Say you've just been sent your building contract. Plans attached, a folder's worth of specifications you've mostly skimmed. The obvious move for a lot of people now is to load the whole thing into ChatGPT and ask it to find the holes.
Is this risky?
What am I missing?
It feels like due diligence. It feels like something a careful person does before signing anything that big.
We don't actually know how many people are doing this. Nobody's told us, and we haven't gone looking. But enough of our clients are open about using AI day to day that we've started assuming at least some of them are pointing it at their contract before they've raised a concern with us directly. Which would be fine, except for what tends to happen next.
Ask a general AI model to find holes in something, and it will usually find some. Not necessarily because your contract has holes in it, but because that's the shape of the question you asked. It's built to hand back an answer that sounds complete and confident, and "nothing to worry about here" rarely reads as either. So it reaches into everything it's ever absorbed from the internet, most of it with nothing to do with Adelaide, your project, or the two people who actually wrote the document in front of you.
AI doesn't know whether there's a real problem in your contract. It knows how to sound like there might be one.
And then you're back in front of your builder with a list of questions three pages long, half of them about clauses that are completely standard, and the conversation starts from suspicion instead of "can you talk me through this." That's the opposite of what anyone actually wants from a tool that's meant to make things easier.
So what actually goes wrong when you do that?
Picture you're on site, and you catch half a conversation. The builder's talking to the roofer about getting the right fall on a box gutter. You didn't quite follow it, so later you ask ChatGPT what the correct fall on a box gutter is. Reasonable enough question.
Except it doesn't know if you're in South Australia or Victoria, and the two don't run under quite the same requirements. It doesn't know your climate zone, whether that's actually a box gutter or a boundary gutter (different thing, different rules), or anything about the roof it's actually sitting above. So it makes some assumptions anyway, and hands you back an answer stated with total confidence, because that's what these tools do whether or not they actually know. The National Construction Code alone runs across three separate volumes covering commercial buildings, domestic buildings, and plumbing and drainage for the whole country. A generic question has no way of knowing which parts of that apply to your job.
You compare that answer to what the builder actually said, and if it doesn't quite match, the instinct is to assume the builder's wrong. Might be. Or it might be that you asked a tool with no idea what climate zone you're in to answer a question that depends entirely on knowing your climate zone.
So is there actually a good way to use it?
There's a Google product called NotebookLM that works differently on this specific point. Instead of drawing on the whole internet, it only answers from documents you actually load into it yourself. So if you put in your contract, your plans, and the emails and messages you've exchanged with your builder, then ask it a question, it can only answer from that material. Nothing else to pull from, so nothing else to get wrong.
That makes it a genuinely useful place to ask the kind of questions that used to mean digging through a long contract by hand. Question like:
- What's a provisional sum, and how do they work?
- What margin does a builder typically charge on a variation?
- What's the actual process for raising a variation?
- Can I reject a variation?
Load your documents in, ask, and it'll point you to the exact section it pulled the answer from, so you can go and check it yourself rather than taking its word for it.
It's just as useful the other way round: the small, specific things you're trying to confirm rather than understand. Did the stone benchtop you talked about actually make it into the specifications? The tiles you picked? The tree out the back that was supposed to come out? Ask it, and it searches everything you've loaded and tells you, rather than you scrolling through a PDF at 11 at night trying to remember which document that was in.
What do you do with the answer?
The part that actually matters most isn't the tool. It's what you do with what it tells you.
If it turns something up that looks missing, that's not proof your builder's trying to pull one over on you. It's usually something far less dramatic: an honest oversight, something that was agreed verbally and never made it onto the page, or something left out for a reason nobody's explained yet. Either way, the next step is the same: go back to your builder and ask. "We talked about this benchtop a few weeks ago, I can't find it in the contract, can you talk me through it?" That's a completely different conversation to walking in with an AI-generated list of everything that might be wrong with your build.
Being upfront that you used AI to get there is worth doing too. It's not something to hide. Most builders are using it themselves, for much the same reasons: faster admin, fewer errors, more time to actually talk to clients instead of drafting emails.
Doesn’t AI always give a neutral response?
There's a reason people lean on it the way they do. AI feels like it has no stake in the outcome, so its answer can feel more trustworthy than the person actually building your house, who you might assume wants to keep you calm and moving along.
People generally trust an AI's answer over their builder's because it seems to have no agenda. It doesn't, but it also has no way of knowing whether what it's telling you is actually true. A large language model isn't checking facts against reality. It's predicting what a plausible-sounding answer looks like, based on everything it's ever read, and it hands that answer over with the same confidence whether it's right or completely invented. Ask it something loaded, like what happens if a builder hits you with a variation, and it might come back with a dramatic story about someone overseas who lost a fortune to a dodgy contract. None of which tells you anything about your contract, your builder, or what's actually likely on your job.
If you're genuinely worried your builder's doing the wrong thing, that's worth listening to. But the fix isn't a longer conversation with a chatbot. It's a shorter, more direct one with the person actually building your house. Tell them what's worrying you. Very little doesn't get sorted with a straightforward conversation, and if it turns out something can't be, that tells you something too.
How much do you actually need to know?
There's a second way AI can send you down a path that doesn't help anyone, and it's less about mistrust than about volume. You don't need to know exactly how the frames go up, or why something on site got redone, or where the offcut pile went, and AI will happily walk you through all of it if you ask.
Building isn't a factory process. It happens outside, in the weather, by hand, to tolerances measured in millimetres, and things shift a little along the way because every house and every site is different. That's not a flaw in the process. It's just what building is. Chasing every one of those details through an AI chat is a good way to spend a lot of time worrying about things that were never actually a problem.
Builders know their clients are using AI. Clients know their builders are using it too, for drafting, for scheduling, for the admin that used to eat half a day. None of that needs hiding. Used well, it should free up time for the conversation that actually matters, the one where you sit down with your builder and work through what's on your mind, rather than either of you disappearing into a chat window and coming back with a printout.
What this looks like at ZJ Building
At ZJ Building, we’d rather you ask the question than sit there worrying about something because you didn’t want to bother us. If loading your contract and plans into NotebookLM helps you get your thoughts in order before a site meeting, do it, and tell us you did. We're using the same kind of tools on our side of the desk, and we'd rather have that out in the open than pretend neither of us is.
If you're still weighing up who to build with and want to know what that kind of upfront relationship actually feels like before you sign anything, book a discovery call with our team. It's a reasonable place to start.
