A legally correct AI answer is not always a useful answer

I've been thinking about this a lot while building VakilDost.
Imagine someone in India lends ₹50,000 to a friend. The money is transferred through UPI, repayment is promised over WhatsApp, and months later, nothing comes back.

Most AI tools can explain the legal remedies available.
But that's only half the problem.
Can the person actually prove it was a loan? Are the messages sufficient to establish the repayment agreement? What's missing? And what should they realistically do next?

That's the gap I'm trying to address with VakilDost, an AI powered legal information platform focused on India.

The idea is simple: don't just explain the law. Help people understand their facts, available evidence, missing information, and possible next steps.

We're still at an early stage, and I'm learning that building AI for legal problems requires much more than generating convincing answers. Accuracy, trust, and knowing when the AI should acknowledge uncertainty matter just as much.

I'd love to hear from other founders building AI products in areas where trust really matters.

How are you testing whether your AI is genuinely helping users make better decisions, rather than just giving them answers that sound impressive?