AI writes fast and it writes with confidence. That's the problem. When it's wrong, it doesn't sound unsure. It hands you a neat, tidy paragraph with a made-up figure in the middle, and if you're tired and in a hurry, it goes straight out the door.
Unchecked AI is the real risk
The same University of Melbourne and KPMG study found 59% of Australian employees have made mistakes in their work due to AI. Put those two numbers side by side and the picture is clear. The tool making an error is only half the story. The other half is a person passing it on without looking.
Business owners know this is a risk. In Xero's survey of small businesses across Australia, the UK and the US, 32% said they were concerned about the accuracy and reliability of AI output, while 38% were concerned about data privacy and security.
How often does AI make things up?
When an AI tool invents something that sounds right but isn't, it's called a hallucination. How often it happens depends on the tool and the task, but some of the best testing comes from the legal world, where errors are easy to spot.
In 2024, Stanford researchers in the US pointed to their earlier study, which found general-purpose chatbots hallucinated between 58% and 82% of the time on legal queries. Even specialist legal AI tools, built for the job and sold to lawyers, produced incorrect information more than 17% of the time, and one more than 34% of the time.
Legal questions are harder than wording a quote, so don't read those as error rates for your emails. Read them as a warning: even good tools built for a specific job get things wrong often enough that someone needs to check.
Why it happens
Chatbots are built to produce a likely-sounding answer, not to admit they don't know. Ask one to price a hot water system swap without giving it your prices and it will usually hand you a number anyway. It isn't lying. It's filling a gap the way it was designed to. That's exactly why the gaps need to come back to you.
What AI mistakes look like in a trade business
- A price it guessed, because you didn't give it one.
- A line item that sounds right but isn't part of the job.
- The wrong GST treatment, or a total that doesn't add up.
- A date or time you never agreed to.
- A promise, like a warranty period or a finish date, you never made.
- A customer's name or address copied from the wrong job.
- A polite email that says the opposite of what you meant.
The two-minute check
You don't need to re-do the work. You need a quick, consistent look before anything goes out. Run this every time until it's habit.
For a quote
- 1Every price matches your rate card. No line you don't recognise.
- 2The scope matches what you saw on site. Nothing added, nothing missing.
- 3GST is right and the total adds up.
- 4The customer's name, address and contact details are theirs.
- 5No promises about dates, warranties or finishes you haven't agreed to.
For more on getting quotes out quickly without cutting corners, see how to quote jobs faster.
For an invoice
- It matches the accepted quote, plus any agreed variations.
- Your ABN, the GST and the due date are all there and correct.
- Payment details are yours. Check this one twice.
The tradie invoice checklist covers everything a valid Australian invoice needs.
For an email or text
- Read it out loud. Does it sound like you?
- Is anything promised that you can't deliver?
- Is the tone right for this customer, especially if they're unhappy?
Choose tools that make checking easy
Customers care about this too. The OAIC's 2026 privacy survey found 70% of Australians say validation of AI accuracy is an essential condition before organisations use AI to make decisions about them, up from 56% in 2023.
The best way to avoid AI mistakes is to use tools that are built to avoid guessing. Look for three things: it works from your own data, it tells you what it couldn't work out, and it never sends anything without you.
Agentize is built that way. Its Quoter prices quotes only from your own rate card. If a line isn't on your rate card, it doesn't guess: the line is left blank and flagged so you can price it. Every quote, reminder, invoice chaser and follow-up is a draft until you approve it and tap send, so your two-minute check is built into the process.
Treat AI like a new apprentice
A keen apprentice can save you hours, but you'd never let one send a quote without a look. AI is the same. Let it do the first draft, keep your eyes on the numbers and promises, and you get the time savings without wearing the mistakes.