A dental practice should show up in all four engines, but in a specific order of effort: Gemini and Google AI Overviews first for local intent, ChatGPT with search next for treatment questions, Perplexity for researched comparisons, and Claude for long conversational prompts. All four share one fixed question set, one city and one run window, so the numbers stay comparable.

The order exists because the engines are not copies of each other. Each one retrieves from a different place and prefers a different kind of source, so the same practice can be named in one and absent from another.

Engine 1

ChatGPT

The same 40 questions, every answer recorded.

Engine 2

Gemini

The same 40 questions, every answer recorded.

Engine 3

Perplexity

The same 40 questions, every answer recorded.

Engine 4

Claude

The same 40 questions, every answer recorded.

Take a practice that ranks well in Google and has a careful website. It is usually visible in Gemini for local questions, often in ChatGPT for treatment questions, and hard to pin down in Claude, where answers arrive without a source name.

Why The Engines Behave Differently

Four Differences That Matter

Search infrastructure differs, so the same page can be reachable in one engine and unseen in another. Source selection differs, because a directory entry and a treatment page are not equally useful for a question. Citation behaviour differs, and so does personalisation.

Each vendor documents the crawler that fetches pages for its search results, and those documents appear below where they matter. A blocked crawler leaves a page that no answer can draw on. The article on which AI crawlers to allow and which to block covers that decision.

ChatGPT With Search

What It Draws On

ChatGPT searches the live web when a question needs current or local information. OpenAI documents the crawler responsible as OAI-SearchBot, and states that a site which disallows it will not appear in ChatGPT search answers, though it may still appear as a navigational link (OpenAI's crawler documentation). The implication for a practice is simple: ChatGPT works from your live pages.

What It Favours

Observed tendency: ChatGPT tends to build answers from short, self contained passages. An answer naming one or two practices and then describing what each does is common, and a ranked list of ten is not. That favours pages answering one question on their own.

Personalisation cuts both ways. A signed-in patient who has already described a chipped tooth may get a different answer than your cold measurement run. That is one reason to measure from a clean session.

The wording you test with decides what you learn, so keep it fixed and patient shaped. The article on 40 patient questions to ask ChatGPT gives the set grouped by intent, with the city placeholder and the rules that keep a monthly run comparable.

Gemini and Google AI Overviews

The Index Advantage

Gemini draws on Google Search, and AI Overviews are built on the Google index. That is the single largest structural difference between this engine and the other three: a page already in Google's index is already reachable.

Google publishes its own guidance for these features, and two points in it matter for a local practice. No special markup is required for AI features, which means the work is good publishing rather than a technical project. And Google says it does not use llms.txt or similar machine readable files to decide inclusion. The same guidance points to a generative AI performance report in Search Console, which is the honest place to watch how AI features affect your pages.

The Business Profile Advantage

Local answers lean on Google data, and that data sits in your Business Profile: category, services, hours, address, photos and review activity. Google's own Business Profile help pages set out what a complete profile needs. Where your website and your profile disagree, you have built the confusion that makes an engine pick another practice.

If your patients ask for a dentist near them on a phone, this is the surface to watch first. The article on Google AI Overviews for dentists covers the measurement side.

Perplexity and Claude

Perplexity Is a Research Engine With Visible Sources

Perplexity is built around research answers, and its citation behaviour is part of the product: the answer carries numbered links back to the pages it used. It documents its own search crawler separately from the user triggered one (Perplexity's crawler documentation). The pages it leans on tend to make a specific, checkable claim: a price range with its variables, a treatment explained on one page, a directory listing whose hours are current.

For a practice that means Perplexity rewards what a patient rewards, which is specific and current. It is also the engine where directories do most of the work, because a directory page is a compact, structured claim about a practice.

Claude Retrieves Without Always Naming The Source

Claude behaves differently in a way that frustrates anyone who counts only mentions. It retrieves and synthesises, and it frequently answers without naming the page it drew from. Anthropic documents a dedicated search crawler for Claude, separate from the one that collects training data (Anthropic's crawler documentation). Your pages are being read. The credit simply does not always come back.

Observed tendency: Claude tends to favour passages that read as complete explanations rather than as bullet lists. Write the treatment page so that one paragraph, lifted on its own, is a correct and useful answer. If being named matters to you, put the practice name inside that paragraph.

Which Engine Matters For Your Patients

The order at the top of this article is a sensible default. For your practice it may be wrong, and there are three ways to find out.

  1. Ask at the front desk. For thirty consecutive new patients, record how they heard about you, with options for a search engine, an AI assistant, a referral, a map and a plan. The share naming an assistant is your first answer.
  2. Read what you already have. Search reporting tells you which queries bring people in, and server logs tell you where sessions came from. A rise in referrals you cannot explain is often a rise in assistant use.
  3. Map your questions onto engines. Local availability and opening hours sit with Gemini and AI Overviews. Treatment comparisons sit with ChatGPT and Claude. Cost research with a demand for sources sits with Perplexity.

Then weight your reading time by that map. A practice whose patients ask who is open today should watch the local surface. A practice competing for implants should watch treatment questions. Context matters too: a patient on a phone at nine in the evening asking for the nearest practice is a different journey from someone comparing three providers at a desk.

Measuring Across All Four Without Losing The Plot

What To Capture For Every Run

Four engines, one question set, one city, one run window. The full protocol is in the article on measuring AI visibility for a dental practice. Score every run the same way, in three states: named and recommended, named without an endorsement, or absent. The third is the gap, the second is the opening, and the first is the result.

Swipe the table sideways to see every column.

EngineWhat It Draws OnExtra Field To RecordWhat a Good Answer Looks Like
ChatGPT with searchLive pages, often the one with the cleanest passage on the questionWhether the answer carried a link or the page nameYour practice named with a detail about how you handle the case
Gemini and AI OverviewsThe Google index, plus Business Profile and directory dataWhether your profile facts were visible in the answerYour practice named with address, hours and a matching service
PerplexityA short list of pages treated as evidence, often directoriesThe cited source, so you know which page wonYour practice named with a source the reader can open
ClaudeRetrieved pages, synthesised into an explanation that omits the sourceWhether your wording appears in the answer at allA correct explanation describing your practice, name or no name

Capture the same five fields for every run whatever the engine: question, engine, date and time, the answer text, and whether you were named. Keep the answer text, because without it you cannot check the scoring later.

Two rules keep the four numbers comparable. Do not delete the runs you did not like, because an answer set with holes in it will move for reasons that have nothing to do with you. And when a model changes substantially, re-baseline and write the date down.

drsmiles.ai measures all four engines on one fixed question set for exactly this reason: a practice cannot tell whether it is improving or reading a different engine on a different day.

  • The question set is written down and the wording has not changed since the first run.
  • Every run records the engine, the date and time, the city, the answer text and whether we were named.
  • Scoring uses three states and one rubric across all four engines.
  • No run is deleted after the fact, including the ones that were inconvenient.
  • Model updates are dated in the log so a baseline can be reset honestly.
  • Crawler access is checked in robots.txt so every engine can fetch the pages.

For the short version of what this work involves, the dental AEO page sets out the process, and the blog index holds the rest of the library in order.