Writing the Check-Before-You-Use Rule Into Your AI Policy
An email went to a customer with a made-up detail in it, and the person who sent it assumed the tool was right. The check-before-you-use rule is meant to stop that.
Why a rule is needed
AI tools can sound sure when they're wrong. The cyber security agency says "AI systems can also make up information, called hallucinations".1 Government staff guidance puts it another way:
"Be aware that generative AI can produce convincing but inaccurate content"2.
The agency also describes a 2025 case where "a lawyer used AI to prepare a court document" that cited false cases the lawyer hadn't checked. The court barred the lawyer from operating and owning a law practice.1 A small business is unlikely to face that exact outcome, but the pattern is the same: no check, real consequences.
Fix one: say who is responsible
The first line of the rule should settle blame. The staff guidance includes this principle: "Remain responsible and accountable for content you create, share or use."2 In plain words, "The AI wrote it" isn't an excuse. The person who sends it owns it.
Fix two: say what to check
"Review it" isn't enough. Staff need a short list. Government advice says to "verify AI outputs for inaccurate answers, biased language, unethical or irrelevant responses".1 The staff guidance adds "Check for fairness, accuracy and bias in generative AI outputs."2
Turn that into questions a person can answer:
- Are the facts, names, dates and figures right? Check each against a real source.
- Is anything in it made up, such as a quote, a link or a rule?
- Does it say something we wouldn't say, or promise something we can't do?
- Is the tone right for this customer?
- Did it use anything that belongs to someone else?
Fix three: add a second reader for risky items
Not every task carries the same risk. A reworded internal note needs a quick read. A contract clause, a refund decision or advice to a customer needs more.
The agency tells businesses to "have a human involved in the decision-making process especially in high-stakes or sensitive operations".1 On customer-facing chatbots it says that where legal, medical or financial advice is involved, businesses should "ensure a qualified person reviews chatbot responses before taking any action".1
Decide in advance which kinds of work need a second reader, and name them in the policy. Common candidates are anything legal, anything about money, anything about a person's health or employment, and anything published under the business name.
Health businesses should use AI for front-desk and admin work only, and never for anything clinical. The check rule doesn't change that.
Fix four: don't let AI make the final call
Staff guidance for government says that generative AI "must not make final decisions on government advice, services or outputs".2 That's written for agencies, but a small business can borrow the idea: a person decides, and AI helps with the draft. Say so in the policy, especially for hiring, pricing, refunds and complaints.
Sample wording
Adapt this to fit your business:
AI tools may help us draft and sort work. They don't replace a person's judgment. Before anything produced by an AI tool goes to a customer, supplier or the public, the person sending it must read all of it and check the facts. That person is responsible for what's sent. Anything about legal, financial, employment or health matters must also be read by the policy owner.
For more sentences, see the sample wording page. Customers may need to know when AI was involved, which is covered in telling customers about AI use.
Notes and sources
- Artificial intelligence for small business, Australian Cyber Security Centre (ASD)
- Staff guidance on public generative AI, Digital Transformation Agency (Australian Government)