AI Visibility for Local Government and Public Services

Residents now ask assistants the questions they used to phone in. The answers they get are short, confident and often missing the caveat that matters. Here are realistic prompts, using a hypothetical city called Riverton:

• "What time does Riverton City Hall close on Fridays?"
• "How much does a residential building permit cost in Riverton?"
• "Do I need a permit to build a fence in Riverton, and how tall can it be?"
• "How do I pay a Riverton parking ticket online?"
• "When is bulk trash pickup in Riverton this month?"
• "What documents do I need to register to vote in Riverton County?"
• "¿Dónde renuevo mi licencia de negocio en Riverton?"

Each one has a single correct answer that changes over time. Hours shift on holidays. Fee schedules change with the budget. Procedures move online. When an assistant gets one wrong, the resident pays for it, not the assistant. The same pattern shows up in other high-stakes fields, which is why the approach here overlaps with AI visibility for hospitals and health systems.

Residents use whatever assistant is already on their phone or in their search results. That means you should check at least these, as of October 2026:

• Google AI Overviews and AI Mode. Google says a page must be indexed and eligible to show with a snippet to appear as a supporting link, with no additional technical requirements.
• ChatGPT. Search inside ChatGPT depends on OpenAI's OAI-SearchBot. ChatGPT also answers without searching at the model's discretion, and then it relies on training data with a fixed cutoff. For OpenAI's GPT-5.6 models that cutoff is February 16, 2026, and the Instant path's is August 31, 2025. A fee change after those dates cannot appear in an unsearched answer.
• Microsoft Copilot, which grounds on Bing's index.
• Perplexity and Claude, each with its own search crawler.

Every engine retrieves differently, and retrieval fires per query. One blended score would hide the engine that is wrong. Check each engine separately and in the languages your residents use, since AI visibility tools can test answers by country and language only if you set them up to.

When an assistant cites another site instead of yours, that site is shaping the answer. Being named correctly on those sites is associated with appearing in the answer, though nothing guarantees it. Check these first:

• State and county portals that list your offices, fees or forms, often copied once and never updated.
• Your Google Business Profile and Bing Places listings for each office, including holiday hours.
• Unofficial directories that republish hours and phone numbers for city halls, DMVs and courts.
• Local news articles about a fee increase or a new online portal, which can outrank the source page for months.
• Permit expediter and contractor sites that summarise your process to sell a service.
• Community forums where residents describe how a process worked years ago.

You cannot edit most of these. You can contact the owner, update the listings you control, and make sure your own page states the current fact plainly. The local business fact check for AI answers walks through the listing side step by step.

Nobody has published a census of AI errors about municipal services, so treat this as a list of what to check, not a ranking. These facts carry the most risk because they change and because the cost of a wrong answer is high:

• Office hours, especially holiday closures, lunch closures and walk-in versus appointment-only windows.
• Permit and license fees, including which fiscal year's schedule applies.
• Eligibility rules, such as residency requirements or which projects need a permit at all.
• Deadlines, like tax payment dates, appeal windows and registration cutoffs.
• Where to go, after a department moves buildings or merges with another.
• How to pay or apply, after a process moves from paper to an online portal.
• Jurisdiction, when an assistant gives city rules for a county service or confuses two towns with similar names.

A specific fact stated on a fetchable page gives the engine a sentence it can quote. That is the one on-page signal with experimental support, and it is why the same logic applies to publishing prices so AI answers get them right.

Public-service websites often keep the authoritative fact in a PDF: a fee schedule, an ordinance, a form instructions sheet. That creates three problems. The fee sits on page nine instead of in a sentence. Older PDFs stay online with last year's numbers. And how each AI crawler handles PDF text is not consistently documented.

JavaScript is the second gap. Vercel's crawler study found that ChatGPT and Claude crawlers fetch JavaScript files but do not execute them, so they cannot read client-side rendered content. Googlebot does render JavaScript. If your office hours load from a widget, or your fee calculator only shows a number after a click, many AI crawlers see an empty box.

The fix is plain. For each high-stakes fact, put one sentence in the HTML of a public page: "A residential fence permit costs $X as of July 1, 2026." Keep the PDF as the legal source and link to it. Remove or clearly label superseded PDFs.

Access is the gate before anything else. Run these checks on your main service pages:

1. robots.txt, per bot. Rules are independent per crawler. OpenAI says GPTBot governs training and OAI-SearchBot governs ChatGPT search, and each setting is independent. A blanket AI block can remove you from ChatGPT search without anyone deciding that. If you want search visibility, allow the search crawlers explicitly:

User-agent: OAI-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Claude-SearchBot
Allow: /

2. Firewall and CDN. OpenAI says eligibility needs both robots.txt permission and a host or CDN that allows its published search bot IP addresses. A web application firewall can return a 403 to a crawler that robots.txt allows, and no dashboard will tell you. The AI Crawler Checker requests a page as each AI crawler and shows which ones your firewall turns away.
3. Snippet controls. Look for noindex, nosnippet, max-snippet:0 and data-nosnippet inherited from a template. Google and Bing document these as controls that keep content out of AI answers.
4. Rendering. View the raw HTML of each key page with JavaScript off and confirm the fact is there.

If your IT or security team owns the firewall, bring them the specific crawler names. Blocking training bots and allowing search bots is a legitimate, separate choice.

Week 1: inventory. List your 20 highest-volume resident questions from call logs, front-desk notes and site search. For each, record the correct answer, the page that should state it, and the date it last changed.

Week 2: test and log. Ask each question in ChatGPT, Google AI Overviews or AI Mode, Copilot, Perplexity and Claude, logged out. Record engine, date, answer, the source cited, and whether the answer was right. The free checker on this site gives a snapshot for up to three buyer questions if you want a starting point. Note which wrong answers cite a third-party page.

Week 3: fix access and pages. Run the access checks above. Move each high-stakes fact into a plain HTML sentence with an effective date. Label or remove superseded PDFs. Update the listings you control. For broader local context, see how AI is changing local SEO.

Week 4: retest and set a cadence. Re-ask the same questions and compare against your log. Some engines will update quickly and some will not, and unsearched answers stay frozen until the next model. Schedule a recheck after every fee change, holiday schedule and fiscal year start. Nobody can promise the answers will change. You control what the engines can reach and read.