You asked the AI a simple question. It gave you back a confident, well-written, professional-looking answer. There was just one problem: it was wrong. If that has happened to you, you are not looking at a rare glitch. You are looking at the single biggest risk of using AI in a small business. The issue is not that AI gets things wrong — every tool gets things wrong. The issue is that AI gets things wrong in a way that looks completely finished. This is a practical guide to what you should actually do when it happens, so a bad answer costs you five minutes instead of a customer.
A bad answer doesn’t look like a bad answer
When a person is unsure, you can usually tell. They hedge, they pause, they say “I think”. AI does none of that. It delivers a wrong figure inside a tidy table and an invented statistic alongside a plausible-sounding source, in the same confident tone it uses when it is completely correct. The industry term for this is a hallucination, and it is baked into how the technology works.
The rate is never zero. Independent testing through 2025 found that even the most reliable large language models still state false information around 0.7% of the time, while weaker or older models get it wrong as often as 30% of the time. You cannot tell which answer fell into that percentage just by reading it, because a fluent wrong answer and a fluent right answer look identical on the page.
If it can happen to Deloitte, it can happen to you
In 2025, one of the most high-profile examples came from the top end of town. Deloitte delivered a report worth roughly $290,000 to the Australian federal government’s Department of Employment and Workplace Relations. Reviewers later found it contained fabricated citations, references to research that did not exist, and a made-up quote — after generative AI had been used to help fill gaps in the analysis. Deloitte agreed to refund part of the fee.
Sit with that for a moment. A global professional services firm, with review processes most small businesses could only dream of, still shipped AI-invented content into a government report. If it can slip past them, it can absolutely slip past a busy owner checking a draft between two other jobs. The lesson is not “avoid AI”. The lesson is that a verification habit is not optional — it is part of the cost of using the tool at all.
The three checks that catch most bad answers
Here is something you can put in place today, before you use AI for anything that matters. Whenever an AI answer contains one of these three things, treat it as a red flag that means “stop and verify” rather than “copy and paste”.
First, check that the source exists. If AI gives you a statistic, a legal reference, a citation, a web link or a direct quote, that is your cue to confirm it independently before it leaves your business. Ask the AI “where exactly did that come from?” and then go and look for that source yourself. If you cannot find it in two minutes, assume it was invented.
Second, check it against what you already know. You are the expert on your business, your industry and your customers. If an answer contradicts your own experience or common sense, trust your gut and dig in. AI is very good at sounding authoritative on subjects it has gotten wrong.
Third, check the maths. AI can write a beautiful paragraph and still add up a column of numbers incorrectly. Any figure you will act on — a quote, a margin, a tax estimate, a forecast — gets checked in a calculator or a spreadsheet, every time. These three checks take a couple of minutes and catch the overwhelming majority of errors that would otherwise reach a client.
Why this keeps happening (and why it isn’t your fault)
AI use has spread through Australian small business faster than the habits needed to use it safely. Deloitte Access Economics reported in late 2025 that around two-thirds of small and medium businesses are now using AI, and the National AI Centre put SME adoption at roughly 43% over the December-to-February 2026 quarter. The tools are everywhere. The verification discipline is not.
The consequences are already showing up. In one 2025 cross-industry survey, close to one-third of organisations reported a negative consequence tied specifically to AI inaccuracy. The core reason is simple: AI does not know when it does not know. It is built to produce the most plausible-sounding response, not to tell you when it is unsure. Expecting it to flag its own mistakes is like expecting a calculator to warn you that you typed the wrong number.
Build the check into the process, not the end
The businesses that use AI well do not check less — they check earlier, and they set the task up so there is less to get wrong. Three habits make the biggest difference. Give AI the raw material instead of asking it to recall facts: paste in the actual invoice, the real policy, the genuine numbers, so it is working from your source rather than its memory. Ask it to show its working, so you can see how it reached a figure instead of just trusting the total. And explicitly tell it to flag anything it is unsure about — it will not do this unless you ask.
Do that, and AI stops being a slot machine you pull and hope on, and becomes a fast, capable assistant whose work you can actually stand behind. A bad answer becomes a two-minute catch, not a reputation problem.
If you are not sure where AI might be quietly creating risk in your business — or, just as importantly, where it could be saving you real hours each week — that is exactly what our free AI Opportunity Audit is built to surface. It walks through how your business actually works and shows you where AI is worth trusting, where it needs a checkpoint, and where the genuine time savings are hiding. Take the free audit here and start using AI with confidence instead of crossed fingers.