How to Measure Your Home Care Agency’s Visibility in ChatGPT

Your agency’s ChatGPT visibility is not a single ranking. It is the percentage of relevant, repeatable questions for which your agency is mentioned accurately, cited as a source, or sends a measurable visit or inquiry to your website.

For a Minnesota home-care agency, the practical way to measure that visibility is to build a fixed prompt set around your services and service areas, test it on a consistent schedule, record mentions and citations, verify every factual description, and connect the results to website traffic and inquiries. The goal is not to “win” one screenshot. It is to see whether your agency becomes more consistently discoverable across the questions families actually ask.

What “visibility in ChatGPT” actually means

Traditional search reporting gives you familiar metrics such as impressions, clicks, click-through rate, and average position. ChatGPT does not provide a comparable agency-by-agency ranking report.

Instead, measure four separate outcomes:

1. Mention: Is your agency named in the answer?
2. Citation: Is your website linked or cited as a source?
3. Accuracy: Are the agency name, services, location, service area, and contact path described correctly?
4. Action: Did the answer lead to a website visit, call, form submission, or other qualified inquiry?

These outcomes are related, but they are not interchangeable. A mention without a citation can build awareness. A citation without a prominent mention can still send traffic. An inaccurate recommendation can create risk rather than value. A visit matters more when it becomes a real inquiry.

OpenAI says ChatGPT responses that use web search may include citations, and users can open those citations to review the sources. OpenAI also says referral links from ChatGPT search include the parameter utm_source=chatgpt.com, which makes attributable visits visible in analytics when tracking is configured correctly.

Start with a controlled baseline

A useful baseline should reflect your actual market, not a handful of flattering prompts written around your brand name.

Create a prompt library that covers five intent groups:

Use natural language. Include the cities you genuinely serve, the services you actually provide, and neutral questions that do not contain your agency name.

A practical starting set is 20 prompts:

• Four discovery prompts
• Four service-specific prompts
• Four comparison prompts
• Four trust or verification prompts
• Four cost, eligibility, or next-step prompts

Twenty is not an industry standard. It is a manageable internal sample that is broad enough to reduce the influence of one unusual answer.

Separate branded and unbranded prompts

Branded prompts answer, “Does ChatGPT understand our agency?” Examples include:

• “What services does [Agency Name] provide?”
• “Where does [Agency Name] operate?”
• “How do I contact [Agency Name]?”

Unbranded prompts answer, “Does ChatGPT discover us when a family does not already know our name?” Examples include:

• “Home care agencies near Brooklyn Park”
• “In-home respite care in Anoka County”
• “Minnesota home care provider for an aging parent”

Track the two groups separately. Strong branded accuracy with weak unbranded visibility means the entity is understood but not yet being surfaced for category demand.

Use a repeatable testing protocol

AI answers can vary with wording, location, timing, available web results, and conversation context. Treat each answer as an observation—not a permanent rank.

For cleaner comparisons:

1. Use the same prompt wording each month.
2. Start each test in a new conversation.
3. Use the same account state and location settings when possible.
4. Record whether web search was used.
5. Save the answer date, answer text, cited URLs, and named competitors.
6. Repeat priority prompts more than once before drawing a conclusion.
7. Avoid testing immediately after telling ChatGPT about your agency in the same conversation.

Never enter a prospective client’s name, diagnosis, care plan, phone number, or other sensitive information. The prompt library should use generic scenarios and public business facts.

Record what the answer actually says

For every prompt, capture:

• Prompt text
• Test date
• Platform and mode
• Your agency mentioned: yes or no
• Position in any presented list
• Your site cited: yes or no
• Cited page URL
• Business facts accurate: yes, partly, or no
• Competitors or directories named
• Notes about the answer
• Screenshot or exported evidence, where appropriate

Position can be useful context, but it should not become the sole KPI. ChatGPT may return prose, a short list, a map, or a sourced summary. “Third” in one answer is not equivalent to a stable third-place search ranking.

Build a simple AI visibility score

A score helps summarize the baseline, but the underlying observations should remain available.

One transparent internal model is:

• 40% mention rate: percentage of prompts that name the agency
• 25% citation rate: percentage that cite the agency’s website
• 25% accuracy rate: percentage with correct public business facts
• 10% action rate: percentage of attributable AI visits that complete a meaningful conversion

For example:

AI Visibility Score = (Mention Rate × 0.40) + (Citation Rate × 0.25) + (Accuracy Rate × 0.25) + (Action Rate × 0.10)

This is a management framework, not a score published or endorsed by OpenAI. Keep the weights consistent so month-to-month changes mean something. If you do not yet have enough referral traffic to calculate an action rate, report that component as “insufficient data” rather than assigning zero.

Also report the raw counts. “Mentioned on 6 of 20 prompts” is easier to understand than a score alone.

Verify accuracy against authoritative sources

A home-care agency should not count an answer as successful merely because its name appears.

Check public facts against:

• The agency’s website
• Its Google Business Profile
• The Minnesota Department of Health provider directory
• Current service and location pages
• Current contact information
• Other official profiles the agency controls

The Minnesota Department of Health provides a public directory for home care and home health agencies that includes licensure or registration information. Use it as a verification source where applicable; do not turn a missing or ambiguous AI statement into a licensing claim.

Flag each answer if it:

• Names a service the agency does not provide
• Places the agency in a city it does not serve
• Uses an outdated phone number or URL
• Confuses the agency with a similarly named provider
• Makes a licensing, insurance, pricing, or availability claim without support

Accuracy problems should become priority fixes. They may point to conflicting directory listings, vague service pages, weak entity signals, or outdated information across the web.

Track citations and the pages earning them

A citation tells you which page ChatGPT considered useful for that answer.

Classify cited pages into groups:

• Homepage
• Service page
• Location or service-area page
• Educational article
• About or credentials page
• Contact page
• Third-party profile or directory

Then ask two questions:

1. Is ChatGPT citing your own website or only third parties?
2. Is it citing the page you would want a family to land on?

If a directory is cited while your agency is only mentioned, the directory may currently provide clearer or better-supported information. That is a content and entity opportunity—not proof that the directory has a special partnership with the platform.

OpenAI’s publisher guidance says public sites can appear in ChatGPT search and recommends allowing OAI-SearchBot if publishers want content to be discoverable and cited. OAI-SearchBot is the search crawler. It should not be confused with GPTBot, which OpenAI identifies separately for potential model training controls.

Measure ChatGPT referral traffic in GA4

Mentions and citations show presence. Analytics show whether that presence produces visits.

In Google Analytics 4:

1. Open Reports → Acquisition → Traffic acquisition.
2. Change or add the dimension Session source / medium.
3. Search for chatgpt.com.
4. Add landing page as a secondary dimension.
5. Review sessions, engaged sessions, and configured key events.
6. Compare the current period with the previous period and the same period last year when enough data exists.

Google describes the Traffic acquisition report as the place to understand where new and returning website visitors come from. OpenAI says ChatGPT search referral URLs automatically include utm_source=chatgpt.com.

Track meaningful actions such as:

• Contact-form completion
• Click-to-call
• Appointment request
• Email click
• Download of a care guide
• Directions or service-area interaction

Do not report every session as a lead. For phone calls, use call tracking or a documented intake question such as “How did you hear about us?” while respecting privacy and consent requirements.

Also understand the limitation: some AI-influenced visits may arrive as direct traffic, through a copied URL, or after a later branded Google search. Treat GA4 referrals as confirmed traffic, not the complete universe of AI influence.

Use Search Console as a supporting signal

Google Search Console does not measure ChatGPT mentions. It does show whether the pages you want AI systems to discover are visible in Google Search.

As of August 31, 2026, Google says its dedicated generative-AI performance insights have rolled out to websites worldwide. The report covers visibility in features such as AI Overviews and AI Mode, including impressions, pages, countries, devices, and dates.

Use that report alongside standard Search Console data to watch:

• Which service and educational pages appear in generative search
• Whether impressions are growing
• Which pages have no visibility
• Whether key Minnesota service-area pages are indexed
• Whether technical problems are limiting discoverability

This is not a substitute for direct ChatGPT testing. It is a second data source that helps show whether your content is becoming visible across AI-assisted discovery.

Diagnose the gap, not just the score

The point of measurement is to decide what to fix.

Do not assume that adding schema alone will solve a visibility gap. Schema is structured data that helps machines classify a page and its entities. AEO, or answer engine optimization, improves the clarity and usefulness of information for answer systems. GEO, or generative engine optimization, is a related term focused on visibility and citation in generative answers. All three still depend on accessible, accurate, useful content and credible external signals.

Create a monthly scorecard

A concise monthly report should include:

• Total prompts tested
• Branded mention rate
• Unbranded mention rate
• Citation rate
• Accuracy rate
• Competitor and directory share of mentions
• ChatGPT-referred sessions
• ChatGPT-referred key events
• Top cited pages
• Newly incorrect facts
• Three prioritized actions for the next month

Use a three-month rolling view when the sample is small. Month-to-month volatility is normal; a rolling view makes durable progress easier to see.

A practical 90-day cadence

Month 1: Baseline and cleanup

Build the prompt set, verify analytics and conversions, check OAI-SearchBot access, reconcile business information, and document the baseline.

Month 2: Improve weak evidence

Strengthen the service pages, location pages, FAQs, author information, citations, and internal links tied to the missed prompts. Correct conflicting public data.

Month 3: Retest and attribute

Run the same prompt set, compare mention/citation/accuracy changes, review ChatGPT referrals and conversions, and keep only actions supported by evidence.

What not to do

Avoid these common measurement mistakes:

• Testing only prompts that contain your brand name
• Treating one answer as a permanent ranking
• Counting an inaccurate mention as a win
• Changing the prompt set every month
• Using screenshots without logging dates and sources
• Claiming that a crawler visit equals a recommendation
• Reporting third-party tool estimates as platform-owned data
• Promising a specific ChatGPT position
• Collecting protected or personally identifying health information in test prompts

Takeaway

Measure ChatGPT visibility as a chain: prompt coverage → accurate mentions → citations → visits → inquiries.

For a Minnesota home-care agency, the most defensible program uses a fixed set of family-intent questions, neutral testing conditions, authoritative fact checks, citation logs, GA4 referral data, and a monthly action plan. That replaces isolated screenshots with evidence you can compare over time.

Want a baseline without building the tracking system yourself? Request Total Web Advantage’s free AI Advantage Report. We will show you how your agency appears across relevant AI answers, which sources are being cited, and the highest-priority gaps to address.

Frequently asked questions

Can I see my home-care agency’s exact ranking in ChatGPT?

No single, stable ranking report exists. Track repeatable mentions, citations, accuracy, referral traffic, and inquiries across a fixed prompt set instead.

How often should I test ChatGPT visibility?

Monthly is usually practical for a full benchmark. Monitor priority prompts after material website or business-information changes, but avoid interpreting daily fluctuations as a trend.

Does ChatGPT referral traffic appear in Google Analytics?

It can. OpenAI says ChatGPT search referral links include utm_source=chatgpt.com. In GA4, review Traffic acquisition by session source/medium and landing page.

Is being mentioned the same as being cited?

No. A mention names the agency; a citation links to a source. Track both because either can occur without the other.

Does schema guarantee that ChatGPT will recommend an agency?

No. Schema can clarify entities and page content, but it does not guarantee a citation or recommendation. Crawlability, accurate information, useful content, local authority, and external corroboration also matter.