Opens in a new tab

How Patients Use AI Search to Choose a Physio

and how your clinic gets named

By , Quarc Digital · 27 August 2026

Patients now put health questions to AI search tools, including which physiotherapist to see, and the tools name clinics whose sites answer the exact question. A physiotherapy clinic gets named when it publishes a specific, accurate answer for each condition it treats, in the city where it treats it.

Short answer: AI search tools match a patient’s question to the clinic page that answers it most directly. A clinic gets named when it has a specific, accurate page for each condition it treats, listings that match its website, and a way to check the result. Ask ten patient-style questions in four tools to see where you stand.

A physiotherapist guiding a patient's raised arm during a shoulder assessment on a treatment table

Photo: Marc Zeman on Unsplash

01 — The data

What are patients asking AI search about physio?

AI search means a chatbot or answer engine, such as ChatGPT, Gemini, Perplexity or Google’s AI Mode, that replies in sentences instead of a list of links. Patients type questions like “physio for sciatica near me” or “who treats a rotator cuff injury in Langley.” The tool reads the question, looks up pages, and writes one answer.

Health questions are already a large share of what people ask. Pew Research Center surveyed 3,488 U.S. adults in June 2026 and found that 34% had used an AI chatbot for at least one of eight health reasons.[1] Fifteen percent used one to help decide whether to go to a doctor. KFF’s March 2026 poll put past-year use at 32% and found that 16% used AI to help decide whether to see a doctor or seek care.[2] Use falls with age: Pew found 44% of adults aged 18 to 29 use chatbots for health, against 17% of those 65 and older.

A woman seated on a sofa typing a question into her phone

Photo: Vitaly Gariev on Unsplash

Canadians look similar. A Liaison Strategies poll of 1,526 adults, fielded May 4 to 16, 2026, found that 46% had used an AI chatbot for medical advice in the past 12 months.[3] That was 65% of Canadians aged 18 to 34 and 26% of those 65 and older.

Choosing a provider is a narrower question, and we found no independent survey that measures it for healthcare. BrightLocal’s 2026 consumer survey shows the wider shift for local businesses: use of ChatGPT and other generative AI for local recommendations rose from 6% to 45% in a year.[4] Health-information use is the closest proxy for clinics, and the polls above suggest it is already common.

02 — The mechanism

How does AI search decide which clinic to name?

AI search retrieves first and writes second. It finds passages that match the question, then composes an answer from them. Your clinic can only be named if one of its passages is found.

Perplexity’s February 2026 release shows how that matching works. The company published two embedding model families, pplx-embed-v1 and pplx-embed-context-v1, and describes them as built for web-scale retrieval.[5] An embedding model turns a piece of text into a list of numbers so that texts with similar meaning land close together. That is why a page headed “Shoulder pain and injury physiotherapy” can match a question about a rotator cuff injury without containing the question’s exact words.

The second model, pplx-embed-context-v1, targets document chunks where surrounding context matters. Perplexity’s API documentation lists it for chunks that share context, such as paragraphs from the same article.[6] Our read is that each paragraph is then judged with the rest of its page in view. Perplexity’s paper reports testing on search scenarios built from 1 billion production web pages.[7] The material we reviewed does not say whether these models power Perplexity’s live answers, and it says nothing about ChatGPT or Gemini. Elsewhere, search vendor Vespa quotes Perplexity as saying the early stages of its ranking use “lexical and embedding-based scorers,” with heavier rerankers last.[8] That is a secondhand quote from October 2025, so read the release as a description of how modern retrieval is built, not as a ranking rule. Perplexity’s own guidance for chunking documents says to split structured pages on headings or paragraph boundaries[9], which is one more reason to give each section a clear heading. Our guide to local SEO, AI Overviews and ChatGPT covers how those engines differ for service businesses in general.

A person typing into an AI chatbot on a smartphone, with the prompt "What can I help with?" on screen

Photo: Zulfugar Karimov on Unsplash

Google says the same thing from the other direction. Its documentation states there are no additional requirements to appear in AI Overviews or AI Mode, and no special schema is needed.[10] A page only needs to be indexed and able to show with a snippet. Google also says these features may use “query fan-out,” issuing several related searches across subtopics before writing an answer. For a physio clinic, one question about sciatica can turn into lookups on the condition, the city, the cost and how to book. A site with a page for each of those subtopics gives the tool more to link.

Selection is tight. SOCi, a local marketing software vendor, asked ChatGPT, Perplexity and Gemini to recommend 10 businesses per industry and market. A location counted as recommended if it landed in the first five. ChatGPT recommended 1.2% of brand locations, Perplexity 7.4% and Gemini 11.0%, compared with 35.9% for the Google local 3-pack.[11] SOCi’s study page says the analysis covered 350,000 business locations and 3.2 million AI queries.[12] It focused on multi-location brands rather than clinics, so treat the percentages as direction only. Our article on AI search versus the map pack weighs what those numbers mean for where a business should invest.

03 — The problem

Why does a generic Services page give AI nothing to cite?

A generic Services page gives AI search a weak match for every condition it lists. The typical page names ten services in ten lines, and each line says little beyond “physiotherapy for back pain” or “sports injuries.”

This is our reading of how passage matching works, not a measured result. A question about sciatica is closest in meaning to a passage that is about sciatica. A page that mentions sciatica once among forty other topics gives the tool a diluted passage, and a competitor’s dedicated page gives it a sharp one. The same logic applies to the paragraph level: a paragraph that begins “We also treat this” has no clinic, condition or city in it, so it stands on nothing once it is lifted out of the page.

04 — The fix

What should a physio clinic publish?

A physio clinic should publish one page per condition or treatment that it wants to be found for, starting with the three it sees most. Each page answers the question a patient would type. Here is an illustrative layout for a sciatica page. It is a template, not a client example, and your clinicians have to supply every bracketed item.

  1. First sentence: “[Clinic name] in [city] treats sciatica with [the approaches your clinicians use].” It names the clinic, the condition and the place before anything else.
  2. Who treats it: the title and registration of the clinician, written as a plain fact.
  3. What the first visit involves: what happens in the assessment at your clinic, in your clinicians’ words.
  4. When to see a doctor first: the situations in which your clinicians refer patients out.
  5. A typical course of care: visit counts and timelines only if your own records support them. Skip the numbers otherwise.
  6. Practical details: hours, location, how to book, whether you direct bill, and what a first visit costs if you publish prices.

Three habits keep each page useful once an AI tool pulls a paragraph out of it. Open each section with the clinic, condition and city rather than “this” or “we.” Keep one condition per page. Give a specific number or step wherever a vague adjective would otherwise sit.

An empty physiotherapy treatment room with a treatment table, a window and an equipment trolley

Photo: Nightingale Home nurse on Unsplash

05 — The checks

What else do AI tools check before they name you?

AI tools lean on facts they can check elsewhere, and wrong details cost clinics bookings. An engine has to match your name, address and phone number to one business before it can name you, a step our guide to how service businesses get cited in AI Overviews and ChatGPT walks through. A rater8 survey of 992 U.S. adults (April 2026) found that 66% of the 465 who had used AI to research a provider said they had met incorrect provider information.[13] rater8 sells reputation-management software, so read this as vendor data. TechTarget’s coverage lists wrong addresses, phone numbers, insurance details and office hours as examples.[14] Sixty percent said they trusted the AI summary without checking it further. A clinic with a stale phone number on a directory listing can be named correctly and still lose the patient.

So the basics matter. Google lists keeping your Business Profile information up to date among the SEO items relevant to its AI features.[10] Check your name, address, phone number, hours and services on your Google Business Profile, your website and every directory that lists you. Reviews matter too. SOCi’s analysis found that locations ChatGPT recommended averaged 4.3 stars, again for multi-location brands.[11] We would not set a star target from that, but a clinic with a low rating and unanswered reviews is not helping its case.

06 — The rules

What does BC’s marketing standard change for these pages?

Every condition page counts as marketing, so BC’s marketing standard covers it. The College of Health and Care Professionals of BC (CHCPBC), which regulates BC physiotherapists, issued a Marketing, Advertising, and Promotion practice standard that took effect April 1, 2026.[15] Its scope takes in websites, social media, directory listings and the use of testimonials.

Four parts of it shape this work:

  • A licensee is responsible for marketing prepared by an agency or other third party and must ensure by contract that they are fully informed of it (preamble).
  • A licensee may not make false, misleading, unsubstantiated or disparaging claims of superiority over other providers (clause 2.1), so “best physio in Langley” is out.
  • Health information has to match current, evidence-informed practice (clause 3), and messaging cannot be sensational, aggressive or fear-based (clause 4.2).
  • A licensee must monitor third-party content that references their services, such as online reviews and directory listings, and promptly ask the responsible party to correct or delete inconsistent information (clauses 7.1 and 7.2).

The last clause turns listing accuracy into a compliance duty as well as a booking problem. This article is marketing information, not legal advice. Check the standard itself, or ask the College, before you publish.

07 — The limits

What do these numbers not tell you?

The surveys support the direction of travel, not the size. Pew and KFF measure health information, not clinic choice. The Liaison release we read does not describe a provider-search question. BrightLocal and SOCi measure local businesses broadly, and SOCi covers multi-location brands. We found no independent or peer-reviewed study of AI search use for physiotherapy, and the physio-specific statistics in vendor marketing cited no source. The one provider-research survey we found, from reputation-management vendor rater8, gives no margin of error, so we use only its figures on incorrect information and leave out its adoption numbers.

What the evidence does support is narrower and still useful. Many patients already ask AI tools about their symptoms and about whether to seek care, use is highest among younger adults, and the tools sometimes return wrong provider details. None of it tells you how many of your own patients arrive this way.

08 — The test

How can you check where your clinic stands?

Run the test yourself this week. It takes about an hour.

  1. Write 10 questions a patient would ask: five conditions you treat, each paired with your city, plus questions about cost, booking and direct billing.
  2. Ask each question in ChatGPT, Gemini, Perplexity and Google’s AI Mode. Ask twice, because answers vary between runs.
  3. Record whether your clinic is named, whether your site is cited, and whether the details are right.
  4. Fix wrong details first, starting with the one that appears most often.
  5. Publish a condition page for the top three gaps, then rerun the test in a month.

For a faster first read on one page, run it through our free AI Visibility Check. It scores a condition page or your homepage on whether AI tools can crawl, understand and cite it, and it needs no email. It checks how the page is built, so it complements the test above rather than replacing it.

Measuring the result is harder than running the test. Our article on why Google still won’t count AI clicks covers how to track visits in your own analytics.

Quarc Digital builds condition pages and Business Profile clean-ups for physiotherapy clinics; see our physiotherapy marketing work, how answer engine optimization fits with local search, or book a consultation.

09 — Answers

Frequently asked questions

No special markup is required for Google’s AI features. Google’s documentation says there is no special schema.org structured data you need to add, and that a page needs to be indexed and eligible for a snippet. Schema can still describe your clinic accurately, but it is not a lever for AI visibility.

The fundamentals are the same and the unit of work is smaller. Google says SEO best practices remain relevant for its AI features. What changes is that the match happens at the passage level, so each paragraph needs to name its clinic, condition and city.

Testimonials fall inside the CHCPBC standard’s scope, which means they must meet the same accuracy and honesty requirements as the rest of your marketing. The standard text we read names testimonials in that scope and does not contain a separate testimonial ban. Confirm current wording with the College before you rely on that.