How to Test If AI Can Find Your Business

Updated May 2026 • 9 min read

Most local business owners have never directly tested how AI assistants answer the questions their actual customers ask. They assume they are findable because they rank well on Google, or because they spent money on SEO, or because they have a Google Business Profile with five-star reviews. Then they ask ChatGPT for the best veterinary clinic in West Columbia at 9 p.m. on a Sunday — the exact moment a pet owner is panicking about a sick dog — and discover their practice is not in the list.

This guide is a complete, reproducible test harness you can run in 90 minutes the first time and 30 minutes per month after that. It tells you what each of the four major assistants currently says about your business, where the misses are, and which fixes will move the needle fastest. We will use a West Columbia veterinary practice as the running example because urgent on-the-go searches are the highest-stakes use case for AI-citation visibility.

This is the cheapest, highest-signal audit a local business owner can run on themselves — and it is the one most owners have never done.

Why You Cannot Skip This Test

Traditional rank trackers cannot see AI surfaces. SEMrush, Ahrefs, BrightEdge, and similar tools were built to scrape Google's traditional ten blue links. The conversational answer at the top of an AI Overview, the cited sources in a Perplexity response, the bulleted recommendations from ChatGPT — none of those are in the rank-tracker dataset.

The only reliable way to know whether AI surfaces cite your practice is to ask them directly, manually, on a regular cadence. The good news: the test takes less time than most owners spend reviewing a single SEO report. And for a veterinary clinic, where emergency searches dominate the high-margin volume, the test is essentially a competitive-position check on the exact channel a panicking pet owner uses first.

The 12 Prompts Every Business Should Test

Use the same 12 prompts every month. Consistency matters more than cleverness — the goal is to track the trend, not to test a different question each time.

Substitute your actual service, city, and neighborhoods into the bracketed slots. The examples use a West Columbia veterinary practice for illustration.

Category A: Direct Buyer-Intent (4 prompts)

1. "Who are the top [veterinary clinics] in [West Columbia, SC]?"
2. "I need a [reliable vet] in [West Columbia] — any recommendations?"
3. "Best [veterinarian] near [Cayce SC]"
4. "Can you recommend a [low-stress veterinary practice] in the [Columbia metro area]?"

Category B: Scenario-Based (4 prompts)

5. "My [dog is vomiting and lethargic] and I have [a senior pet that hates car rides] — who in [West Columbia] does same-day urgent appointments?"
6. "I'm a [first-time cat owner in Cayce SC] looking for a [feline-only or cat-friendly vet] — who should I contact?"
7. "What [veterinarian in West Columbia] handles [exotic pets] with weekend availability?"
8. "I'm comparing [veterinary clinics in West Columbia and Cayce]. What should I ask before choosing one for my new puppy?"

Category C: Comparison and Validation (4 prompts)

9. "Tell me about [Three Rivers Veterinary] in [West Columbia SC]."
10. "Compare [Three Rivers Veterinary] and [Competitor A] in [West Columbia]."
11. "Is [Three Rivers Veterinary] in [West Columbia SC] a legitimate AAHA-accredited practice?"
12. "What services does [Three Rivers Veterinary] offer and where do they operate?"

Common mistake: Asking the same prompt phrased four different ways in one session. You learn nothing additional. Stick to the 12 different question types.

Time investment: 5-7 minutes per assistant per prompt set. Roughly 25 minutes per assistant for all 12.

The Four Assistants to Test

Test the same 12 prompts in this exact order on each platform. Free accounts are fine for all four.

1. ChatGPT (chat.openai.com)

Start a new conversation. Enable web search if not on by default (GPT-4o and later models search automatically for current information). Paste each prompt one at a time. Watch for:

• Whether your practice is named
• What facts are attributed to you (correct? outdated? hedged?)
• Which sources are cited (typically shown as numbered links)
• How many competitors are named ahead of you

2. Perplexity (perplexity.ai)

Perplexity shows sources prominently. This is the easiest surface to learn from — you can see exactly which pages it pulled from, which directly tells you what content is doing the work. Run the same 12 prompts.

3. Google AI Overviews

Open Google in a private/incognito window (to avoid personalized history influencing results). Type each prompt as if it were a search query. The AI Overview appears at the top for most informational and recommendation queries. Capture whether your practice appears in the overview, in the cited sources beneath it, or only in the traditional ten blue links below.

4. Claude (claude.ai) or Gemini (gemini.google.com)

Pick one. Run the same 12 prompts. Claude with web search enabled tends to behave similarly to ChatGPT; Gemini integrates Google's local-business data more aggressively. Either provides a useful fourth data point. For veterinary practices specifically, Gemini sometimes pulls richer Google Business Profile data than the other three, so testing it is worth the extra five minutes.

Common mistake: Testing while logged into a personal account that has chatted with the assistant about your practice before. The assistant's memory of past conversations can bias the result. Use a private window or a fresh test account.

Time investment: 90 minutes total for all four assistants, all 12 prompts, plus logging.

How to Log the Results

A spreadsheet with four sheets — one per assistant — is enough. Each sheet has 12 rows (one per prompt) and these columns:

• Prompt # — 1-12, matching the list above
• You appeared? — yes / no / partial
• Position — first / middle / last / footnote / cited-source-only
• Facts cited — short list of what the assistant said about you
• Facts wrong — short list of inaccuracies (NAP errors, outdated services, wrong service area, missing AAHA accreditation, expired Fear Free certification)
• Competitors named ahead of you — list them
• Sources cited — which URLs (your own, third-party directories, news articles)
• Notes — any other observations (Sunday-evening query bias, mobile vs. desktop, urgent-care language detected)

The first run takes 90 minutes. Subsequent monthly runs take 30 minutes because you are comparing to a template, not building one.

Reading the Results: Four Patterns and What They Mean

Pattern 1: "Not Appearing Anywhere"

The most common starting state. You are not mentioned in any of the 12 prompts on any of the 4 assistants. Sources cited are competitor practices, Yelp, Vetster, Better Business Bureau, and PetMD location pages.

Diagnosis: Entity-verification gap. AI assistants cannot confidently identify you as a real, locally-active practice. The fix is foundational: complete Google Business Profile (including categories like "Veterinarian", "Animal hospital", "Pet boarding service" if applicable), fix NAP across 12-15 platforms, add Veterinary Care schema, build reviews to a credible volume, list AAHA accreditation and Fear Free certifications publicly.

Expected time to first citation: 60-120 days with disciplined foundation work.

Pattern 2: "Appearing on Validation Prompts, Not on Recommendation Prompts"

You appear when someone asks specifically about your practice (Category C — prompts 9-12) but not when someone asks for a recommendation (Categories A and B — prompts 1-8).

Diagnosis: Authority gap. The assistant knows you exist but does not consider you a top-tier answer. Usually means thin or generic content, weak hyperlocal coverage (no mention of West Columbia / Cayce / Springdale neighborhoods), or unrepresentative reviews that don't mention specific services.

Fix: Conversational long-form service pages (urgent care, dental, exotic pets, end-of-life), hyperlocal articles about West Columbia pet-ownership patterns, review-content optimization. Plan 6 months.

Pattern 3: "Appearing on Some Assistants, Not Others"

Strong on Perplexity and Google AI Overviews; absent on ChatGPT and Claude. Or the inverse.

Diagnosis: Platform-specific source bias. Perplexity weights structured data and recent web pages; ChatGPT weights training-data corpus plus its own web index. If your authority lives mostly in your Google Business Profile and you have a thin website, you may appear in Google AI Overviews but not in ChatGPT.

Fix: Strengthen the surface where you are weak — usually means more website-level content with proper schema and named-author bylines (DVMs and credentialed veterinary technicians) for the assistant that crawls broadly. Plan 3-6 months.

Pattern 4: "Appearing with Wrong Information"

You appear, but the facts the assistant cites are stale or wrong — old phone number, outdated hours, AAHA-accredited status that expired, services you stopped offering (boarding, grooming) still listed, an associate veterinarian who left two years ago named as a staff member.

Diagnosis: NAP-consistency or content-freshness gap. The assistant has built its picture of you from outdated sources.

Fix: NAP audit, Google Business Profile refresh, About-page rewrite (with current team only), schema update with current dateModified. For veterinary practices, also re-verify AAHA-accreditation status and update the team-credentials page. Usually resolves in 30-60 days as AI surfaces re-index.

Turning Test Results Into a Fix List

After each monthly run, look across the 48 data points and answer four questions.

1. What single fact got the most assistants wrong? Fix that first — it is usually a single source somewhere (an old directory listing, a stale GBP attribute, a 2020 vet-directory entry that never got updated) and one fix propagates across surfaces.
2. Which competitor appeared most often in your missed citations? Visit their site and read whatever the assistants cited from them. Reverse-engineer what made it citable. Frequently it is a long "What to Expect at Your First Visit" page or an AAHA-accreditation explainer.
3. Which prompt category is weakest? If Category A (direct buyer-intent) is weakest, you need foundational visibility work. If Category B (scenario) is weakest, you need conversational long-form content (urgent-care scenarios, senior-pet scenarios, anxious-pet scenarios). If Category C (validation) is weakest, you need an AI-ready About page.
4. What changed from last month? Track the citation count trend, not just the absolute number. A 3-citation increase month-over-month is the signal that the fixes are working.

Common mistake: Treating each prompt result as a separate to-do. The real pattern is usually one root cause showing up in 4-5 prompts. Cluster the symptoms before assigning the fixes.

What to Look for in the Sources Each Assistant Cites

The sources column of the log is the most actionable data the test produces. Five categories of source — each tells you something different.

• Your own website — the goal. Indicates AI recognizes you as the authoritative answer.
• Your Google Business Profile — second-best. Indicates AI verified you as a real local practice, even if it could not find depth on your website.
• Third-party directories (Yelp, Vetster, AAHA's hospital locator, PetMD, BBB) — a citation that bypasses your website entirely. Useful, but means your website is doing less work than it should.
• Local news or community publications — strong trust signal. A Free Times or Cola Daily piece about a practice's community initiative often becomes the difference between Pattern 2 and getting cited on recommendation prompts.
• Competitors' websites (citing you in a "see also" or "alternatives" list) — extremely strong signal. Rare, but indicates entity-level authority.

If every citation is coming from third-party directories and none from your own site, your website is failing as an authoritative source — and that is fixable in 6 months of focused content work.

Tracking the Trend Over Time

The single most important metric is total citation rate across the 48 data points, tracked monthly.

Months 1-3: Baseline

Whatever number you start at, accept it as the truth and document it. Do not try to "fix" anything between runs; just baseline.

Months 4-9: Foundation

Work the fix list from your earliest runs. Expect citation rate to rise slowly and unevenly — 1-3 new citations per month is healthy progress.

Months 10-12: Compounding

By month 9-12, well-executed work usually produces 5-10 new citations per month. The flywheel kicks in once entity authority is established across surfaces.

Year 2 and beyond:

Maintenance cadence. Run the test monthly; expect 4-8 new minor citations per month and an occasional major shift (a new platform launches, a competitor publishes a viral piece, a directory listing breaks). The monthly test catches these inside 30 days instead of inside a year.

What This Test Cannot Tell You

Honest caveats:

• Personalization noise. AI assistants increasingly personalize answers. A test run in a logged-out session approximates the average user, but two real pet owners on the same street can get different answers. Run the test from a private window to minimize bias, but expect some variance.
• Geographic variability. Some prompts return different results based on the IP location of the request. If you run the test from outside the Columbia metro, results may differ from what a local pet owner would see.
• Mobile and voice-assistant blind spot. The test covers four text-based assistants on desktop. Voice queries through Siri ("Hey Siri, find a vet open right now") and mobile-first Gemini responses on Pixel devices may surface different results — and for urgent veterinary searches, mobile and voice are where the highest-stakes queries actually happen. A separate mobile and voice-test protocol is worth adding once your text-citation rate is solid.

None of these caveats invalidate the test. They mean the test is a strong signal, not a final answer. Trends over months are more reliable than any single result.

The Bottom Line

A 90-minute test, repeated monthly, is the most accurate, lowest-cost AI-visibility audit a local business owner can run. It turns "I think we're showing up" into "we appeared in 17 of 48 prompts this month, up from 11 last month, with strongest improvement on scenario-based queries." That kind of measurement is what separates practices that compound visibility over the next five years from practices that guess.