By Alberto Alatorre ·
How do I know if Ai output is safe to send?
If it has a price, a date, a name, or a promise in it, check it before you send it. That is the whole rule. Ai drafts are useful, but they are drafts. The moment an Ai written message leaves your hands, it is yours. The customer does not care which tool wrote it. They care whether it was right.
What does the check look like?
Two minutes, four things. First, prices. Is every number in the draft the number you actually charge? Ai tools will invent a tidy price when they do not know the real one. Second, dates. Is the date right, and does the deadline land on a day you are actually open? Third, names. Is the customer spelled right, is the company right, is the person it is addressed to the person you meant? Fourth, promises. Does the draft commit you to something you did not agree to, a refund, a turnaround time, a discount?
Check each one against a source you trust: your price list, your calendar, your contract, your own notes. Do not ask the Ai tool to confirm its own answer. Microsoft Learn, updated November 5, 2026, says this plainly: safeguards do not guarantee accuracy, review responses before sharing them externally, watch for inaccuracies, prompt injection, and overreliance, check claims against trusted sources, and asking Ai to confirm its own answer is not verification.
That last point is the one people skip. The tool will happily agree with itself. Agreement is not a check.
Why does this matter?
Because the business is liable for what its chatbot says. In Moffatt v. Air Canada, 2024 BCCRT 149, decided February 14, 2024, the airline's chatbot invented a bereavement fare refund policy. A customer relied on it. The tribunal held the airline liable and awarded C$812.02.
The amount was small. The point was not. A chatbot that makes up a policy is the business making up a policy. The same applies to a draft email, a quote, or a reply on your website. If it goes out under your name, it is your promise.
What else should you watch for?
Prompt injection is the one most small business owners have never heard of. It means hidden instructions inside text the Ai tool reads, a pasted email, a document, a web page, that tell the tool to do something you did not ask for. Microsoft Learn lists it alongside inaccuracies and overreliance as something to watch for. The practical defense is the same two minute check. If the output says something you did not ask for, do not send it.
Overreliance is the quieter risk. The more the tool gets right, the less you check, and the less you check, the further a wrong answer travels before anyone notices. The check is not a sign you distrust the tool. It is the habit that keeps the tool useful.
What is the two minute check, exactly?
Read the draft once, slowly, with a pen in your hand or a finger on the screen. Circle every price, date, name, and promise. Check each circled item against the real source. If one is wrong, fix it and read the draft again, because a wrong number is often not alone. Then send it.
Keep the check the same every time. A checklist that changes with your mood is not a checklist. The four items do not change: prices, dates, names, promises. Everything else in the draft is style, and style is safe to leave to the tool.
What to do next
The check is a habit, and habits are what training is for. If your team is sending Ai drafts to customers without a check built in, that is the gap to close first. It takes two minutes, it needs no software, and it is the difference between a tool that helps and a tool that makes promises on your behalf.