You can’t just trust your AI. A huge percentage of what it tells you, even when it sounds completely certain, is made up. I was blown away when I saw an analysis from this industry pro who spent 30 days and over 200 prompts to figure out what actually stops AI hallucinations.
The expert’s deep dive into ChatGPT, Claude, and Gemini revealed that a staggering 34 percent of factual queries contained false information! The good news is that the creator found that specific prompting techniques can slash these errors by over 70%. It’s not about writing longer prompts; it’s about writing smarter ones.
Here are the most important insights from the author’s incredible testing.
📌 The Single Most Powerful Technique The author found that one simple instruction reduced hallucinations by 52%. Adding this to your prompt forces the AI to be honest about its confidence. I’m definitely adding this to my workflow.
Here’s the exact phrase the post’s author recommends adding to your factual queries:
“If you’re not completely certain about something, say ‘I’m uncertain about this’ before that claim. Be honest about your confidence levels.”
✅ Stack Your Safeguards for Max Power The person who shared it discovered that combining techniques is even better. For serious factual research, the original poster recommends combining four specific instructions to achieve a massive 71% reduction in false claims. This includes demanding source attribution and setting temporal constraints on the AI’s knowledge.
💡 Stop Wasting Your Time on Useless Instructions This might be my favorite part of the analysis. The LinkedIn user tested all the common advice and found that some popular ‘tips’ have zero impact. Stop telling your AI to “be accurate” or “think carefully”, as the testing showed it does absolutely nothing to improve reliability.
The big takeaway here is that while these techniques make AI far more trustworthy, they don’t make it perfect. The expert was clear: you still need to verify critical information.
This is just a glimpse of the findings. For the full ranked list of all 12 techniques and the different combinations for various tasks, you have to check out the original post.
How to Stop AI from Making Up Facts – 12 Tested Techniques That Prevent ChatGPT and Claude Hallucinations (2025 Guide)
byu/Over_Ask_7684 in