Is AI search actually driving patient bookings yet?

Yes, but unevenly. Younger urban patients (25-40) increasingly start with ChatGPT or Perplexity when researching a service. Older patients still default to Google. For categories like pelvic-floor physiotherapy, mental health, naturopathic medicine, and aesthetic services, we are seeing 10 to 25 percent of new-patient inquiries originating from AI-mediated discovery. For dental and physiotherapy, lower but rising.

What is llms.txt and do I need one?

llms.txt is an emerging convention (proposed by Jeremy Howard in 2024) where you publish a plain-text file at /llms.txt summarising your site, its key pages, your positioning, and the topics you cover well. LLMs that respect it use the file to understand the site at training and retrieval time. It is not yet universally read, but the AI providers that read it weight the data heavily. Cheap to add, asymmetric upside.

Does Google's AI Overviews use the same signals as regular search?

Mostly, plus a few extras. AI Overviews favour content that has clear authoritative framing, includes specific numbers and examples, and exists on sites with strong E-E-A-T signals. The same on-page work that helps traditional rankings tends to help AI Overviews appearance. The unique addition is structured content (clean h2/h3 hierarchy, lists, tables) that the AI can extract directly into the overview.

How does an AI decide which clinic to recommend?

From what we can observe, three signals dominate: (1) content depth on the specific service and location, (2) brand mentions on authoritative third-party sites (BC health publications, news, partner clinics, registered association directories), and (3) structural clarity of the site so the AI can extract recommendation-worthy snippets. SEO basics matter, but the unique AI-recommendation factor is breadth of mentions across credible sources.

What is the highest-ROI move for AI search visibility right now?

Publish an llms.txt file, structure your service pages with clean h2/h3 hierarchies and FAQ schema, and start earning brand mentions on third-party BC publications and partner clinic blogs. The first two are cheap and immediate. The third is a slow build but creates the citation graph AI systems read.

AI search visibility for clinics: Perplexity, ChatGPT, Gemini

In 2026, a meaningful share of patients researching healthcare services in Vancouver start in ChatGPT, Perplexity, Gemini, or Google’s AI Overviews. They ask “best naturopath in Vancouver for hormonal issues” or “physiotherapy clinic in Mount Pleasant that takes ICBC” and get an AI-summarised recommendation, sometimes with one or two clinic names mentioned by the model.

A clinic that gets named in those AI responses is in front of high-intent patients before they ever look at a search results page. A clinic that does not is invisible to a growing slice of the market.

This article is what we know about how AI search recommends clinics and the content patterns that work.

What changes in AI-mediated search

Three differences from traditional Google search:

The AI gives an answer, not a list. A user asking “what is the best physio clinic for runners in Vancouver” gets a recommendation, not a 10-blue-link page. Whoever the model recommends captures most of the consideration.
The model decides which sources to read. Traditional SEO optimises for search rankings. AI search optimises for being one of the trusted sources the model pulls from when constructing its answer.
The patient gets a synthesis, not a brand exposure. The AI may paraphrase content from your site without showing the user your clinic name. Visibility happens differently than impression-driven traditional search.

This is not a future scenario. It is the current behaviour of ChatGPT, Perplexity, Gemini, and Google’s AI Overviews in Vancouver healthcare queries.

What signals AI systems read

We do not have direct access to how each model ranks clinic recommendations. From observed behaviour, three signal categories seem to dominate:

1. On-page depth and structure

AI models extract structured information well. Pages that win the extraction game share traits:

Clean h2/h3 heading hierarchy that maps to a logical content structure
Lists, tables, and step-by-step breakdowns that the AI can pull cleanly
FAQ sections with explicit question-answer markup
Specific numbers, ranges, and examples instead of marketing-speak
Clear authorship (named author, credentials, publish date)

A clinic with a service page that has 1,500 words of well-structured content beats a clinic with a 300-word page on the same service in AI recommendations.

2. Brand mentions on authoritative third-party sites

This is the lever most clinics miss. AI models read your site for content but read other sites to verify your authority.

For Vancouver clinics, the third-party signals that matter:

Mentions in BC health publications (HealthLink BC referenced articles, BC magazine features)
Listings in registered association directories (PABC, RMTBC, CTCMA, BCCA, etc.)
Partner clinic blog mentions, cross-referrals in writing
News mentions, community publication features
Citations in industry reports or local-business roundups

A clinic with 8 to 12 authoritative third-party mentions related to a specific service gets recommended for that service. A clinic with none does not, regardless of how good their own site is.

3. Brand-mention frequency in the broader web graph

How often your clinic is mentioned by name in relevant contexts (blogs, social media, podcasts, healthcare directories) shapes how confident the model is that you exist and are credible.

Most clinics under-invest in earned mentions. The fix is not paid placement; it is being the kind of clinic other people in the BC health ecosystem talk about by name.

What to actually build

1. Publish an llms.txt file

llms.txt is a plain-text file at the root of your site (yourclinic.com/llms.txt) that summarises your site for AI models. The proposed format (Jeremy Howard, 2024) is:

# Clinic Name

> Short description of what the clinic is and where.

## What we do
- Service one
- Service two

## Who we work with
- Patient cohort
- Patient cohort

## Key pages
- Home: https://yourclinic.com/
- Services: https://yourclinic.com/services
- Specific service pages

## Common patient questions
- Question 1
- Question 2

## Contact
- email, phone, address

Click Chemistry publishes one at clickchemistry.ca/llms.txt. The file is read by ChatGPT and Perplexity (at minimum) and gives the model a structured understanding of the site. Cheap to add, useful asymmetric upside.

2. Structure service pages for AI extraction

Every service page should have:

One H1 that names the service and the location
Two to four H2 sections covering the major patient questions
An FAQ section with 5+ question-answer pairs (also serves traditional SEO)
Specific data: typical visit duration, typical number of sessions, insurance coverage, cost range, what to expect
Named practitioner with credentials and a short bio

This is the same structure that helps traditional rankings (see our local SEO cornerstone). The bonus for AI search is that the structured content is much easier for an LLM to extract and quote.

3. Earn third-party mentions

This is the slow work. The patterns that produce mentions:

Original data or perspectives. A clinic that publishes its own data on patient outcomes, insurance coverage trends, or service-specific insights gets cited by journalists and other writers.
Practitioner authorship beyond the clinic site. Practitioners writing for industry publications, association newsletters, or guest posts on partner clinic blogs builds a citation graph.
Local community visibility. Health-related events, partnerships with community organisations, and visible presence at BC health conferences create earned mentions over time.
Partner clinic content. Reciprocal mentions between non-competing clinics (a physiotherapist mentioning a specific naturopath they refer to, or vice versa) create a credibility graph the AI reads.

This is a multi-year project, not a campaign.

4. Use FAQ and structured data schema

JSON-LD schema markup helps AI search the same way it helps Google. The schemas that matter most for clinics:

MedicalClinic / Physiotherapist / Chiropractor / NaturopathicPractitioner schema with full NAP, services, and credentials
FAQPage schema on every service page
Article schema on every learning content piece
Review / AggregateRating schema where authentic data exists

LLMs that crawl your site read the schema and use it to construct accurate answers about your clinic.

What does NOT work

A few patterns to avoid:

Generic content. “Welcome to our clinic, where we provide excellent care” earns zero AI recommendations.
Stuffing the llms.txt with marketing copy. The file is for honest information, not promotional language. LLMs that read it weight the data, but they detect manipulation.
Trying to game the AI with hidden keyword stuffing. Modern LLMs are robust to this. The effort is wasted.
Spinning up fake third-party content. AI-generated blog comments, paid placements with no editorial review, and link-farm citations are detected and discounted.

What is changing

AI search is still evolving. What works in 2026 may shift in 2027. Two trends worth watching:

Direct booking integrations. Some AI platforms are testing direct-booking handoffs (the patient asks the AI for a recommendation and the AI offers to schedule). Clinics that are ready to integrate booking platforms with AI agent endpoints will have a structural advantage.
Citation transparency. Models are increasingly required to show which sources they pulled from. The clinics that get cited by name in transparent citation lists get clicked back to.

The fundamentals are stable: be the most useful, most authoritative source on your specific service in your specific location. Everything else follows.

For an audit of your current AI search visibility (we look at how Perplexity and ChatGPT actually respond to queries about your service in your area), the Clinic Growth Review includes this in the broader local-presence assessment.