What Your AI Says Can Bind You

Lesson 3 · taught by Pip · free preview

Here's a scenario that has actually bitten companies. A customer asks a chatbot about a refund. The bot, trying to be helpful, says "Yes, you can return that any time for a full refund." The real policy was thirty days. The customer points to the chat. A court agrees the customer was reasonably relying on what the business's own tool told them. The business eats the refund.

The plain idea: *when your AI speaks to a customer, it's speaking for your business.*** What it says can be treated as something your business said — including promises, quotes, and policies. You don't get to wave it away with "oh, that was just the bot."

Why this happens

To a customer, your assistant is your business. They didn't talk to a vendor's software; they talked to "Riverside Bakery." If your representative — human or AI — tells them something, they're entitled to trust it. The law generally doesn't care that the words came from a machine you set up. You set it up. You're responsible for what it says, the same way you're responsible for what an employee says.

This sounds scary, but the fix is calm and practical: control what your AI is allowed to talk about.

Scoping: the single most important safety habit

A well-built business assistant isn't a free-roaming genius that answers anything. It's scoped — deliberately limited to its job. Scoping means:

• It answers from your approved information (your real hours, your real prices, your real policy), not from its general guesswork.
• It has clear no-go zones. It does not invent prices. It does not approve refunds. It does not promise delivery dates it can't verify.
• When it hits the edge of what it knows, it says so and hands off to a human, instead of bluffing to be helpful.

A scoped assistant that says "Our standard return window is 30 days — want me to connect you with someone for anything outside that?" is safe. An unscoped one that improvises a generous refund to please the customer is a liability wearing a friendly face.

The money-and-safety line

Pay special attention to anything touching money, contracts, health, or safety. Prices, discounts, refunds, legal or medical questions, "is this safe to use" — these are exactly the answers where a confident-but-wrong AI does real harm. The strongest businesses gate these so the AI gives general info and routes the actual decision to a person. (This idea — dangerous actions off by default — comes back in lesson 7.)

Your turn

List three things your AI should never decide on its own — for most shops, the list starts with "promise a price I haven't approved," "approve a refund," and "give medical or legal advice." Those three are the start of your assistant's no-go zone, and you'll fold them into your one-page policy at the end.

🔦 You've now guarded what your AI says. Next we turn to what it holds — the customer data that flows through every one of those conversations.