I Trust the AI That Shows Its Sources, and You Should Too

I stopped asking open chatbots for facts I'd be embarrassed to get wrong. A general model will hand you a confident answer with no way to check it, and when the topic is something I'd stake my name on, that's a problem dressed up as convenience.

Perplexity ships every claim with a clickable citation, so I can see where it came from. Elicit and Scite go further: they show how a paper is cited, supporting or contradicting the point, which is the only kind of answer I trust for anything approaching a decision. The source is the product, not a footnote.

The habit that changed my output was leading with one question: where did this come from? If a tool can't answer that, it's entertainment, not research, and I treat its output accordingly. I'll brainstorm with a black box all day; I won't cite one.

None of this means the source-tied tools are perfect - citations can be weak, and a paper that's cited a lot isn't automatically right. But at least I can follow the thread, and that's the difference between an answer I can defend and one I have to hope is correct.