A woman asked ChatGPT how to dye her hair at home. The AI laid out a confident process: apply color to the ends first, wait 20 minutes, then do the roots.
The instructions in the actual box said something different: mix the formula and apply everything at once. That's how this specific product was designed to work.
She almost followed ChatGPT anyway.
The "AI Said So" Reflex
This scenario is becoming remarkably common. People are defaulting to AI-generated instructions over manufacturer directions, professional advice, and their own lived experience. The pattern is consistent: ask ChatGPT, get a confident answer, treat it as the final word.
The issue isn't that ChatGPT is always wrong. It often gives reasonable general advice. But "reasonable general advice" and "correct for your specific situation" are fundamentally different things. ChatGPT didn't know which brand of hair dye was being used. It didn't know the chemical formula. It pattern-matched against common hair dyeing processes from its training data and presented the result with the unearned confidence of an encyclopedia.
That confidence is precisely the problem. ChatGPT doesn't hedge or qualify. It doesn't say "I'm guessing based on general principles and I haven't read your product's instructions." It presents answers as though it has access to specific information it simply doesn't have.
Research backs this up. A 2025 study from the University of Melbourne found that participants were more likely to follow AI-generated health advice than advice from a qualified nurse, even when told the AI had no medical training. How information is packaged matters more than where it actually comes from.
Hair Dye Is Low Stakes. Medication Isn't.
The hair dye case is mostly harmless. A slightly uneven color result isn't the end of the world. The real danger is this same reflexive trust applied to medication dosages, electrical work, food safety, legal questions, or financial decisions.
When ChatGPT confidently tells you the wrong drug interaction information or gives incorrect guidance about load-bearing walls in your home, the consequences extend well beyond a bad hair day. And the AI will be just as confident in its wrong answer about acetaminophen dosing as it was about hair dye timing.
A Practical Trust Framework
For daily AI users, the boundary isn't complicated:
Trust AI for: Brainstorming, drafting, summarizing, explaining concepts, code generation, answering well-documented questions where broad consensus exists.
Don't trust AI for: Anything specific to YOUR situation that it can't actually see. Your specific product's chemical formula. Your health condition. Your electrical panel. Your tax circumstances.
When AI output conflicts with the instructions that came with whatever you're doing, follow the instructions. They were written by people who designed the product, tested it, and are legally liable for accuracy. ChatGPT carries none of that accountability.
The most productive AI users aren't the ones who trust it the most. They're the ones who've learned exactly where to draw the line.