When to trust the AI, when to check the source

We were setting up a Shopify store. The question came up: when a customer pays through Razorpay, how long before the money reaches the merchant?

I asked Shopify Sidekick — Shopify’s own AI support tool. The answer: T+2 to T+7 days. Plausible. Reasonable. And vague enough to be useless.

The problem with “T+2 to T+7” is that it’s not an answer. It’s a hedge. Somewhere between two days and a week is not a settlement policy. It’s a range wide enough to cover almost any outcome.

So I went to Razorpay’s documentation directly. Two minutes of reading: domestic payments settle in T+2 working days. International: T+7. Not a range — two distinct answers for two distinct scenarios.

The habit worth building

Treat the first AI answer as a starting point, not a conclusion. Especially when the answer involves money, timelines, or anything a business will depend on.

The pattern that emerged from this session:

  1. Ask the platform AI (Shopify Sidekick, in this case) — fast, contextual, usually directionally right
  2. Go to the primary source (Razorpay’s official documentation) — authoritative, specific

Two steps. Five minutes. You end up with an answer you can put in a document and share.

The operator who skips step two is operating on a hedge. The one who runs both has a fact.

The thing about AI answers

It’s not that AI answers are wrong. But they could be. That’s the thing. Shopify Sidekick gave us T+2 to T+7 — which happened to be in the right ballpark. But we only know that because we checked. Had it said T+3 to T+10, we might have accepted that too. AI answers are confident by nature. They don’t come with uncertainty attached. The only way to know if you’re holding a fact or a hallucination is to go to the source.

Precision requires going to the source. AI makes going to the source faster than it’s ever been — finding the right page, reading the right section, cross-referencing in minutes rather than hours. The habit of verification isn’t a tax on working with AI. It’s what makes AI-assisted research actually reliable.

Trust the AI to find the right door. Check the source to confirm what’s on the other side.