AI hallucinations: what they are and how to keep them away from customers

An AI tool can give an answer that sounds certain and is simply wrong. This is called a hallucination, and every business using AI should plan for it.

What a hallucination is

Language models generate likely-sounding text. They do not look facts up unless they are connected to a source. When the model lacks information or the question is ambiguous, it can still produce a fluent answer that is invented or outdated.

Where it hurts a small business

• A chatbot quoting a price or policy that does not exist.
• A drafted email stating a wrong date or promise.
• A public assistant describing your business incorrectly, for example the wrong service area.
• A summary that quietly changes a number.

Safeguards that work

• Ground answers in your own information. Limit an assistant to approved content instead of letting it improvise.
• Limit what it may say. It should not quote prices, make promises or give legal or medical advice unless that content is approved.
• Hand off to a person when it is unsure or the topic is sensitive.
• Review before anything goes live, and sample live conversations afterwards.
• Log and correct. When it errs, fix the source information.

When others get it wrong about you

If an outside assistant misdescribes your business, the fix is usually at the source: clear, consistent facts on your site, structured data and listings. See how assistants choose sources.

The takeaway

Treat AI output as a draft from a fast but unreliable assistant. Verify before you rely on it. See why every automation needs a human in the loop.

Common questions

Keep reading

• AI for business

  Human in the Loop: Why Every AI Automation Needs Review

  2 min read
• AI for business

  Good vs Bad AI Practices for Small Business

  2 min read
• AEO

  How AI Assistants Find and Choose Sources

  2 min read