Truthfulness is a real problem in AI. Not "sometimes wrong." Structurally wrong — with a confident voice, a tidy paragraph structure, and sources it invented. The model doesn't know it's wrong. There's no blush. There's no "actually, wait." There's just the output, presented as truth.
If you build a business decision on that output without checking it, that's on you. Here's how to not get burned.
Cross-reference across multiple systems. ChatGPT, Gemini, Grok, Claude, Perplexity — run the same prompt. Disagreement = contested ground = go find human sources.
Use capability prompts, not command prompts. "Can you find evidence for X?" vs "Tell me about X." Capability prompts reduce hallucinations because they frame the task as finding something that may or may not exist.
Never ask AI to recommend AI tools. The circularity is poisonous. Go to Reddit. Go to actual humans who've used the thing.
AI should accelerate your thought process, not replace it. The moment you stop having the domain expertise to catch the errors, you've stopped being the boss.
Read the full piece on Medium →