ChatGPT Is Good. Just Not for This.
I use ChatGPT. I use Claude. I use multiple AI tools every day in running my business. These are powerful, useful technologies. For the right tasks. Drafting correspondence. Summarising long documents. Research assistance. Brainstorming. They save time.
But when someone asks me whether they can use ChatGPT to manage their NZS 3910 contract obligations, my answer is the same every time: you can, the same way you can drive a nail with a screwdriver. It will work sometimes. But it is the wrong tool for the job, and eventually it will cost you.
The gap between generic AI and purpose-built construction intelligence is not a technology gap. It is a knowledge gap. ChatGPT does not know your contract. It does not track your deadlines. It does not understand what happens when Special Conditions amend Clause 13.3.1 from "as soon as practicable" to a 10-working-day absolute bar.
What Generic AI Gets Wrong
I have tested this extensively. Ask ChatGPT to summarise NZS 3910 obligations and it will give you a reasonable overview. Ask it about a specific clause interaction between the standard form and a set of Special Conditions, and the quality drops sharply. Here are the patterns I see:
It hallucinates citations. AI models will confidently cite reports, cases, and publications that do not exist. I have personally verified AI-generated citations that referenced reports with plausible-sounding names from real organisations. The reports were not real. The numbers were not real. The AI presented them with complete confidence. On a construction project where a wrong clause reference could affect a claim worth hundreds of thousands of dollars, that is not an acceptable error rate.
It does not understand Special Conditions. The standard NZS 3910 is a known document. But every real project modifies it through Special Conditions. Government agency contracts can contain anywhere from 7 to 310 Special Conditions. Each one changes the risk profile. ChatGPT can tell you what Clause 12.4 says about retention. It cannot tell you how your specific Special Conditions have modified the retention release mechanism on your specific contract.
It has no memory. You ask ChatGPT a question about your contract on Monday. You ask a follow-up question on Wednesday. It does not remember Monday's conversation. It does not know that a notice period started running, that a payment claim was issued, or that a variation instruction was given. Every conversation starts from zero. That is fundamentally incompatible with contract administration, which is entirely about tracking things over time.
It does not monitor. Contract administration is not about answering questions. It is about knowing what needs attention before anyone asks. A CCA payment schedule deadline does not announce itself. A notice period runs silently. The system needs to flag these proactively. ChatGPT sits and waits for you to ask. If you do not ask the right question at the right time, the deadline passes.
Under the Construction Contracts Act 2002, missing a payment schedule response deadline turns the full claimed amount into a debt payable immediately. "I asked ChatGPT and it didn't mention the deadline" is not a defence. The legislation does not care which tools you use. It cares about dates.
What Purpose-Built Construction Intelligence Does Differently
The difference is not that construction AI uses a better model. It often uses the same underlying technology. The difference is in three areas:
Domain configuration. A system built for NZS 3910 contract administration knows the clause structure, the common Special Conditions amendments, the CCA payment framework, and the notice requirements. It knows that Clause 9.2.1 requires the contractor's programme within 10 working days of the contract. It knows that Clause 9.8.3 gives the engineer 10 working days to respond to variation pricing. It knows these because someone who has administered these contracts for 10 years configured it.
Persistent state. The system knows your contract. Not NZS 3910 in general. Your specific agreement, with your specific Special Conditions, with your specific deadlines. It tracks obligations continuously. When a payment claim arrives, the response deadline is calculated and monitored automatically. When a notice period starts running, the system counts the working days and flags when action is needed.
Proactive surfacing. The system does not wait for you to ask the right question. It tells you what needs attention today. It flags the payment schedule response due in three working days. It highlights the variation assessment that is overdue. It shows you the notice period that expires next Tuesday. That is the fundamental difference between a question-answering tool and an intelligence system.
ChatGPT is a brilliant research assistant who knows a lot about construction but has never worked on a site, forgets everything between meetings, and only answers when spoken to. A purpose-built system is a contract administrator who knows your project, tracks everything, and taps you on the shoulder when something needs your attention.
The "Can't We Just Use ChatGPT?" Conversation
This comes up in every meeting. Someone on the board has used ChatGPT to write an email and now wonders whether the firm needs to pay for project intelligence when they have a $20/month AI subscription.
The answer is the same as it has always been in construction: you get what you configure. You could run your project accounts in a spreadsheet instead of an accounting system. You could do your programming in Microsoft Project instead of Primavera. You could track your correspondence in email instead of a document management system. Some firms do. They accept the risk.
The question is not whether ChatGPT can do something with your contract. The question is whether you are comfortable managing a multi-million dollar contractual obligation on a tool that has no memory, no monitoring, no domain configuration, and no accountability.
For a $30M+ project where a missed CCA deadline can make a six-figure sum payable overnight, where a missed notice period can extinguish a legitimate claim, where 900+ adjudications have been filed through the BDT since 2008, the answer should be obvious.
Use the Right Tool for the Right Job
ChatGPT is excellent for many tasks. Use it for drafting, research, and analysis where the output will be reviewed by a qualified professional before anyone acts on it. That is a sensible, cost-effective use of the technology.
But do not use it as your contract administration system. Do not use it to track obligations, monitor deadlines, or manage compliance. That requires a purpose-built system, configured by someone who understands the contracts, the legislation, and the way projects actually run.
The firms that understand this distinction will use both: generic AI for productivity gains across their business, and domain-specific intelligence for the contract work where the stakes are highest.
The firms that do not understand it will find out the hard way. Probably when a deadline passes that nobody asked ChatGPT about.
Provan builds AI-powered operating systems for infrastructure and engineering businesses, covering six domains: Pipeline, Contracts, Projects, People, Finance, and Risk. Unlike generic AI tools, the Contracts domain maintains persistent state on your agreements, tracks every obligation and deadline, and surfaces what needs attention before you have to ask. Built from 10 years managing projects from $10M to $750M.
This article provides general commentary on AI tools in construction. It is not legal or technology advice. References to ChatGPT are based on publicly available capabilities as of March 2026. AI tools evolve rapidly. Consult qualified professionals for decisions about your contract administration systems.
See the Difference for Yourself
Book a working session and we will show you what a purpose-built project intelligence system looks like next to a generic AI tool. Same contract. Different outcomes.
Book a Working Session →Or email stephen@provan.ai