Why cited-to-source beats confident-and-wrong
A general chatbot will answer any legal question with total confidence, including the ones it's making up. The only durable defense is a citation you can open.
The single most dangerous property of a general-purpose language model, for a lawyer, is that its confidence is uncorrelated with its accuracy. It will cite a case that does not exist in the same tone it uses to cite one that does. There is no tell.
The sanction is not hypothetical
The profession already has its cautionary tale. Attorneys have been sanctioned for filing briefs with citations a chatbot invented: fake cases, fake quotes, real consequences. The lesson is not "AI is bad." The lesson is that unverified output is bad, and confidence is not verification.
What "cited to source" actually means
A verified answer isn't one that sounds right. It's one where every assertion is bound to a specific, openable source (a page, a paragraph, a line) that you can read and confirm before you rely on it. If the source can't be produced, the claim doesn't get made.
This is the difference between an answer you have to trust and an answer you can check. For legal work, only the second one is usable.
Design for the absence, not just the presence
The subtle part is what happens when there is no support. A verification-first system has to be willing to say "I couldn't find a source for this," and to say it loudly, rather than smoothing over the gap with something plausible. Surfacing the absence of a citation is as important as producing one.