The 5-Check Review: How to Use AI at Work Without Getting Burned

By Camderson Academy

AI tools can draft an email, summarize a report or outline a plan in seconds. They can also state something false with complete confidence. The fix is not to avoid AI. It is to build a short, repeatable review habit so nothing leaves your hands until a person has checked it. This guide gives you that habit: five checks, in order, that take a few minutes and protect your reputation.

Why a review habit matters

The companies that build these tools are open about their limits. OpenAI's help center says ChatGPT "can produce incorrect or misleading outputs" and lists fabricated quotes, studies and citations among the problems to watch for. Its advice is simple: use ChatGPT as a first draft, not a final source, and check important information.

The U.S. National Institute of Standards and Technology (NIST) names the same issue in its Generative AI Profile (NIST AI 600-1, 2024). It calls it confabulation: confidently stated but erroneous or false content, known informally as hallucinations. The same document lists data privacy as a separate risk, which matters every time you paste work material into a tool.

None of this means AI is not useful. It means the person who sends the work is still responsible for it. A review habit is how you keep the speed and drop the risk.

Treat every AI output as a confident first draft from a new colleague: often helpful, sometimes wrong, and never sent without your review.

The 5-Check Review

Run these five checks on anything AI helped you produce before it goes to a customer, your manager or the wider team. The order matters. Start with the checks that can sink you, and finish with polish.

1. Facts: is every claim true?

Highlight every name, number, date, quote and statement of fact in the draft. Confirm each one against a source you trust, such as your own records, the original document or an official website. If you cannot confirm a claim, cut it or rewrite it as an open question. Numbers deserve extra care, because a wrong figure looks exactly like a right one.

2. Sources: do the references exist and say what is claimed?

If the draft cites a study, article, policy or law, open it. Check that it exists, that the link works and that it actually supports the sentence it is attached to. A citation that looks real is not proof. If you would be embarrassed to have someone click the link, do not send it.

3. Data: did anything sensitive go in, or come out?

Ask two questions. First, did you paste anything into the tool that your organization's policy does not allow, such as customer details, employee records, health information or confidential numbers? Second, does the output repeat sensitive details that the audience should not see? If your workplace has an approved AI tool and a policy, follow it. If it has neither, ask before you use work data at all.

4. Fit: does it answer the real request?

Reread the original ask. Does the draft answer that question, for that audience, in the right length and tone? AI often produces something polished that misses the point, adds generic filler or uses a tone that does not sound like you or your company. Cut what does not serve the reader.

5. Ownership: would you sign your name to it?

Read the final version as if you wrote every word, because once you send it, you did. Make any judgment calls yourself. If your organization asks you to disclose AI assistance, do so. If the content affects money, people's jobs, legal matters or safety, have a qualified person review it as well.

A worked example

Imagine you manage a small customer service team. Your company has just changed its refund policy, and you ask an AI assistant to draft a short team update from the two-page policy document.

The draft comes back in seconds and reads well. Here is what the 5-Check Review catches:

• Facts: the draft says customers have 45 days to request a refund. The policy says 30. The tool filled a gap with a plausible number. You fix it.
• Sources: the draft mentions "industry best practice guidelines" that the policy never cites. You delete the line.
• Data: you notice you pasted an email thread that included a customer's name and order number. Next time you will paste only the policy text. The output itself contains no customer details, so it is safe to send.
• Fit: the draft is 400 words. Your team needs the three changes and the start date. You cut it to 120 words with a short list.
• Ownership: you add one line on who to ask with questions, read it once more and send it under your name.

Total time: about five minutes. Without the review, your team would have promised customers 15 extra days.

Template: the 5-Check Review card

Copy this into your notes, or print it and keep it by your desk.

Make it a habit, not a hurdle

The review gets faster with practice. A few tips help:

• Prompt for checkability. Ask the tool to list the facts and assumptions it used in a separate section. That gives you a ready-made list for check one.
• Give it your source. When you paste the actual document and ask the tool to work only from it, you have less to verify.
• Match effort to stakes. A brainstorm for yourself needs a light touch. Anything external, financial or about people gets all five checks.

The takeaway

AI is a fast first-draft partner, not a final authority. Five checks, Facts, Sources, Data, Fit and Ownership, turn a risky shortcut into a dependable way of working. Run them every time, and you get the speed without the surprises.

Sources

• OpenAI Help Center, "Does ChatGPT tell the truth?"
• National Institute of Standards and Technology, Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile (NIST AI 600-1), July 2024

This article shares general good practice for using AI tools at work. Always follow your organization's AI and data policies.