What patients ask an assistant before choosing a practice

The questions a patient types before choosing a practice

The question has the practice's constraints in it before it has a name. Typed the way people type: "family doctor in Mesa accepting new patients that takes my insurance", "pediatrician near me with same week appointments", "dermatologist in Tucson who sees adults for acne and does video visits", "clinic in Calgary with a doctor who speaks Punjabi". Once a name is in play, the follow-ups are about one practice: "is this practice taking new patients", "what insurance do they accept", "how soon can a new patient get in".

We should say what we have not measured. Nobody outside the assistant companies knows how often each of these is asked, and this site does not pretend to. What we have measured is the other side of the exchange: what a practice's own site states when an AI assistant fetches it to answer one of these questions. That is where the answer gets decided, and it is the part you control.

Two kinds of question arrive together, and the assistant treats them differently. A question about a condition gets general information and, usually, a line about seeing a clinician. A question about which practice to see gets names, and the names come from pages the assistant retrieved at the moment of asking: your site, the map listing, insurer directories, review sites. The practice that is named is the one whose page states the constraints the patient typed.

What an answer can take from a practice's pages

The facts the patient asked about, stated on a page the assistant can read and attribute to you. Not the quality of care, which no page can show it, and not the design of the site, which a text fetch does not carry. The questions above ask for six plain facts: whether you are accepting new patients, which insurance you take, how soon a new patient is typically seen, what you see patients for, whether you offer video visits, and which languages are spoken.

It is easy to state two or three of the six and miss the rest, and the missing ones tend to be the ones patients ask first. "Accepting new patients" may live in the map listing but not on the site, or on the site only as words inside a hero image. The insurance list is a PDF, or a sentence that says "most major plans". Availability is nowhere, because it changes. Telehealth is a banner from 2020 nobody has re-read. Languages appear in one clinician's bio and not in the practice's own description.

The second thing it reads for is whether you are one entity. A practice that trades as one name on the sign, another in the insurer's directory and a third on the billing statement, with a different phone number on each, is hard for a machine to hold together. When it cannot, it names a practice it can. Google's Business Profile guidelines ask for the real name and the real address, and the site, the profile and the directory listings should agree with each other to the character.

What the practice site has to state

Each row is a patient's question, the sentence on your site that answers it, and where that sentence belongs. Everything in the middle column is a fact about the practice, stated in plain text on a page a crawler receives. None of it is advice, and none of it says anything about outcomes.

The markup that carries these facts is LocalBusiness schema in its medical form: MedicalBusiness and its subtypes, of which Physician is the one for a doctor's office or clinic. It takes the same address, telephone and opening hours that Google's local business documentation lists, plus the medical properties in the table. Write it as JSON-LD, one node for the practice and one Person node per clinician, and only after the visible page says the same thing.

Stating a specialty without giving advice

A practice site states what the practice does, for whom, and how to get in. It does not tell the reader what is wrong with them or what will happen. The line is easy to hold once it is written down: "we see adults and children for asthma and manage it between visits by phone" is a fact about the practice; "asthma is usually well controlled with the right plan" is advice.

Credentials belong on the page as facts, in the form the certifying body uses: the degree, the board certification by its name, the state or provincial license, hospital affiliations, the year the clinician joined. Not "highly experienced" and not "top-rated". A named person with stated credentials can be matched against the bar in a patient's question ("a board-certified dermatologist"). An adjective cannot.

Reviews need more care here than in other trades. Your own privacy obligations decide what you may quote and with what permission, and nothing on this page asks you to quote a patient without it. What you can do without quoting anyone is state where patients review you and carry a rating count sourced from where they left it. Above all, put the six facts on the page; they are the evidence the patient's question is asking for.

What our crawl found on practice-sized sites

In our audit of 9,483 small business websites across 40 trades and 90 metros in the US, Canada, Europe and the UAE, run between late August and early September 2026, 61.5% had no question-and-answer content of any kind. For a practice the questions are already written; they are the ones the front desk answers on the phone all morning. Doctors were among the 40 trades, but we do not publish a practice-only cut of this figure.

43.1% were inconsistent about who and where they are across their own three main pages, and 34.1% had no postal address a machine could parse anywhere. A practice with a suite number, a building name and a campus address can present all three and be parsed as none. The fix is a single address line, formatted the way the postal service writes it, in the footer of every page and in the markup.

17.8% carried no schema.org markup at all, and most of the markup on the rest described the website (WebSite, WebPage, BreadcrumbList) rather than the business. A practice whose Physician markup states its address, hours, specialty and whether it is accepting new patients has told a machine what it is; markup that describes only the website has not.

What to fix first, in order

1. Settle the identity: one practice name, one address line, one main number, written identically on every page, on the map listing, and in every insurer directory you can edit. Label the billing and after-hours numbers for what they are.
2. Write the new patients page: accepting or not, for which clinicians, how booking works, how soon a new patient is typically seen, what to bring. Date it, and change it the week the answer changes.
3. Publish the insurance list as text, plan by plan, with a line for plans not listed and for self-pay. Take it out of the PDF.
4. Rewrite the services page in plain names, per clinician where they differ, with video visits and languages stated alongside. One question per heading where the patient would ask it, the answer under it. This is the answer page for your practice.
5. Add MedicalBusiness or Physician markup as JSON-LD carrying the same facts, plus a Person node per clinician with credentials as the certifying body writes them. Google's introduction to structured data says the markup must describe what a visitor can see, so it follows the page.
6. Check what a crawler receives: open the new patients page with scripts off, confirm the text is in the HTML, and run the crawler access check. A patient portal's security layer can put a challenge page in front of the whole site without anyone deciding to.

What not to do

Do not write patient reviews, and do not reward patients for them without saying so. Google's review snippet documentation excludes fake reviews and undisclosed incentivized ones, and a practice has more to lose than most from being caught doing it.

Do not stuff the page with place names and specialties. "Family doctor Mesa, Mesa family medicine clinic, primary care Mesa" states nothing a patient asked. Google does not read the keywords meta tag, as its list of supported meta tags shows, and an assistant reading prose is looking for the sentence that answers the question.

Do not make promises: not about outcomes, not about wait times to the day. And do not accept one from anyone about being named by an assistant. No agency has a lever inside ChatGPT, Claude, Gemini or Perplexity. What can be done is to measure how often your practice is named on a fixed set of patient questions and fix what the measurement shows; how we measure it is written down.