What ChatGPT Cites When a Patient Asks for a Med Spa

By Jeremy Bengtson, founder of The Search Sherpa.

Here is a run, reconstructed from a patient-shaped prompt about a suburban Chicago clinic. It is a composite, not a case study. The prompt was plain. Who is the best place near me for lip filler.

ChatGPT answered in two short paragraphs and named four clinics. Underneath the answer sat the sources it had used.

One was a clinic’s own treatment page. One was a Yelp listing. One was an aesthetic directory profile with a provider bio on it. And one was a Reddit thread from a local subreddit. In it, a handful of people argued about which injector they trusted with their own face.

The first source is the kind of page a clinic writes and rewrites. The last one is the kind a clinic almost never reads.

That gap is what this page works on. Your treatment pages are the source you control, and the source you already fix. The public discussion about your practice is the source you cannot buy and cannot control. Most clinics have never logged it. It sits in the citation list anyway, working for you or against you.

Why the clinic down the street gets named and yours does not

The clinic down the street is not out-ranking you. It is being assembled into an answer, which is a different thing.

When someone asks ChatGPT, Perplexity, Google AI Overviews, or Gemini for a med spa, the engine does not rank clinics the way a search results page does. It assembles a short answer from sources it already trusts. Your website is one of those sources. It is rarely the only one that matters.

So the useful question is not how you climb. It is which sources an engine reaches for. Then it is whether your clinic is described accurately inside them.

That difference changes what work is worth doing. Ranking work asks how one page can outperform another page. Assembly work asks whether the facts about your clinic agree wherever an engine finds them. A clinic can hold a strong position in Google Maps results and still be missing from an assembled answer. It can also be named in an answer while its own website sits on the second page.

Think about what assembly means in practice. The engine reads several pages about your area, then writes a short summary in its own words. The clinics that survive that summary tend to be the ones described the same way in several places. Contradiction is expensive for an engine, and the cheapest way to avoid it is to name someone else.

The general model of how AI engines choose any local business is not a med spa question, and it is not repeated on this page. That model lives on the five inputs AI uses to decide who to recommend, where it is worked through properly. It applies to a roofer the same way it applies to a clinic. Want the wider version first? Read how AI search engines pick which local businesses to recommend, then come back here.

Google’s own help documentation names four inputs behind the business information it shows. Those are publicly available information such as crawled web content, licensed data from third parties, contributions from users including owners who claim a profile, and information based on Google’s own interactions with a local business (Google Business Profile Help).

Read that list slowly. Two of those four are things a clinic edits directly. The other two are things a clinic can influence at best. Nobody can log in and change them. That is the honest shape of the problem. I would rather show it to you than pretend the whole answer is yours to control.

One thing needs clearing off the table before we go further. Being named in an answer is not the same purchase as software that answers your phone. Both get sold with the same word on the label. Front-desk automation converts demand that has already found you. This work decides whether that demand ever hears your name. The money side sits on what this work costs a med spa. The delivery detail sits on our AI search optimization service.

Two warnings belong here too. Be cautious of anyone who sells placement inside ChatGPT as a promise. No provider controls what a model returns on a given day. Be equally cautious of anyone selling raw review volume with no descriptive content behind it. A hundred five-star ratings that all say great experience give an engine almost nothing to quote.

The source nobody optimizes: the public discussion about your clinic

Start here rather than with your website. Your website is the part you already fix. It is also the part every other page on this subject wants to talk about.

Public discussion about your clinic is indexed. Reddit threads, community question posts, and local forum conversations are crawlable pages like any other. The same systems that read your treatment page read those threads. They are not a back channel or a loophole. They are ordinary web content that happens to be written by patients instead of by you.

There is a commercial reason this surface goes unmentioned in most advice written for clinics. You cannot sell it as a deliverable. There is no checklist to run, no field to fill in, and no dashboard that turns green. A provider who bills by the task has nothing to bill for here.

A clinic owner does not much want to hear it either. It is uncomfortable to learn that strangers are writing part of your answer. In public, under usernames, at two in the morning.

It is still the surface. Pretending otherwise does not make it smaller. The citation log later on this page will put it in front of you regardless. What follows is what sits on each surface, what a clinic can do about it, and where the line is.

Reddit threads where patients compare injectors by name

Read a few of these before you decide how much they matter. The shape is consistent once you have seen ten of them.

Someone posts in a city or suburb subreddit and asks who does good Botox near a named town. Replies come in over the following week. People name providers. They compare what they paid. They mention who is heavy-handed, who rushed the consult, and who talked them out of a treatment they did not need. Occasionally a whole thread is one long argument about a single clinic.

Notice what gets named. Patients frequently name the injector, not the practice. They say a first name and a last name, or a first name and a neighborhood. The clinic is the container that person works in. It is not always named at all.

That detail matters more than it looks. If your provider bios are thin, the engine has a person’s name in a forum thread. It has no page on your site connecting that person to your clinic. The corroboration breaks at exactly the point where it would have helped you. The fix for that is not on Reddit. It is on your own site, and it comes further down this page.

Two other patterns show up often. Patients describe outcomes in ordinary language rather than treatment names. A thread can be about your service without ever using your service’s word. Someone writes about the lines between their brows softening. Nobody in that thread types the word neuromodulator.

They also answer the questions your website will not. What something really cost, once the second syringe was added. How long the swelling actually lasted. Whether the follow-up appointment was included. Whether the consult felt like a sales meeting.

An engine looking for something that reads like first-hand experience finds that content easily. Your page cannot sound like that, and it should not try. What your page can do is be accurate, specific, and easy to check against the thread.

The aesthetic directories and review platforms that describe you

This is the middle ground, where you have influence rather than control. It is also the part most clinics half-finish and then forget.

Your Google Business Profile, Yelp, RealSelf, Apple Maps, and aesthetic industry association listings each publish a description of your clinic. An engine trusts a fact more when it finds that same fact in several independent places. When those descriptions disagree about your hours, your services, or your staff, the engine has to pick one. It may not pick yours.

BrightLocal’s Local Consumer Review Survey, run on a representative panel of 1,002 United States adults, found that the average consumer uses six different review sites when choosing a local business (BrightLocal, 2026). That is corroboration seen from the patient’s side of the counter. They are already cross-checking you across six places. The engine is doing a faster version of the same thing.

The disagreements are rarely dramatic. A suite number is missing from one listing. Holiday hours were updated in one place and nowhere else. A treatment you dropped last year still sits in a directory category. A former injector still has a bio on a platform nobody logs into. Each one is small on its own. Together they teach an engine that facts about your clinic are unreliable.

So the work here is a sweep, not a project. Pull every listing that names your clinic, including the ones you never created. Aggregators and industry directories build profiles from licensed data, and those profiles exist whether you claimed them or not. Fix the name, the address, the phone number, and the hours. Then make the service list read the same way your own website reads.

We keep our 42-item Google Business Profile checklist on its own page rather than repeat it here. Consistency across every listing is the whole job.

The pages you own, and the work they actually do

Your own pages sit underneath everything above. They are the part every other page on this subject already covers. So I will be direct about the job they do, rather than hand you another checklist.

A treatment page is not there to win an argument. It is there to be the thing a directory listing and a forum thread get checked against. When all three agree, an engine has a corroborated fact. When they disagree, it has a judgment call, and you do not get a vote in it.

That reframes what belongs on the page. Name the treatment the way patients name it, then name it the clinical way in the same paragraph. Say who performs it, by name, and link to that person’s bio. Say what happens at the appointment and roughly how long it takes. Say what recovery looks like in ordinary language.

Provider bios deserve far more attention than they usually get. Remember what the forum threads do. They name a person, not a practice. Give the bio page a full name, credentials, years in practice, and the treatments that person performs. That is what lets an engine connect a thread to your clinic. Without it, the praise in that thread belongs to a name and not to your clinic.

Do the same for locations. One page per location, with its own address, its own hours, and its own staff named on it. A single page listing three suburbs is easy to write and hard to corroborate.

The method behind all of this is written up in the treatment page and schema work behind it. How the pieces connect is covered in the wider med spa marketing and AI visibility system.

What a clinic may and may not do on a discussion surface

You cannot buy your way onto this surface. That is not a rule someone imposed on you. It is simply what the surface is.

Attempts to fake it are visible to moderators and to readers. Communities that discuss injectables are unusually good at spotting a clinic account. For a medical practice, being publicly caught astroturfing is a reputational event, not a slap on the wrist. It can outlive the thread it started in.

Paying a patient to post is worse. It creates a disclosure problem and a substantiation problem at the same time. It also puts your practice on the wrong side of platform policy and healthcare-ethical marketing.

Here is what a clinic can actually do. Post under a real identity, with the practice name attached, and say plainly who you are. Answer the clinical questions people ask in public, with no pitch and no link. Read the rules of each community first, because many bar promotion outright and a few allow verified professionals to answer.

Keep the answer general and keep it educational. Do not diagnose a stranger from a photograph. Do not comment on another clinic’s work, even when invited to. Those two habits separate a provider a community keeps from an account a community reports.

The same rules apply to your staff. An injector who answers questions in a community she already belongs to is fine, as long as she says where she works. A front desk account created last week to praise the practice is not. If you are going to be present, be present as people.

The point is not to seed mentions. The point is simpler. A real provider answering a real question is the kind of content these communities keep, and kept content gets crawled.

I will be honest about the limit. No provider controls that conversation, including this one. What you can control is the rest. Make sure nothing else an engine finds contradicts the version of you sitting in that thread.

How to test what the engines say, without fooling yourself

The strongest objection to everything above is a fair one. Every agency tells you to open ChatGPT and ask it for the best med spa near you. That test is worthless, because the answer changes every time you run it.

Your analytics will not settle it either. Google Search Console reports how your own site performs in Google search. It does not tell you which clinic an assembled answer named, and it says nothing at all about ChatGPT, Perplexity, or Gemini.

The fix is not a better prompt. It is a method you can repeat, and it produces evidence rather than a screenshot. None of it requires you to take my word for anything.

Whitespark’s Local Search Ranking Factors study covers 187 individual factors, scored by 47 working local search experts. Its 2026 edition added a separate column for AI search visibility for the first time (Whitespark, 2026). The measurement side of that question, in its general form, is covered in the five signals we use to measure AI search results. What follows is the clinic version.

To test what the engines say about a med spa, build a prompt set from your own treatment menu. Fix the number of runs before you read any answer. Log the sources cited beneath each answer. Then count how many of your prompts each rival clinic appears in.

Build the prompt set from your own treatment menu

Open your own price list and your own booking system. That is the source document, not a keyword tool.

Pull the treatment names you actually perform. Pull the brand names patients say out loud, because a patient asks for Botox and Morpheus8 rather than for neuromodulators and radiofrequency microneedling. Add your neighborhood or suburb. Add the financing question, because it gets asked constantly. Add near-me phrasing.

Write each prompt the way a patient speaks, not the way a marketer types. Who does the best lip filler near me. Is there anywhere near a named suburb that does payment plans for laser. Which med spa should I go to for acne scarring if I have darker skin.

Cover the whole menu, not only the treatments you enjoy selling. A device you run twice a month still earns citations. It is also usually less contested than the injectables every clinic in your radius advertises.

Then put your own suburb name into about half of them and leave the other half unlocated. The two halves behave differently, and the difference is informative on its own. Located prompts lean on listings and maps data. Unlocated prompts lean harder on written content, and that is where a thin website shows.

Twenty to thirty prompts is enough for a single location. Write them down in a fixed list and stop editing the list. The moment the prompt set changes, every number you collected before it becomes uncomparable.

Fix the run count before you read a single answer

This is the step that decides whether the whole exercise means anything. Do it before you look at any output.

There is no published standard for run counts in AI search testing. What follows is our house rule and not an industry rule. I will keep saying so every time it comes up.

We run each prompt ten times per engine, in a clean session, across ChatGPT, Perplexity, Google AI Overviews, and Gemini. We count a clinic as present when it appears in at least half the runs for that engine. That threshold is written down before the test starts, not chosen afterwards.

A clean session matters as much as the count. A signed-in account carries your own history, your own location, and your own past questions. It will show you a friendlier world than a patient sees.

Google AI Overviews adds one wrinkle worth planning for. It does not fire on every query, so record the runs where no overview appeared at all. A prompt that never triggers an answer is a different finding from a prompt that triggers one without you in it.

One appearance is not a result. One absence is not a failure. The habit to avoid, and it is a common one, is rewording the prompt until your clinic finally shows up. That produces a number that feels good and tells you nothing you can act on.

Score each of the four engines on its own line and never average them. They diverge, sometimes widely, and a blended figure hides the one engine where you are invisible.

Log the citation, not the verdict

Presence is the score. The sources are the reason, and the reason is the only part you can act on.

So save the source list behind every answer, not just whether your name appeared. A spreadsheet with one row per run and one column for each cited source is enough. It is tedious for the first hour and then it starts paying.

Keep the row honest. Record the date, the engine, the exact prompt, whether your clinic appeared, which rivals appeared, and every source named underneath. Paste the full link rather than the site name. A directory has thousands of pages and only one of them is describing you.

After a hundred rows the log tells you which surface is carrying the answer in your market. If the citations are directory pages and review platforms, corroboration is doing the work. If they are forum threads, the public discussion is doing the work, and you now have named threads to go and read. If they are your own treatment pages, you have earned something most clinics in your radius have not.

The log also catches the thing a presence score cannot. You will find rival clinics cited from sources you did not know existed. You will find your own clinic cited from a page you forgot you published.

Presence without attribution is a number you cannot use. Attribution without presence is still a to-do list. Keep both columns.

Score share of voice against the clinics that keep appearing

By the third or fourth engine you will notice the same three or four clinic names recurring. Write them down. Those are your real competitors in this channel, and they are often not the ones you assumed.

Then count. For each named rival, how many of your prompts did that clinic appear in, out of the full set. Do the same for yourself. Now you have a per-rival appearance count across a fixed prompt set. That is the only competitive number on this page worth quoting to anyone.

Read it per rival and never blend it. A single visibility score, averaged across engines and competitors, is the most comfortable and least useful number you can build. It hides the exact thing you ran the test to find, which is where the gap sits and who owns it.

The counts also tell you what to fix. A rival who appears everywhere across every engine has earned broad corroboration. A rival who appears only in Perplexity, and only on unlocated prompts, has earned something narrower. The citation log will usually show you the source doing it.

Run the whole thing once and you have a baseline. That baseline is the artifact, and it is worth more than any impression of who is winning locally.

Once you have run it yourself, you also have a way to check any provider, this one included. Ask which sources the last AI answer about your clinic cited, and ask to see that list. Then ask how many runs per engine sit behind any visibility number you were quoted. A provider who cannot answer either question is guessing in front of you, politely.

When an engine states something false about your clinic

Sooner or later, one of the four engines will say something about your practice that is simply not true. Correcting it is a different job from measuring it.

The failure modes are specific and they repeat. Wrong hours. A price you stopped charging two years ago. A location you closed. An injector who left. A treatment you have never offered, usually because a directory auto-filled a category. A medical director’s name that is out of date. Another clinic’s reviews attributed to you because a listing got merged. Occasionally, a claim about a treatment outcome that you would never make yourself.

The engine is not inventing these. It is quoting something. That is genuinely good news, because a quotation has a source and a source can be corrected.

So the correction path runs source by source, not engine by engine. Chasing the engine is the mistake almost everyone makes first. You cannot argue with the output, and there is no support ticket that rewrites an answer.

Start by finding the sentence. Run the prompt that produced the error a few times and read the citation list each time. The wrong fact almost always sits on one of the pages named there, often in the same words.

Fix your own pages first. They are the fastest thing you control and, more often than owners expect, the origin of the error. An old promotion page, an unlinked location page, a staff page nobody updated after a resignation. All three keep answering questions long after you stopped thinking about them.

Then correct your Google Business Profile and every directory listing carrying the wrong fact. Correct all of them, not the first two you find. A single stale profile can keep an error alive for months, because corroboration cuts both ways.

Some listings will not cooperate. A directory may have no owner login, or a support queue that never answers. Log the attempt, keep the date, and move to the next one. Then outweigh it. One uncontactable page loses to the same fact stated correctly on five pages an engine also reads.

Then work on the review record, through the review system we build for a clinic and the visibility system behind the fix.

Then wait, and re-run the prompt set a few weeks later against the same fixed list. Engines re-read sources on their own schedule and nobody publishes that schedule. Most of the feedback buttons inside these products are requests rather than levers, and it is healthier to treat them that way.

I want to be plain about the limit. No provider can force an engine to change an answer. What a provider can do is remove every source supporting the wrong version. Then give the engine a better-corroborated one to find instead.

One exception deserves its own sentence. If an engine states something false about a medical outcome, a credential, or a license, that has stopped being a marketing problem. Log it, screenshot it, and talk to your counsel before you talk to your marketer.

Where standard AI search advice collides with medical aesthetics rules

This is not legal advice. The Search Sherpa is a visibility specialist, not an attorney or a compliance officer. Check this section with your own counsel and your state medical board. Three collisions come up in almost every med spa conversation.

The first is before-and-after imagery. AI answers reward proof that is easy to extract, and result photographs are the most persuasive proof a clinic owns. They are also governed by consent, release, and disclosure rules that vary by state. Publishing more of them is standard AI search advice. Publishing them without a consent process is a different decision, and not ours to make for you. The workable version is a consent form signed before the first photograph, not chased after a good result.

The second is publishing treatment prices. Written prices are easy for an engine to quote, so the standard advice is to put them on the page. Medical advertising rules in some states constrain how a practice may advertise the price of a procedure. There is also a plain business cost. A published price list is the easiest competitive intelligence you will ever hand a rival clinic. A starting price, with the consultation requirement stated beside it, is usually the version a clinic can live with.

The third is review solicitation. Reviews feed the corroboration surface, so the standard advice is simply to ask for more of them. Substantiation rules, review gating, and incentives are all live issues in healthcare-ethical marketing, and platform policies sit on top of state rules. We keep a compliant review request process documented separately rather than improvise it per clinic.

One thing links all three. Everything on this page that works, works because it is checkable by someone other than you. A consent form, a stated price rule, and an unincentivized review all survive being looked at by a regulator, a platform, and a patient.

That is what compliance-aware messaging looks like … slower to build, and harder to sell in a first meeting. It is also the version that survives a complaint.

Ask about money in the same breath, because the cost drivers behind a med spa SEO retainer are the same drivers here. Cost depends on your site, your competition, and your goals. We will be clear about pricing before any work begins. Weighing this against ads? We wrote how a med spa should split budget between paid and organic as its own page.

The work underneath all of it is local SEO done properly. The general mechanism sits on the page about how engines choose a local business. Delivery is described on our AI search page. The page-level method sits in the med spa SEO guide, and the figures sit on the med spa AI cost page.

None of this is a product you switch on. It is steady work across local SEO, your website, your Google Business Profile, and your content. Done properly, your clinic reads the same way to search engines, AI tools, and customers. That is a connected visibility system. It helps your business show up in an assembled answer and in Google Maps results at once. One body of work, so the right people find you twice.

I want to be plain about what I can and cannot promise. We do not promise rankings, lead counts, or AI citations. No honest provider controls what ChatGPT, Perplexity, Google AI Overviews, or Gemini decide to show on a given day. Parts of our research and drafting use AI assistance, and a person checks every claim before it publishes.

This is founder-led work out of Mokena, Illinois, serving Will County and the southwest Chicago suburbs, for med spas, law firms, real estate professionals, home service contractors, and other professional services. You work directly with Jeremy Bengtson. What I can do is show you the source log, show you the run counts, and report in plain English.

If you want your clinic to be easier to find, easier to understand, easier to choose, that starts with the sources an engine actually reads. Tell me what you sell and where you sell it, and I will tell you which source is holding you back … no pressure and no contract. You can work with Jeremy directly, or ask for a free website and SEO review first. The phone number is (217) 579-8791. No contracts. Clients stay because the work earns it.

Frequently asked questions

How do I know which sources an AI answer used about my clinic?

You read the citation list the engine publishes beside its answer. Copy it into a spreadsheet before you close the tab. Perplexity and Google AI Overviews show their sources openly. ChatGPT and Gemini show them when the answer was built with browsing. Log one row per run, with a column for every source named. After enough runs the pattern is obvious.

Do Reddit threads really affect what ChatGPT says about a med spa?

Sometimes yes, and you can verify it yourself in the citation list rather than taking anyone’s word for it. Nobody outside the engine companies can measure how much weight a forum thread carries against a directory page. What you can confirm is whether a thread was cited for a prompt about your treatment and your suburb. Be cautious of any provider who quotes you a percentage here.

How many times do I have to run a prompt before the result means anything?

Ten runs per engine, in a clean session, is the rule we use. That is a house rule and not a published standard, and there is no published standard to point you at. We count a clinic as present when it appears in at least half the runs for that engine. We set that threshold before the test, not after it. One appearance or one absence tells you nothing.

Can my clinic post on Reddit about itself?

Yes, if you post under a real identity, disclose that you work at the practice, and read each community’s rules first. Answer the clinical questions people ask, with no pitch and no link attached. Many communities bar promotion outright and some allow verified professionals to answer. Fake accounts are visible to moderators and readers, and for a medical practice being caught carries real reputational risk.

How do I tell whether a rival clinic is actually beating me in AI answers?

Count how many of your fixed prompts name that clinic, then compare it to how many name yours. Do it per rival and per engine, across the same prompt set, with the same run count. Never blend those counts into one visibility score, because a single number hides which clinic owns which prompts. The gap you can act on only shows up in the per-rival column.

What do I do if ChatGPT says something wrong about my practice?

Correct the source, not the engine. Fix your own pages first, then your Google Business Profile and directory listings, then the review record. Those are the places the answer was built from. Re-run your prompt set a few weeks later to see whether it moved. No provider can force an engine to change what it says.

How often should I re-run the test?

Re-run the full prompt set monthly, on the same prompts and the same run count. Read the trend by quarter rather than by week, because a month-to-month wobble is variance and not progress. Most local SEO shows measurable movement in about three to six months, depending on competition and the condition of your site. Steady traction across all four engines usually runs longer than that.