Why Concierge and Specialty Practices Are Losing Patients to ChatGPT

For two decades the front door of a healthcare buying decision was a Google query. A patient considering a concierge medicine membership, a Mohs surgery, a labiaplasty, or a complex orthopedic consult would type their question, scan the top three blue links, and click through to the practices that ranked. Pages were optimized accordingly. The economic logic of an entire generation of healthcare SEO was built on top of that behavior.

That behavior is changing fast. ChatGPT now serves more than two hundred million weekly users with a conversational answer that synthesizes a small, opinionated set of sources. Google's own AI Overviews is doing the same on roughly a quarter of all searches. Perplexity is the default research surface for a meaningful share of high-income, high-intent buyers, which is the exact population concierge, dermatology, and surgical specialty groups depend on. The high-value query has not gone away, but the interface has.

That changes everything about what ranking means for a healthcare practice. The new objective is not page-one position. Instead, success is being inside the citation set the model picks when it answers. If you are not cited, you are not visible, even if you are still ranking in classical search. We covered the underlying architecture in our SEO and LLM optimization system, and this essay is the practitioner-level deep dive.

Large language models do not browse the web the way a human does, and they do not pick citations the way Google's classical algorithm picks blue links. Each major LLM has a retrieval layer that pulls a small candidate set of pages from an index at query time. It ranks them with a different objective function than classical results, and then a generative layer composes an answer that cites a subset of the retrieved set.

Two consequences fall out of this process. First, classical SEO ranking is necessary but not sufficient. Pages that do not rank classically are rarely retrieved, but pages that rank classically are routinely passed over for citation. Second, the citation decision rewards a different kind of clarity than the click-through decision, favoring factual specificity, structured extractability, named expertise, and an absence of fluff. Pages written for a hypothetical buyer's emotional state get cited less often than pages written like a clinical reference card.

From instrumented LLM-rank tracking across hundreds of healthcare queries over the last twelve months, the signals that correlate most strongly with citation are mostly structural. Almost all of them are independently good for buyers and for classical search.

• Clear H-tag hierarchy. This includes one H1 and descriptive H2s phrased as questions or claims, along with H3s for sub-claims. Models read the outline before they read the body.
• Factual specificity. Citations favor numeric ranges, durations, eligibility criteria, contraindications, side-effect rates, and time-to-result. Vague pages do not get cited because there is nothing extractable to attribute.
• Original data. Useful data includes internal patient-volume statistics, internal outcomes data where compliant, and internal pricing posture. Aggregator content can be summarized without citation, while original data forces attribution.
• Named, credentialed authors. This involves bylines with credentials and links to provider bios. Models weigh expertise signals heavily for medical content.
• Schema density. Implementation includes MedicalProcedure, Physician, MedicalCondition, and FAQPage. Schema is not a ranking factor, but rather an extraction format that makes the page legible to retrieval.

The under-discussed lever is entity coherence. Models build internal representations of practices, providers, and procedures. If your provider page says Dr. Sarah Lee, board-certified dermatologist, but your homepage says Dr. S. Lee, Mohs surgeon, and your scheduling page says Sarah J. Lee, MD, the model sees three weakly-linked entities instead of one strong one. Citations go to the practices whose entity graph is internally consistent and externally reinforced. This occurs when the same provider name and credentials appear on the practice site, the state medical board record, the hospital affiliation page, and the major directories.

Fixing this is not glamorous work. It requires a multi-week audit of every place a provider name appears, a normalized canonical form, and a coordinated rewrite. But it is one of the highest-leverage interventions we run, especially for surgical specialty and concierge groups whose buyers do an above-average amount of cross-checking.

Before changing anything, you must measure. The audit is straightforward in principle and tedious in practice. Take the ten to twenty highest-intent queries for your service lines, phrased the way a patient would actually phrase them rather than how an SEO tool exports them. Run each query against ChatGPT, Gemini, and Perplexity, in both general and search-grounded modes. Capture the cited URLs.

Three patterns will emerge, including queries where you are cited and need to defend, queries where a competitor is cited and you should attack, and queries where the citation set is dominated by aggregators or generic media. This third bucket is usually the largest territory available for action, because the bar to displace an aggregator is mostly about original specificity. A real practice has this specificity and an aggregator does not.

The same audit, run for you, lives inside the free website audit.

Three content patterns disproportionately earn citations in the verticals we cover most.

1. Procedure deep-dives with internal data. Consider a page on Mohs surgery, GLP-1 management inside concierge medicine, ACL reconstruction protocols, or laser resurfacing. When these are written by a named provider with real recovery timelines, contraindications, and the practice's own outcome data, they get cited because they answer the next question the patient is about to ask.
2. Comparison pieces written without sales fog. An article comparing direct primary care and concierge medicine that is written honestly by a practice will outperform aggregator content because it reads as expert commentary instead of marketing.
3. Decision frameworks. Examples include how to choose a Mohs surgeon, what to ask before a cosmetic procedure, or when membership medicine is and is not worth it. Frameworks attribute well because they are inherently structured and quotable.

All three are also superb assets for paid acquisition and for classical SEO. The same page that gets cited in Perplexity is usually the page that converts a logged-in visitor too. We tie this work into the broader healthcare growth systems stack so the same content asset earns three forms of compounding return.

Classical rank tracking does not work for LLM visibility because there is no stable SERP to scrape. The replacement is a citation-rate metric. For a defined keyword set, track what percentage of model responses cite at least one of your pages across each of the major LLMs, sampled weekly. That number, combined with a per-query break-out, is the dashboard a practice CMO should be watching.

Tied to closed-loop attribution, the citation-rate metric becomes pipeline-defensible. We document the underlying instrumentation in our analytics and reporting system.

You do not need a six-month content overhaul to begin earning LLM citations. A focused 30-day sprint moves the needle for most practices.

Days 1 through 7: Start with an audit. Identify the ten highest-intent queries, capture current citation sets across the three major LLMs, and fix the most obvious entity-coherence issues like provider name normalization, credential consistency, and schema gaps.

Days 8 through 18: Ship two procedure deep-dives written by named providers, one comparison piece, and one decision framework. Each should include proper schema, factual specificity, and internal cross-links to the corresponding service pages.

Days 19 through 30: Transition to measurement. Re-run the citation audit weekly, watch which queries flip into your favor, and iterate the next round of content based on what got cited and what did not.

By day 30, a practice with no prior LLM strategy will typically appear in the citation set for one to three of its top ten queries. That is a per-query attribution surface that did not exist a month earlier. This work compounds because each new cited page strengthens the entity graph for the next.