Fred: A Resident Assistant That Won’t Lock Anyone Out or Guess a Deadline

A resident opens the county site after hours and asks, "when is the absolute last day I can appeal my property assessment?" A generic chatbot answers with a confident date. If that date is wrong, the resident misses the window and the agency is the reason. A second resident, using a screen reader, never gets that far, because the chat widget was built without keyboard or screen-reader support and simply does not respond. The office wanted a tool to deflect routine calls. It got one that invents legal deadlines for some residents and shuts out others entirely.

That is the exposure the threat and standard pieces in this series cover. This article is the answer: what a compliant AI assistant for government actually looks like, and why accessible-by-design and route-don’t-guess are the parts that keep the agency out of trouble.

The Real Choice Is Governed or Ungoverned

Agencies tend to debate whether to use AI at all. Residents settled that already. They expect to find an answer the moment they look, and the office that makes them wait on hold or in line loses their trust. The decision that matters is whether the assistant is governed.

An ungoverned chatbot guesses at filing deadlines, improvises an interpretation of an ordinance, and ships in a widget that a resident on a screen reader cannot operate, which is the access gap Title II of the ADA and the DOJ’s web and mobile accessibility rule are built to close. A governed assistant answers what is published, routes the rest, and works for every resident. Same convenience, none of the failure.

How Fred Holds the Line

Fred starts from the opposite default of a general-purpose chatbot. A generic bot answers anything unless told otherwise; Fred answers only what it is cleared to, and the boundary lives in the system rather than in a prompt it can drift away from.

In practice Fred reads from the agency’s own published pages and documents, answers the routine question, and routes anything touching a statutory deadline, a legal interpretation, eligibility for a benefit, or an enforcement matter to the right staff queue. It does not invent a date, because a wrong deadline from the agency’s own tool can cost a resident a right. It is built to the WCAG-aligned standard the DOJ rule references, so keyboard and screen-reader users get the same service everyone else does. Fred runs more than 50 industry guardrail packs, and the public-sector pack is built around accessibility, deadline and legal-interpretation boundaries, and accurate routing to the responsible department.

The difference shows up under pressure. Ask a prompt-instructed bot for a deadline three different ways and it eventually answers the phrasing it was not warned about. Fred does not depend on recognizing the wording; it decides what is allowed out before the answer exists. "Will not" is a suggestion. "Cannot" is an architecture.

Fred vs. a Generic AI Chatbot

The table is the whole argument in one screen. A generic tool is helpful right up to the moment helpful becomes a wrong legal deadline or an access barrier. Fred is helpful everywhere that carries no risk and structurally careful everywhere that does.

What Your Agency Actually Gets

Set the compliance framing aside and look at the operational case. Fred answers the high-volume questions that flood the front desk and the phone lines, office hours, which form to file, where to pay, what a permit process involves, and it answers instantly, around the clock, in plain language and in the resident’s own words. It routes the rest to the correct queue with the context attached, so staff pick up requests already sorted instead of triaging from scratch.

There is a service-equity angle too. Because Fred is built accessible from the start, the resident on a screen reader, the resident who reads at a different level, and the resident who only has a phone all get the same answer at the same speed. That is not a compliance checkbox; it is the agency actually serving the whole public. Staff time goes to the cases that need judgment, not to repeating the hours and the filing address two hundred times a week.

So the question is not whether residents will expect self-service answers from your site. They already do. It is whether the tool answering them is accurate and usable by everyone, before one wrong deadline or one locked-out resident becomes the story.

Frequently asked questions