Product design collectiveWhat an AI Visibility Audit actually looks like
Most audits are a 40-page PDF nobody reads. Ours is a short, quoted, prioritised document: here is the shape of it, using an anonymised example — a family dental practice.
Note: this is an illustrative worked example, not a real client. The business, competitor names and figures below are representative, chosen to show the method. A real audit quotes the real answers, verbatim, on the day it was run.
1. The buyer questions
We start from the questions your customers actually type, not keywords. For a general dental practice we'd run 25–40 of these; here are five.
- “Who are the best family dentists in town?”
- “I just moved here — who should I see for a cleaning?”
- “Are there dentists nearby that take my insurance and do same-day crowns?”
- “Best dentist near me for kids?”
- “Do I need a dentist or an oral surgeon for a broken tooth?”
2. What the engines said
We ask each question in a fresh session, with no account history, and record the answer verbatim — the same answer a customer would see. A sample of what came back:
ChatGPT · Google AI Overviews
“Who are the best family dentists in town?”
“Here are a few well-regarded options: Silver Sage Dental, Galena Family Dentistry, and Mt. Rose Dental Group…”
Not named
Perplexity
“Best dentist near me for kids?”
“For pediatric and family dentistry, parents often mention Peavine Pediatric Dentistry and Galena Family Dentistry (cited from a local parents' blog and Yelp)…”
Not named · competitors named
Gemini
[Practice name] — do you take [insurance]?
“[Practice name] is a dental office on the south side of town. Hours and accepted insurance are listed on their website…” (and then invented a Saturday opening time that the practice does not have.)
Named — but with an accuracy error
Illustrative answers. In a real audit each is captured with the engine, the date, and the exact response text.
3. The scorecard
| Engine | Named for category? | Named for “near me”? | Facts accurate? |
|---|---|---|---|
| ChatGPT | No | No | — |
| Google AI Overviews | No | No | — |
| Perplexity | No | No | — |
| Gemini | Yes | No | Wrong hours |
| Claude | No | No | — |
4. Why — the diagnosis
- The homepage is invisible to crawlers. The practice site is a single-page app: the raw HTML carries 190 readable words. ChatGPT, Perplexity and Claude, which don't run scripts, see essentially nothing.
- Nothing states the category plainly. The H1 reads “Dentistry, reimagined.” Nowhere does the page say “family dentist in this part of town.”
- Thin third-party signal. 41 Google reviews, no consistent listings on Healthgrades or Zocdoc, and no local press mentions — while two competitors are cited from third-party sources the engines trust.
- Stale facts in the wild. An old address and a Saturday-hours claim circulate in two directories and were repeated by an engine.
5. The fix list (prioritised)
- Ship static, readable pages. Serve real HTML with a sitemap; keep the current design. HIGH IMPACT · WEEK 1
- Answer-first service page. A “Family Dentistry in [Town]” page whose first 100 words state what, where, for whom, hours, insurance and price range.
- Consistent business facts. One canonical set of name/address/phone/hours pushed to Google Business Profile, Bing Places, Healthgrades and Zocdoc.
- Review flow. A simple post-visit ask that grows recency and volume — reviews are what engines actually read.
- One earned mention. A local-press or community item that names the practice and its location.
6. The three ready-to-publish assets
- An answer-first rewrite of the homepage and the family-dentistry page — copy the client can paste in.
- An FAQ set (12 questions) written the way patients ask them, in language an engine can quote.
- A business fact sheet — name, address, phone, hours, services, insurance, price range — formatted for directory and profile entry.
Want this run on your business?
Start with the free mirror test — three real questions, the answers verbatim, and the competitor named instead of you.
Get the free mirror test →