When Your City’s Website Answers for the Government

A resident on the county website at 9 p.m. asks whether their building permit covers the deck they want to add, and whether the filing deadline for an appeal has passed. The chatbot answers both with confidence. It is wrong on the first and outdated on the second. When the inspector and the clerk contradict it the following week, the resident is not just annoyed. They relied on an answer the government’s own website gave them, and now the agency is explaining how that happened.

A general-purpose chatbot on a municipal site is a different animal than one on a store. It speaks in the government’s voice, about deadlines, eligibility, ordinances, and rights, where a wrong answer is not a customer-service slip but a due-process and equity problem. The tool was added to cut call volume at the front counter. What it actually does is issue official-sounding answers no one cleared and, in many cases, do it in a way that locks out the residents the law most protects.

A Wrong Answer From the Government Is Not Neutral

When a business chatbot is wrong, a customer buys elsewhere. When a city chatbot is wrong, a resident misses an appeal window, builds without the right permit, or assumes they do not qualify for a benefit they were entitled to. Residents are entitled to rely on what the government tells them, and the agency cannot wave away a transcript by saying a bot wrote it. Deadlines, eligibility, and code interpretations are exactly the answers that require a human who is accountable for them, and exactly the answers a generic model is happy to improvise.

Accessibility Is Now a Hard Requirement, Not a Best Practice

The deeper exposure is one most agencies have not priced in. The Americans with Disabilities Act bars state and local governments from excluding people with disabilities from their services, and in 2024 the Department of Justice made the web obligation explicit: a rule requiring state and local government web content and mobile apps to meet WCAG 2.1 Level AA, on a compliance timeline now running. A bolt-on chatbot widget that is not built to that standard, one a screen-reader user cannot operate, or that traps keyboard navigation, does not just frustrate a resident. It can put the agency out of compliance with Title II on the very channel it deployed to improve access.

The Record-Keeping Trap

Government runs on records. A chatbot that answers constituents may be generating communications that fall under public-records and retention obligations, and an off-the-shelf widget that logs conversations into a vendor’s cloud, or discards them, can collide with both. If the system also moderates or filters what residents can submit, it edges toward First Amendment territory that private businesses never have to think about. The agency did not set out to build a records system or a public forum. It bought a chat widget, and inherited the duties anyway.

A Second Failure Mode: Equity

The quieter danger is uneven service. A resident who reads English, navigates a screen easily, and phrases a question the way the model expects gets a usable answer. A resident who does not gets a worse one, or none. When a government leans on a chatbot as a front door to benefits, permits, or services, that gap is not a UX footnote; it is the agency delivering its services unequally, on a channel it chose. A bot that confidently tells one resident they "probably do not qualify" for assistance, based on a loose reading of eligibility, can steer a person away from help they were entitled to, and do it without a caseworker ever seeing the exchange. Add the resident data these conversations collect, names, addresses, situations that can be sensitive, and an off-the-shelf widget that routes that information through a vendor’s systems becomes a privacy question layered on top of the accuracy and accessibility ones. The agency inherits all of it at once.

Who Owns the Answer

Strip away the technology and the duty is unchanged. A clerk who gives a resident the wrong deadline creates a problem the agency has to remedy, and public bodies build training and review around exactly that. When an unsupervised automated system gives that wrong answer at scale, in the government’s name, the accountability does not evaporate. It lands on the agency, now with a written transcript, an accessibility obligation attached to the same tool, and residents who were entitled to trust it.

The governments that get burned are not the ones that modernized their service. They are the ones that dropped in a consumer chatbot, assumed "helpful" was the same as "accountable," and found out the difference when a resident, or an advocacy group, pushed back. The fix is not to abandon AI at the counter. It is to use one built for a setting where the answer is official, the channel must be accessible, and the conversation may be a record.

Frequently asked questions