AI Hallucinations and Output Verification

This course contains the use of artificial intelligence.

An AI assistant can draft a polished report in an hour. It will cite sources, quote experts and give exact figures, and some of it may be wrong. AI tools state errors with the same confidence as facts, and fluent writing makes those errors easy to miss. This course gives you a repeatable method for checking AI-generated content before you use it, share it or base a decision on it.

You'll follow one realistic case from start to finish: the operations team at a fictional delivery company receives an AI-drafted vendor recommendation report the day before a leadership meeting. The report sounds certain. As you work through it, you'll uncover a misquoted analyst, a statistic its own source doesn't support, an outdated claim, a missing policy exception and a cost total that leaves out a fee, along with claims that check out and one that simply can't be confirmed yet.

What you'll learn:

• How to break an AI output into checkable claims and decide which to verify first.

• A five-step verification workflow: extract claims, assess risk, locate evidence, compare and check, and document a decision.

• How to check whether a cited source exists, is current and relevant, and actually supports the claim.

• How to compare AI summaries with their source documents and recompute numbers yourself.

• How to write prompts that set source boundaries and ask for traceable evidence and stated uncertainty.

• What AI-side checks, such as web search, self-checks and a second AI's agreement, can and can't prove.

• How to keep a verification record and decide whether to use, revise, withhold or escalate an output.

The course is presentation-based, with worked examples and pause-and-decide exercises throughout. No software, coding or technical background is needed. Downloadable resources include the case evidence pack, a verification record template, a source-bounded prompt template and a one-page output verification checklist.

Whether you draft with AI yourself or review work that others have drafted with it, you'll finish with a method you can use on any AI output you rely on.