GEO Content Strategy

Case Studies

How a Mid-Sized Government Agency Improved Its AI Visibility for Public Health Services

A regional public health department found that residents were turning to AI answer engines for clinic schedules, eligibility rules, and compliance steps — but its deep menus and legacy PDFs left those engines hallucinating hours or omitting the agency entirely. This is a composite, illustrative scenario built to demonstrate the GEO method, not a real client engagement.

Updated June 9, 2026 Hypothetical scenario

Case profile

Industry

Government agencies

Organization size

Mid-sized regional public health department

Team size

2 communications specialists (small team)

Difficulty

Moderate

Estimated timeline

Variable

Figure 1 — profile of the hypothetical agency in this case study
The situation

What problem was the agency facing?

Local residents were increasingly using AI answer engines to find critical community services, clinical schedules, and regulatory compliance rules — but the agency was frequently missing or misrepresented.

Because the agency's web presence relied heavily on deep menu hierarchies and legacy document formats, conversational search models frequently hallucinated operational hours or failed to list the agency as a verified resource.

The challenge

Citizen search behavior has shifted from keyword string queries (like "city health clinic hours") to multi-criteria, conversational strings (such as "I don't have insurance and my child needs school vaccines before Friday, where can I go near the downtown transit center?").

Traditional SEO focuses on keyword density and backlink authority to rank a page. Generative Engine Optimization (GEO), however, requires feeding an AI's internal Knowledge Graph. An AI Knowledge Graph is essentially a massive digital map of connected concepts, entities, and locations. If an AI engine cannot easily map the explicit connections between your organization's physical address, specific service eligibility, real-time operating hours, and jurisdictional authority, it simply leaves your organization out of the generated answer to avoid presenting risky or unverified info to the user.

What citizens are asking AI

  • "Which county clinics offer free ambient air quality testing kits, and what do I need to bring to prove residency?"
  • "I want to open a small catering business from my home kitchen. What are the exact permitting steps and fees required by the local health department?"
  • "Are there any active water boil advisories or food safety recalls issued by the regional government right now?"

Why this matters: When an AI model processes a complex multi-criteria query, it synthesizes an original answer on the fly using the strongest connections it can find in its graph. If your official public data is trapped in an image-based table or written in dense legalese, the AI cannot map those connections reliably, causing it to recommend a different entity or hallucinate outdated instructions.

Baseline GEO audit

To evaluate how effectively major AI models could retrieve and synthesize the agency's public health data, a series of standardized diagnostic prompts were run against the leading generative engines. Ratings are illustrative, not measured.

Category ChatGPT Gemini Claude Perplexity
AI visibility Moderate Weak Weak Moderate
Entity clarity Moderate Moderate Weak Strong
Program/service pages Weak Missing Missing Moderate
FAQ content Weak Weak Missing Weak
Trust signals Strong Strong Strong Strong
Expert profiles Missing Missing Missing Missing
ChatGPT
AI visibility
Moderate
Entity clarity
Moderate
Program/service pages
Weak
FAQ content
Weak
Trust signals
Strong
Expert profiles
Missing
Gemini
AI visibility
Weak
Entity clarity
Moderate
Program/service pages
Missing
FAQ content
Weak
Trust signals
Strong
Expert profiles
Missing
Claude
AI visibility
Weak
Entity clarity
Weak
Program/service pages
Missing
FAQ content
Missing
Trust signals
Strong
Expert profiles
Missing
Perplexity
AI visibility
Moderate
Entity clarity
Strong
Program/service pages
Moderate
FAQ content
Weak
Trust signals
Strong
Expert profiles
Missing

The initial audit revealed a clear pattern: while all four major engines recognized the agency as a trusted, highly authoritative government entity (reflected in the "Strong" Trust Signals due to .gov domain authority), they struggled immensely to extract specific operational details. ChatGPT and Perplexity could occasionally scrape high-level data using live web-browsing capabilities, but their outputs regularly missed nuanced program details. Claude and Gemini completely failed to surface specific clinical programs or localized FAQ data. This tells us that while the AI engines fundamentally trust the domain, the site's layout makes it incredibly difficult for their models to extract actionable answers to user questions.

Key issues found

Three structural gaps explained why a trusted .gov domain still failed to surface actionable answers.

  1. 1

    "Invisible" PDF data caches

    The critical criteria for program eligibility, low-cost sliding scale fees, and environmental health permit processes were locked inside multi-page, non-tagged PDF documents. While AI engines can crawl text within PDFs, they struggle to map contextual relationships inside complex document layouts. If an eligibility table is trapped inside a PDF, the AI model often fails to connect those rules back to the main agency entity, rendering the data invisible during real-time conversational synthesis.

  2. 2

    Ambiguous, promotional headings

    Service pages used ambiguous, vague marketing copy and creative headings (e.g., "Nurturing Our Community's Brightest Futures") rather than explicit, plain-language declarations of what the service actually is. AI engines rely heavily on semantic clear-text headings to quickly classify information blocks. When a heading lacks concrete nouns and entities, the LLM cannot confidently determine whether the section contains a specific medical service, an educational article, or a general mission statement.

  3. 3

    Total lack of backend structured data

    The website featured absolutely no structured schema markup in the source HTML to explicitly define its location networks, service types, or regional jurisdiction. Without machine-readable structured code, an LLM must guess your operational details based entirely on unstructured text. If a page lists multiple phone numbers or addresses without schema tags specifying which number belongs to which clinic branch, the AI engine will often blend the data together or omit it entirely to prevent sharing inaccurate information.

Recommended GEO improvements

Targeted, editorial-first changes the lean team could execute inline within their existing CMS.

Transforming text for direct answers

Before — traditional marketing copy

"Welcome to the portal of our Environmental Health Division. It is the core mission of our dedicated team of regional inspectors to ensure that our beautiful community enjoys the highest standards of culinary safety and sanitation across all of our local dining establishments. If you have recently visited an eatery within our administrative boundaries and encountered conditions that you believe did not align with optimal public wellness or regulatory benchmarks, we provide mechanisms to register these occurrences for internal review."

After — GEO-optimized copy

How do I report a restaurant health violation in [Region Name]? You can report a restaurant health violation directly to the Regional Public Health Department by calling 555-0199 or filling out our online complaint form. Cost: Free of charge. Anonymity: You may file your report anonymously. Response Time: An environmental health inspector will conduct an on-site facility inspection within 48 business hours of receiving your complaint.

Why we chose it: LLMs are trained to reward text that directly answers natural language questions. Giving them a concise, factual summary at the very top of a service page makes it effortless for the model to copy, paste, and cite your exact text as a definitive snippet answer.

Implementing GovernmentService schema

Inject custom JSON-LD schema markup into the HTML header of every core program page, explicitly identifying the service type, provider entity, and service area. Schema provides an absolute baseline of truth that bypasses an LLM's text-parsing guesswork — it lets the agency dictate its exact coordinates, hours, and jurisdiction straight into the AI's data-ingestion pipeline.

Before — standard HTML

HTML 
<div class="clinic-info">
  <h2>Downtown Health Center</h2>
  <p>Come see us at 100 Main Street for immunization
     walk-ins Monday through Wednesday, 8 AM to 4 PM.</p>
</div>

After — with GovernmentService schema

JSON 
{
  "@context": "https://schema.org",
  "@type": "GovernmentService",
  "name": "Public Immunization Clinic",
  "provider": {
    "@type": "GovernmentOrganization",
    "name": "Regional Public Health Department"
  },
  "serviceOperator": {
    "@type": "MedicalClinic",
    "name": "Downtown Health Center",
    "address": {
      "@type": "PostalAddress",
      "streetAddress": "100 Main Street",
      "addressLocality": "Metro City",
      "addressRegion": "ST"
    },
    "openingHours": "Mo-We 08:00-16:00"
  },
  "serviceArea": {
    "@type": "AdministrativeArea",
    "name": "Regional County"
  }
}
A third recommendation — transitioning PDF data into high-density FAQ blocks — rebuilds buried regulatory, permitting, and service-delivery workflows directly on the website as structured web pages built around explicit questions. Web pages parse cleaner in AI scraping loops than embedded document formats, turning unindexed data into a highly crawlable format LLMs can instantly pull from.

Continue learning

Related guides that map directly to the three recommended actions above.

Optimization

Improve Program & Service Pages

Restructure service pages into clean, question-led layouts an engine can lift directly.

Read guide Optimization

Organization Schema

Define your entity, locations, and jurisdiction in machine-readable JSON-LD.

Read guide Optimization

Create AI-Friendly FAQs

Turn buried PDF rules into high-density Q&A blocks engines can parse and cite.

Read guide