AI tools predict plausible text. Most of the time plausible and correct are the same thing. Sometimes they aren't: a statistic that doesn't exist, a law that was amended, a sum that is slightly off, a phrase that lands badly. Your name goes on whatever you send, so the last check is always yours.
The five-point check
1. Facts
Underline every claim that could be wrong: names, dates, rules, product features, "studies show". If you can't confirm it from a source you trust, remove it or check it. Be especially careful with laws and regulations, which change and vary by country.
2. Numbers
Recalculate totals, percentages and growth rates yourself, or in a spreadsheet. AI tools are much better at explaining a calculation than at doing arithmetic reliably in their heads.
3. Sources
If the answer cites a report, article or link, open it. Make sure it exists and says what the answer claims. A citation that looks real is not proof.
4. Tone and fit
Read it aloud as the recipient. Does it sound like you? Is it the right length? Would anything feel cold, over-familiar or presumptuous to this particular person?
5. Fairness and bias
For anything about people (job ads, reviews, feedback, selection), look for gendered or age-coded words, assumptions about background, and criteria that don't relate to the work. Ask: would I write this the same way about a different person?
Make the AI check itself first
Before your own review, ask the tool to critique its answer. It won't catch everything, but it often flags its own weak spots.
Review your previous answer as a sceptical expert would. 1. List every factual claim, number or rule you stated, and mark each as "confident", "unsure" or "should be verified". 2. Point out any assumptions you made about my situation. 3. Flag any wording that could read as biased, unfair or too absolute. Don't rewrite the answer yet. Just give me the list.
When to take extra care
- It goes outside your team: to a customer, candidate, regulator or leadership
- It contains numbers someone will make a decision on
- It refers to a law, policy or contract
- It describes or evaluates a person
- It will be published or kept on record
The more it matters, the more of the work you check yourself. AI can draft a first version in seconds; that time saved is what pays for a careful review.
Build the habit with real tasks
Every lab in AI for HR Professionals ends with a review step, so checking output becomes second nature.