When an AI Answer Needs a Second Person

An AI-written message goes out after one person's quick read. Two days later a customer holds you to what it said. Quick glances are fine for low-stakes text. For some jobs, they aren't enough.

Why one glance misses things

If you asked the question, you know what you hoped the answer would say. A second person comes to it fresh. They read what's on the page, not what you meant.

Australia's privacy regulator tells businesses to think about where a person fits into the process.¹ A second reader is one simple form of that.

Jobs that need a second person

Anything a customer may rely on

In a 2024 Canadian tribunal case, the airline argued its chatbot was a separate legal entity responsible for its own actions. The tribunal rejected that, and held that the airline was fully responsible for the information on its website, chatbot included.² The American Bar Association called it a helpful reminder that companies remain liable for the actions of their AI tools.²

That covers returns, warranties, delivery promises, prices, quotes and booking terms. Have someone who knows your real policy read each AI draft against it. This is general information, not legal advice. For your own case, ask a qualified adviser. Official guidance comes from bodies like the ACCC (Australia), the FTC (United States) and Citizens Advice (UK).

Anything filed with a court, regulator or authority

In Mata v. Avianca, attorneys were sanctioned for using fake case law citations generated by a chatbot, and the court issued a $5,000 fine.³ Tax returns, licence applications, grant forms and formal replies to a regulator all belong in this group. A second person should check every reference and figure against the official record.

Decisions about people

The OAIC treats AI-assisted decisions with a legal or similarly significant effect on a person's rights as likely high privacy risk, and asks for particular care.¹ That guidance is from Australia. Other countries have their own rules. UK readers can start with the ICO, and US readers with the FTC.

Hiring, firing, pay, discipline and credit decisions belong here. Don't let an AI draft stand in for the decision. Employment and privacy law differ by country and region, so check with the official body.

Contracts, financial figures and anything over a set value

Agree a line in your business. Anything above it gets a second reader. The line is your call, but write it down.

Anything about a named person

Don't send a summary of someone's personal information on a quick glance. The OAIC points out that AI-generated information about an identifiable person can still count as personal information, even when it's wrong.¹ Treat it with care.

How the second person should check

• Give them the AI draft and the original sources. Don't give them only your corrected version.
• Ask them to check facts, not style.
• Have them follow the step-by-step checking routine.
• Don't tell them what you expect to find.

If you work alone

Sleep on it where you can, then reread the next morning. Print the draft and read it aloud. Better still, ask a friend in your trade or your accountant to read just the risky parts.

A short record

Note what was checked, by whom, and when. A one-line entry in a shared document is enough. It also shows where mistakes tend to turn up, as the list of common slips describes.