When patients ask ChatGPT, Gemini, Claude or Perplexity where to go for an aesthetic treatment, the answer usually contains only a small number of clinics.
Penydarren studies how those recommendations are formed and helps UK aesthetic clinics become easier to find, verify and recommend.
Our research traced the searches, sources and clinic names behind 1,420 AI-generated answers across London, Manchester and Newcastle.
Testing period: 9 June – 7 July 2026 (two monthly rounds)
Last updated: 16 July 2026
We studied what happens between the patient’s question and the final recommendation.
Rather than assume that AI visibility works like ordinary SEO, Penydarren tested real patient questions across the four major AI engines.
The study covered treatment-led, concern-led, practitioner-led, review-led and location-led questions across fourteen aesthetic treatment categories.
414
patient questions
→
1,420
AI responses
→
2,388
background searches
→
11,465
recorded source instances
→
7,365
clinic mentions
→
1,111
distinct clinics
4 AI engines — ChatGPT, Gemini, Claude, Perplexity3 UK cities — London, Manchester, Newcastle14 treatments (~94% non-surgical)
WHAT WE COUNTED
A clinic was counted as recommended when it was named in the answer itself. A clinic appearing only in a citation footer, source list or search result was not counted as a recommendation.
THE HEADLINE FINDING
There is no single AI result to optimise for.
The four engines did not consistently recommend the same clinics.
63%
of clinics appeared in only one engine
Most clinics named during the study had visibility in ChatGPT, Gemini, Claude or Perplexity, but not across the wider AI search market.
8.9%
appeared across all four
Only around one in eleven of the 1,111 distinct clinics appeared in every engine.
67%
of the specific picks changed between monthly tests
The broader group of credible clinics remained relatively stable, but the individual clinics selected for particular answers frequently changed.
A clinic can be highly visible in one engine and absent from another. It can also appear for a treatment during one testing round and disappear during the next.
AI visibility therefore cannot be understood through a single prompt, one engine or one static ranking.
Each engine has to be measured separately and repeatedly.
THE RETRIEVAL GATE
A clinic cannot be recommended if it never enters the research process.
Before producing an answer, an AI assistant may rewrite the patient’s question, run several searches and retrieve information from clinic websites and independent sources. Search visibility determines which clinics are available for the engine to consider.
41%
positions 1–5
12%
positions 16–20
≈3.4×
a top-five result was ~3.4× more likely to enter the citation set than one near the bottom of the first twenty
But ranking was not the final decision. A high position increased the likelihood that a page would be retrieved. It did not guarantee that the clinic would be named in the answer. The engine still had to understand:
—Who provides the treatment
—Where the clinic is located
—What the practitioner is qualified to perform
—Whether the page answers the patient’s actual question
—Whether independent sources support the clinic’s claims
—Whether the available information is clear, current and consistent
THE TWO LOCKS
First, be retrieved. Then, provide enough evidence to be selected.
LOCK ONE · RETRIEVAL
Search position opens the first lock. It gives the clinic a chance to enter the engine’s research process.
LOCK TWO · SELECTION
The quality and authority of the available evidence open the second.
A clinic can rank and still be left out. It can be cited without being recommended. It can appear in several sources but lose the final place to a clinic whose information is more relevant, authoritative or easier to extract.
90–98% of the recommendations recorded in the study were supported by a page retrieved during the live research process. The remaining recommendations appeared to come from information already available to the model.
For most clinics, live retrieval is therefore the main route into the answer.
FOUND · CITED · VERIFIED · RECOMMENDED
AI visibility is a chain, not a ranking.
Found
The clinic appears in one or more searches created by the AI engine. Being found puts the clinic into consideration. It does not mean that the engine will read the clinic’s website or include it in the final answer.
Cited
The engine retrieves or references the clinic’s website or another source mentioning it. A citation shows that the clinic entered the research process. It does not prove that the clinic was recommended.
Verified
The engine finds enough consistent evidence to understand the clinic, practitioner, treatment and location. That evidence may come from the clinic’s website, professional profiles, registers, directories, reviews and other independent sources.
Recommended
The clinic survives the comparison and is presented to the patient as a suitable option.
Citation visibility matters. Recommendation visibility is the outcome.
THE EVIDENCE LAYER
Your website is only part of what an AI engine reads.
Clinic-owned websites supplied approximately 36% of the recorded source instances during the study. The majority came from third-party domains. These included:
—Professional registers
—Regulatory sources
—Clinic directories
—Practitioner profiles
—Review platforms
—Editorial articles
—Treatment guides
—Local search results
AI visibility therefore cannot be improved through website content alone. The clinic must be represented clearly and consistently across the wider collection of sources an engine may use to research and verify it.
This does not mean appearing on every available directory. It means understanding which sources matter for a particular engine, treatment and location.
WHY THE ENGINES DISAGREE
ChatGPT, Gemini, Claude and Perplexity do not research clinics in the same way.
Each engine can:
—Interpret the patient’s question differently
—Create different background searches
—Retrieve a different mix of websites and third-party sources
—Select different pages from the same search results
—Extract different information from those pages
—Produce a different final shortlist
The result is four related but distinct visibility channels. A clinic that appears frequently in ChatGPT is not automatically visible in Gemini. A clinic with strong Gemini visibility may be missed by Claude. A clinic repeatedly retrieved by Perplexity may still fail to appear in its final recommendation.
Effective AI visibility work must therefore measure the engines separately rather than treating “AI” as one platform.
CONTENT THAT CAN BE USED
More pages do not automatically create more visibility.
The study initially examined count-based signals such as the number of treatment pages and the quantity of keywords present on a clinic website. Those measures were unreliable.
Websites with large numbers of repetitive or padded pages could appear strong by volume while providing little distinctive information for an AI engine to use.
The more useful signal was substantive treatment-page depth. Strong pages made it easy to establish:
—What the treatment is
—Which concerns it addresses
—Who performs it
—The practitioner’s relevant experience
—Where the treatment is available
—Who may or may not be suitable
—What evidence supports the clinic’s claims
—How the treatment differs from related options
Penydarren assesses content for distinctiveness, authority, extractability and resistance to templated padding.
The objective is not to create more words. It is to create information that can be found, understood and used in an answer.
DEDICATED EVIDENCE
A treatment hidden inside a short list is difficult to recommend confidently.
A clinic may technically mention a treatment without providing enough information for an engine to understand its relevance. A short paragraph beneath a generic treatment heading gives the engine less usable evidence than a dedicated, well-structured page. Clear treatment architecture helps connect:
This matters because patients do not only ask for treatment names. They ask about outcomes, concerns, suitability, practitioners, safety, location and trust.
WHAT CLINICS CAN IMPROVE
The signals surrounding a clinic can be strengthened.
AI-generated answers are variable and no agency can guarantee that a particular clinic will be named. But clinics can improve the evidence available to the engines making those decisions.
Search visibility
Improve whether relevant clinic pages appear in the searches the engines create.
Treatment and concern content
Build substantive pages that answer the questions patients ask and the related searches AI assistants run.
Practitioner evidence
Connect qualifications, registrations, clinical experience and treatment specialisms clearly to the clinic.
Location relevance
Make it clear where the clinic operates and which local treatment questions it is relevant to.
Third-party consistency
Ensure that independent sources describe the clinic, practitioners, treatments and locations accurately.
Citability
Present distinctive, authoritative information in a form that can be extracted and used when an engine generates its answer.
Repeated measurement
Test visibility across multiple questions, engines and time periods rather than relying on one prompt.
HOW PENYDARREN USES THE RESEARCH
We identify where your clinic is being lost.
01
Test the questions patients ask
We test relevant treatment, concern, practitioner and location questions across ChatGPT, Gemini, Claude and Perplexity.
02
Trace the research process
We record the background searches, retrieved sources, citations and clinics appearing behind the final answers.
03
Separate discovery from selection
We identify whether your clinic is absent from the search process, retrieved but not named, or included in the final recommendation.
04
Assess the available evidence
We examine the clinic website and the independent sources used to understand and verify the clinic.
05
Strengthen the relevant gaps
We improve the content and external evidence most closely connected to the point where the clinic is being lost.
06
Retest the engines
We repeat the testing to measure whether the clinic is entering more searches, citations and final answers over time.
The study was conducted by Penydarren across two monthly testing rounds (9 June and 7 July 2026). It covered:
·414 patient questions
·1,420 AI-generated responses
·ChatGPT, Gemini, Claude and Perplexity
·London, Manchester and Newcastle
·Fourteen aesthetic treatment categories
·Approximately 94% non-surgical questions
The study did not attempt to represent every UK city or every aesthetic treatment. Findings about geographic visibility should therefore be read within the three locations tested.
AI-generated answers are also variable. Search results, retrieved sources, model behaviour and final recommendations can change between testing rounds.
The statistical models identify patterns within the observed dataset. They are directional and should not be interpreted as proof that any single signal directly causes an AI recommendation.
The clearest result was consistent across the analysis: a clinic must first enter the engine’s retrieval process before its content and supporting evidence can influence the final recommendation.
DEFINITIONS
What the measurements mean
Patient question
A prompt written to reflect how a patient might ask an AI assistant to find, compare or choose an aesthetic clinic.
AI response
The answer generated by ChatGPT, Gemini, Claude or Perplexity.
Background search
A distinct search created by an AI engine while researching its response.
Recorded source instance
An occasion where a domain or page appeared within the recorded source set for a response — whether presented as a citation or retrieved as a supporting source.
Clinic mention
An occasion where a clinic was named in the answer prose.
Distinct clinic
An individual clinic identified after names and domains were checked and consolidated.
Recommended clinic
A clinic named in the answer itself as a relevant or suitable option. Clinics appearing only in a citation footer or source list were not counted as recommended.
QUESTIONS
AI visibility for aesthetic clinics
What is AI visibility for an aesthetic clinic?
AI visibility is whether assistants such as ChatGPT, Gemini, Claude and Perplexity find, understand and name a clinic when a patient asks where to go for a treatment. It includes whether the clinic appears in the searches and sources used to create the answer, as well as whether it reaches the final recommendation.
Is AI visibility the same as SEO?
No, although the two overlap. SEO affects whether a clinic appears in search results. AI visibility examines what happens after that, including which results the engine retrieves, which sources it uses and which clinics it ultimately names. Strong search visibility can help a clinic enter the process, but it does not guarantee recommendation.
Why do different AI engines recommend different clinics?
Each engine can create different searches, retrieve different sources and select different evidence. In Penydarren’s study, 63% of distinct clinics appeared in only one engine and just 8.9% appeared across all four.
Does being cited mean a clinic was recommended?
No. A clinic or source can be retrieved and cited without the clinic appearing in the final answer. Being cited means the clinic entered the research process. Being recommended means it was selected and named to the patient.
Can a clinic improve its AI visibility?
The information and evidence surrounding a clinic can be improved. That includes search visibility, treatment content, practitioner evidence, location relevance and consistency across independent sources. Any effect on recommendations must then be measured through repeated testing.
Does adding schema or an llms.txt file guarantee AI visibility?
No single technical change guarantees inclusion in an AI-generated answer. Technical accessibility matters, but it does not replace relevant content, search visibility, practitioner evidence or credible third-party support.
Which clinics does Penydarren work with?
Penydarren works with UK aesthetic and cosmetic clinics, including doctor-led practices, nurse-led clinics, independent locations and multi-site groups.
Is AI visibility part of local SEO?
Local SEO affects whether a clinic appears in location-based searches and whether its business information is clear and consistent. This can help a clinic enter the research process, but AI visibility also examines which pages are retrieved, which independent sources are used and whether the clinic reaches the final recommendation. Local SEO can open the first lock. It does not automatically open the second.
Is Penydarren an SEO or digital marketing agency?
Penydarren is an AI search optimisation consultancy specialising in UK aesthetic clinics. The work can involve SEO, local search, website content, practitioner evidence and third-party profiles, but each activity is selected according to where the clinic is being lost from the AI recommendation process.
Does E-E-A-T guarantee that a clinic will be recommended?
No. Clear experience, expertise, authority and trust can strengthen the evidence available around a clinic, particularly where medical treatments and patient safety are involved. They do not form a single score and do not guarantee that an AI engine will select a particular clinic.
START HERE
Find out where your clinic enters the recommendation process.
We will show you how your clinic appears across ChatGPT, Gemini, Claude and Perplexity, where it is being found and what may be preventing it from reaching the final answer.