fastest way to catch errors?
Ask the AI to review its own answer for mistakes, then spot-check the highest-stakes claims yourself.