ChatGPT for building law? Where it falls short

“Can’t I just ask ChatGPT?” is the first question we hear in every conversation with an office. It is a fair one. A general language model writes well, knows the basic terms of building law and answers in seconds. For many tasks in the office it is a good tool.

For one kind of question it is not enough: the question whose answer goes into a permit submission, a file note or a decision in the project. There it lacks three things.

First: the right edition

Building law in Austria is state law. Nine building codes, and the OIB guidelines apply only once a state declares them binding, each in its edition and on its date. According to the OIB’s overview, Vienna has applied OIB guideline 6 in its 2025 edition since 15 July 2026, while several states had not yet adopted the 2023 guidelines as of the last overview.

A language model answers from what was in its training. It does not know which version applies in your state today, and it often does not say. The answer sounds equally sure whether it is right or an edition out of date.

Second: your project

Few planning questions are purely legal. “Do we need a fire stop here?” depends on the building class, on the façade in the section, on the insulation thickness in the drawing. “What can we build on this plot?” depends on the zoning plan.

In ChatGPT you can upload files, for that one conversation. But the project lives in the office: drawings in versions, permits, room schedules, and a team that needs the same documents. And how your office solved a detail on the last project is something no model can answer if it has never seen your archive.

Third: the evidence

The most dangerous trait of a language model in building law is not that it gets things wrong. It is that it gets them wrong convincingly, with a citation that looks plausible and does not exist, or says something else. If you then have to look up whether the cited section is right, you have saved no time.

A usable answer names the section, the clause of the guideline or the page of the drawing, and the passage is checked before it is shown.

What ChatGPT is good for in the office

To be clear: a general language model is useful for a lot of office work.

• drafting and rewording texts: building descriptions, cover letters, competition texts
• summarising long documents whose content you then check yourself
• explaining terms and reading up on a new subject
• making emails to clients easier to follow

The line runs where the answer becomes the basis for a decision.

What a specialist tool does differently

That is the gap we are building Piloti for. Piloti answers from the state building codes in RIS, the OIB guidelines, your project documents and your office’s archive. Every citation is checked against the source text before it appears, and opens at the marked passage. Piloti looks at drawings as images when the floor plan matters. When a document is missing, such as the zoning plan, it says so instead of answering as if it had read it. And when needed, the answer becomes a file note in the project.

The full side-by-side is on Why Piloti. How an answer comes about technically is in How Piloti works.

Try it on one question

Take a question from a current project, put it to ChatGPT and send it to us. Afterwards, do not compare which answer sounds more fluent, but which one you can find in the original.