Can AI read a township zoning ordinance for you?

Ask an AI assistant whether a property can run as a short-term rental in a given Michigan township, and it will answer confidently — usually in under ten seconds, usually in plain English, and often close to right. Close is the operative word. Michigan has more than 1,500 townships, cities and villages, each writing and amending its own zoning ordinance, and a general-purpose AI model trained on a snapshot of the internet is not the same thing as someone who called the township this week. A short stretch of West Michigan shoreline can cross a city, the township next door and a village between them, and it is entirely normal for those three governments to answer the same question three different ways.

That's not an argument against using AI to research property rules — it's a genuinely fast way to get oriented. It's an argument for knowing exactly which part of the job it's actually doing.

What it's genuinely good at

Zoning ordinances are long, repetitive and written in a register nobody speaks casually. Feeding one to an AI assistant and asking it to summarize the residential districts, list the setback requirements or explain what an accessory dwelling unit is allowed to be is a legitimate use of the tool, and it saves real time — the kind of reading that used to take an evening can take ten minutes. It's also good at comparison — laying a handful of neighboring townships' short-term rental rules side by side, or turning a dense ordinance into a list of specific questions to bring to the zoning office.

• Translating dense ordinance language into plain English
• Comparing rules across several named jurisdictions at once
• Turning a long ordinance into a checklist of what still needs a human answer
• Drafting the questions to ask a zoning administrator, not answering them

Where it fails

The clearest evidence on how often this goes wrong doesn't come from real estate — it comes from law. A 2024 Stanford RegLab study of AI legal research tools built specifically for attorneys, with retrieval systems designed to ground answers in real case law, still found hallucination rates between 17 and 33 percent, including confident citations to sections that didn't exist. A general-purpose assistant asked to interpret a township ordinance with no such retrieval system behind it is working with less structure than that, not more.

Municipal zoning adds its own layer of risk on top. Ordinance text lives in different places — a township's own website, or a codification service like Municode or American Legal Publishing — and those services say plainly that their libraries can lag behind a municipality's most recently adopted legislation. An AI model trained on whatever was crawled months or years earlier inherits that lag with no way to flag it. Michigan's short-term rental rules are also actively in motion: as of this year there is still no statewide law governing them, competing bills have sat in committee since 2024, and regulation happens entirely at the local level in the meantime — which means an answer that's accurate today can be overtaken by a township vote next month.

• Ordinance text baked into the model's training data going stale without warning
• A confident, specific-sounding section number that doesn't exist in the current code
• Conflating a township with a same-named city or village that has its own, different ordinance
• Missing an amendment the township board adopted in the last year
• No signal at all that state legislation on the topic is actively pending

The verification workflow that actually works

Use the AI pass for what it's fast at — a first read of the ordinance, a plain-English summary, a list of open questions — and then verify every citation against the township's own posted code before it informs an offer, a renovation plan or a rental business plan. That means pulling the actual ordinance text, not a summary of a summary, and calling the zoning or planning office directly for anything with a permit or a purchase price riding on the answer. A five-minute phone call with the person who administers the ordinance settles in one conversation what an AI model can only guess at from the outside.

Treat an AI assistant here the way you'd treat a knowledgeable friend who used to live in the township — worth listening to, worth using to get organized fast, and never the last call before you sign something.