Last updated: August 2026
Can AI interpret building codes is a question with a useful answer and a dangerous one, depending on what you mean by interpret. AI can explain what a code concept means and summarize a passage you give it, which genuinely speeds up research. It cannot be trusted to tell you which edition your jurisdiction adopted, to quote the correct section, or to know your local amendments, and it will state a non-compliant value with total confidence. The tell is that even the code publisher’s own AI tool warns you to verify everything against the source. This guide separates what AI does well from what it gets wrong, with the evidence, and shows how to use it without getting burned. It pairs with our guide on which standard applies to your design and, for the toolkit, the hub on the best AI tools for civil engineers.
The short answer
Partly, and only as a research aid. General AI can explain code intent in plain language and summarize a passage you paste in, but it does not know your locally adopted edition or amendments, it hallucinates section numbers, and it cannot tell when it is wrong. Even ICC’s own AI Navigator, trained on the I-Codes, links every answer back to the referenced sections and tells users never to take the AI’s word for it. Treat any AI reading of a code as a lead to verify, never as compliance. The adopted edition and the authority having jurisdiction govern, not AI and not this article.
What AI can and cannot do with building codes
The line runs between explaining and citing. AI is reasonable at the first and unreliable at the second.
| Reasonable, with verification | Risky, do not trust |
|---|---|
| Explaining a code concept or intent, such as what an occupancy classification or a fire-resistance rating means | Citing the specific edition your jurisdiction adopted, which varies widely and lags the model’s training |
| Summarizing a code passage you paste in, grounded on the text you provide | Quoting a section number or a numeric limit from memory, which it will sometimes invent |
| Pointing you toward which code area likely applies, as a starting point | Accounting for local amendments that your state or city made to the model code |
| Drafting questions to bring to a plan reviewer or the authority having jurisdiction | Stating whether a design is compliant, which is the reviewer’s and the authority’s call |
Why AI gets code citations wrong
The failure is not random, and it is measured. When researchers asked models specific, verifiable legal questions, hallucination rates ran from about 58 percent for GPT-4 to about 88 percent for one open model, and the models often could not tell when they were wrong or correct a false premise (Dahl et al., 2024). Those are legal-domain figures, used here as an analogy, but code citation carries the same risk: a confident, specific, and wrong reference. The pattern shows up directly in engineering documents too, where a multimodal benchmark found leading models struggle to reliably retrieve the relevant rules from a rulebook (DesignQA, 2024).
On top of that, a general model has no reliable view of your jurisdiction. Editions of the IBC and IRC vary from place to place and lag the newest release, cities and states amend the model code, and the model cannot know which set governs where you are building. It is the same reason AI can get units wrong: the answer looks authoritative and is not.
Even the code publisher says verify
The most telling evidence comes from the code publisher itself. The International Code Council offers an AI Navigator inside its Digital Codes Premium platform, trained on the I-Codes and a growing list of adopted state codes. It can answer questions and summarize sections, but every response links back to the referenced code sections, and ICC’s own guidance is to never take the AI’s word for it and to read the full context yourself (ICC Digital Codes). When the organization that writes the code builds an AI on its own authoritative text and still tells you to verify, that is the ceiling for what to expect from a general chatbot working from memory.
Academic work points the same way. A study presented at a 2023 construction-automation conference used ChatGPT to help convert building-code requirements into a computable form and found it still lagged a purpose-built rule-based approach on accuracy, while showing promise as a way to speed the work (ISARC 2023). The consistent theme is accelerator, not authority.
How to use AI on codes safely
Used carefully, AI is a real time-saver for code research. The discipline is to keep it grounded and never let it be the authority.
- Paste the text, do not rely on memory. Give the model the actual code passage and ask it to explain or summarize that, which keeps it grounded on real text rather than a guess.
- Use it to orient, then verify. Let it suggest which code area to look at, then read the authoritative clause from the adopted edition yourself, with the discipline in how to verify an AI engineering answer.
- Confirm the edition and amendments with the authority. The adopted year and any local amendments come from the authority having jurisdiction, not from a model, and they decide the outcome.
- Never submit AI output as compliance. AI code analysis is a research note, not a determination, and the professional responsibility stays with you, the same as when you decide whether to use AI on stamped drawings.
Read alongside the broader limitations of AI in engineering and our guide on whether AI can read technical drawings, the pattern is consistent: AI is strong at explanation and weak at the precise, jurisdiction-specific detail that compliance turns on.
Frequently asked questions
Can ChatGPT read and correctly apply the building code for my city?
It can explain code concepts and summarize a passage you paste, but it does not reliably know which edition your city adopted or your local amendments, so it cannot correctly apply the code on its own. Use it to understand and orient, then verify against the adopted code and confirm with the authority having jurisdiction.
Does AI know which edition of the IBC or IRC my jurisdiction adopted?
Not reliably. Adopted editions vary by jurisdiction and lag the newest release, and a general model has no dependable view of what your authority has adopted or amended. Always confirm the edition in force with the building department rather than trusting the model.
Is the ICC AI Navigator accurate, and can I rely on its answers?
It is grounded on the official I-Codes and links every answer to the referenced sections, which is more trustworthy than a general chatbot. Even so, ICC tells users never to take the AI’s word for it and to read the full code context, so treat it as a fast way to find the right section, not as a final answer.
Will AI invent building-code section numbers?
Yes, it can. Research on specific, verifiable questions found high hallucination rates and models that could not tell when they were wrong, and benchmarks show trouble retrieving the right rule. Never quote a section number or limit that came from a model without confirming it in the adopted code.
Can I submit AI-generated code analysis to a plan reviewer or the authority?
No. AI output is a research aid, not a compliance determination, and the professional responsibility for the submission is yours. Verify everything against the adopted code, and let the plan reviewer and the authority having jurisdiction make the compliance call.
Sources
- Legal hallucination rates, 58 to 88 percent on specific questions: Dahl et al., 2024 (arXiv:2401.01301)
- Models struggle to retrieve relevant rules from a rulebook: DesignQA, 2024 (arXiv:2404.07917)
- AI Navigator grounded on the I-Codes, with a verify-everything framing: ICC Digital Codes
- ChatGPT for automated code-compliance checking, accelerator not authority: ISARC 2023
About the author: this guide was written and edited by the CognitiveFuture editorial team, which researches how AI tools fit real professional workflows. We cite primary sources for the claims we make and update our guidance as tools and research change. We do not test products ourselves; our assessments synthesize peer-reviewed research, primary documentation, and practitioner reporting.
This is general information, not legal or code-compliance advice. Adopted editions and amendments vary by jurisdiction, so confirm requirements with the authority having jurisdiction and a qualified professional before relying on them.
Richard Johnson
Richard Johnson is an AI specialist at one of the world's largest technology companies, where he has spent the past three years helping organizations adopt AI. CognitiveFuture extends that work publicly: gathering the available evidence on each tool, from vendor documentation to independent reviews and user feedback, and cutting a crowded market down to the right choice for the job in front of you.