Direct answer first, bias disclosed: we build one of these systems, and in our judgment the best AI receptionist for a small city is whichever one can pass a five-part test, answers only from knowledge staff approved, transfers a caller to a department live, alerts a human instantly with a summary, never places an outbound call to a resident, and keeps records of everything. Rocketship was built to pass that test. Hold any vendor, including us, to it.

The problem, stated by a city itself

When Chicago's Office of Inspector General reviewed the city's 311 operation, the report's language was blunt enough to make news: callers described a "black hole." That is a big city with a real 311 budget. Small cities run the same constituent load, permits, utility billing, court dates, facility hours, "who do I even call about this", through a handful of clerks who also do everything else, on lines that close at 4:30 PM. The result is predictable: independent call studies find roughly 62% of calls to small organizations never reach a live person during business hours, and after hours the number is effectively 100%.

The five requirements

1. Approved knowledge only, with no improvisation. A chatbot that guesses a fence-permit rule creates liability a city cannot accept. The system must answer exclusively from content staff reviewed, and say plainly when something is not covered. In Rocketship, Sophie reads the city website, staff review and edit every fact she learned before going live, and her standing instruction is to never answer past that playbook.

2. A live path to a human. Not "we'll email someone." Mid-call, the system should connect the caller to a named destination: the Permits desk, Utilities, the front office, and when no one is available, the right person should get an instant alert with the caller's number, the question, and a summary. Both behaviors ship in Rocketship today.

3. Inbound only. A city system that can place outbound calls to residents is a headline waiting to happen. Rocketship's government configuration is inbound service only, as policy and as design: she never cold-calls a resident.

4. Records. Every call should produce a transcript, an AI summary, and a logged request a clerk can act on Monday morning. Constituent service that leaves no record is not service a public body can stand behind.

5. Keep the published number. Cities should not reprint letterhead for a vendor. Conditional call forwarding, after-hours only or overflow, points the existing published line at the AI and reverses with one carrier code.

The bar for government AI is not a good demo. It is "defensible in a council meeting."

The honest limits

Three things a small city should know before deploying any AI line, ours included. Emergency traffic is out of scope: 911 and true emergency dispatch stay on their own systems, full stop; an AI receptionist belongs on non-emergency and administrative lines. Language coverage should be tested, not assumed, require a live demonstration in the languages your residents actually use before it is promised in public materials. And documents matter: fee schedules and code excerpts live in PDFs, so ask how the knowledge base ingests them, not just the website.

A one-hour test you can run this week

Do not evaluate from a brochure. Pick the five questions your front desk answered most often last week, plus the one that always goes wrong, such as a fee that changed or a permit that depends on the parcel. Put the approved answers in writing first, so you know what a correct reply sounds like.

Then call the line and ask all six. Check that it answers from approved text and says so when it does not know. Ask for a department and see whether the transfer happens live or ends in a promise to email someone. Hang up mid-call and look at what the staff member received: the caller's number, a summary, the time. Finally, ask the vendor in writing whether the system can ever place an outbound call to a resident. If the answer is anything other than a flat no, you have learned what you needed.

Run the same script against every vendor on your list. A one-hour test with identical questions tells you more than any reference call, and it gives the council a record of how the choice was made.

What it costs, and how to test it

Rocketship's government configuration runs on the Mission Control plan, $149/month with a dedicated number and 150 voice minutes included, a figure a clerk can compare against a single answering-service invoice or a staff hire. The government configuration is inbound service only, so the plan is used to answer residents, not to reach out to them. The evaluation costs nothing: call our live line at (405) 873-9151, ask it the hardest question your front desk got this week, and then ask your other vendors for the same phone number. The full government page, including the no-cold-calls policy in writing: deployrocketship.com/government.