Skip to content
Building an AI-Powered Product or ServiceLesson 8 of 13

8 · Quality and trust: handling wrong answers, disclosure, and support

Your AI will be wrong sometimes — that's a fact about the technology, not a bug you can fully fix. The founders who win design for that honestly. How you handle a wrong answer, whether you disclose AI, and how a customer gets help is your product's trustworthiness.

Handling wrong answers — design the mistake before it happens:

  • Catch it before the customer does. This is the human-in-the-loop (Lesson 4) for anything that matters — verify claims, prices, and policies before they ship.
  • Make uncertainty visible. A feature that can say "I'm not sure — here's how to check / reach a person" is more trustworthy than one that's confidently wrong. PAIR calls this designing for graceful failure and giving users control (PAIR, 2021).
  • Make it easy to fix and report. Let customers correct or flag a bad answer. Every report is free quality data and a trust-builder.

Disclosure — be honest that it's AI:

  • Tell customers when they're interacting with AI, and give a clear path to a human. Pretending a bot is a person breaks trust the instant someone notices, and transparency is core to trustworthy AI (NIST, 2023).
  • Don't overstate what the AI does. The FTC is explicit: AI performance claims must be truthful and substantiated — you can't exaggerate your AI's abilities, and "the AI said so" is no defense (FTC, 2023). This is a legal line, not just an ethical one.

Support — the human safety net:

  • Always offer a way to reach a person for problems the AI can't solve. An AI-only wall with no human exit is a trust-killer.
  • Use AI to help support, not to hide from customers. Drafting replies for you to approve = good. A bot that stonewalls upset customers = brand damage.

Trust DNA: customers forgive an honest small business that's clearly trying. They do not forgive being deceived — by a hidden bot, an overstated claim, or a wrong answer no one would fix. Handle the mistake well and you can actually gain trust.

Check yourself. Name one way to handle an AI wrong answer that builds trust, and one disclosure rule you must follow — and say why the disclosure rule is also a legal matter.

Sources

Sign in to track your progress