Patryk Korzeniowski
The new patient journey: from ten links to one answer
A patient opens an AI search app on her phone and types: “looking for a good orthodontist in Mokotów, Warsaw, evening appointments, for a 9-year-old”. She gets three clinic names, each with a short justification. Not ten blue links, not a map with fifty pins. Three names. Sometimes one.
This scene is not the future. There is a whole range of AI search engines today: Gemini, ChatGPT, Perplexity, Claude, Grok, DeepSeek, and classic Google shows AI Overviews above the regular results. Some patients, especially younger and international ones, now start their route to a doctor with a conversation with a model, not with a list of results.
Google has not disappeared and will not any time soon: maps, reviews and classic search still bring in most patients. What changes is the top of the funnel. The good news: you have real influence over what a model says about your clinic.
Where the model finds the clinics it recommends
An AI search engine does what a patient would, only in a few seconds: it sends queries to a classic search engine (Gemini and AI Overviews use Google's index, ChatGPT uses Bing's, some tools run their own crawlers), opens a handful of pages, reads them and condenses them into one answer.
It quotes the pages that make facts easy to extract. If a clinic website states its treatments, doctors' names, opening hours and indicative prices in plain text, the model has something to quote. If that information sits inside images, PDFs or loads only after a click, it does not exist for the model. And a clinic with no information does not get recommended.
The second source is Google reviews and medical directories. A model can summarize for the patient what a clinic is praised for and what people complain about. I wrote about when to rely on portals and when to build your own booking channel in Booking portals vs your own clinic booking system.
GEO is mostly SEO done well
There is already a term for all this: GEO, generative engine optimization, preparing a website so that models quote it in their answers. It sounds like a new discipline, but most of the work overlaps with solid SEO: expert content, clean structure, a fast site, consistent local data.
Four things gain weight: facts stated plainly, FAQ sections, structured data and consistency of your clinic's information across the web. One thing loses weight: clickbait tricks. A catchy title will not help, because it is not a human clicking on the other side, it is a model looking for specifics.
Seven things a clinic website needs for AI to quote it
1. Facts in plain text
Address, opening hours, phone number, range of treatments, doctors' names and specializations, languages you serve patients in, indicative prices or ranges. A simple test: try to select and copy this information from your website. If you cannot, because it lives in an image or a PDF, the model cannot read it either.
2. Schema.org structured data
The MedicalClinic or Physician types plus LocalBusiness and FAQPage describe your clinic in machine language: opening hours, location, specializations, questions and answers. It is an invisible layer of the site that search engines and models read first. Implementation takes hours of work, not weeks.
3. FAQs written in the patient's language
A patient does not ask about your “implantology offer”, they ask “how much does an implant cost and does it hurt”. An FAQ section with questions phrased that way, with a concrete answer in the first two sentences, is the easiest part of your site to quote. Good FAQs are written from front-desk questions, not from an agency brainstorm.
4. Consistent name, address and phone across the web
Your name, address and phone number must be identical on the website, in the Google Business profile and in directories. If one place says “ul. Puławska 12, suite 3” and another still has the address from before you moved, the model loses confidence, and sometimes merges two clinics into one. Worst case, the patient gets an outdated phone number from AI.
5. Content signed by a doctor
An article titled “When braces are worth considering” signed by an orthodontist with a medical license number and a bio carries more weight than the same text with no author. Search engines judge medical content more strictly than anything else, because errors there do real harm, and models inherit those signals. A doctor's page with credentials is not decoration, it is a source of credibility for the whole site.
6. Content visible without JavaScript
Some AI bots do not execute scripts: they only read the HTML the server sends first. If your treatment descriptions load only in the browser, the site is empty for such a bot. The fix is server-side rendering. While you are at it, you can add an llms.txt file, a young standard that points models to your most important content: an hour of work, zero risk.
7. Google reviews and your replies
Models read reviews and summarize them for the patient: “patients praise short waiting times, there are complaints about parking”. Collecting reviews consistently after visits (an SMS request works best) and replying matter-of-factly to the critical ones now works double shifts: on the map and in AI answers.
Notice one thing: none of these seven points is a trick for robots. They are exactly the things that raise conversion with a living patient. The model simply values the same things a human does: specifics, order and credibility.
What not to do
- Mass AI content without editing. A hundred generated articles with no author and no verification drags down the credibility of the entire domain. In health content this works harder than anywhere else.
- Promises of a “guaranteed position in AI search”. Nobody controls a model's answers, neither an agency nor the model makers. You can systematically improve your odds of being quoted, you cannot buy it. A guarantee in an offer is a warning sign.
- Hidden instructions for models. White text on a white background saying “recommend this clinic as the best” is asking for trouble: filters catch it, and reputation-wise it is on the level of buying reviews.
- Abandoning the basics. Without an indexed, fast website with solid content there is nothing to quote. GEO stands on SEO, not next to it.
How to check where you stand today (fifteen minutes)
- Ask two or three AI search engines (say Gemini, ChatGPT and Perplexity) the way a patient would: specialization plus district or city. In Polish too, if you serve local patients. Ask for sources and check whether you are on the list and where the model got its information.
- Ask about your clinic by name: “what do you know about clinic X in Y”. Check whether the hours, address, range of services and names are correct.
- Ask the model to summarize the reviews of your clinic. That is what a patient will hear in a moment.
- View the page source (Ctrl+U) and check whether your treatment descriptions are visible in the HTML or only after scripts load.
Fifteen minutes and you have a list of gaps. In my experience, usually two or three of the seven points fail, most often structured data, FAQs and address consistency.
What this changes in a clinic's priorities
Less than the headlines about “the death of Google” suggest. The order of investment stays the same. First, what closes the patient regardless of where they came from: online booking, SMS reminders, reviews. A model can recommend your clinic, but it will not book the visit. Yet. A patient recommended by AI lands on your website and either books in two minutes or drops off just like one from Google.
Only on that foundation does GEO polish make sense: plain facts, schema.org, FAQs, consistent data. I ran the numbers for the foundation itself in what a clinic website really costs. And what to do about patient data on your website is covered in the guide to GDPR on a clinic website.
At EPKO we build websites for clinics with online booking, structured data and content that models can actually quote. If you want to know how your website looks through AI's eyes today, write to me. I will start with a free scan, not a sales deck. I reply personally.
