Two Drug Makers Hold 80% of AI's Injectable Citations. What It Says About Your Med Spa, It Cannot Verify.

Ask about Botox and the assistant answers about the molecule: AbbVie 47%, Galderma 33%

Put a patient's question to ChatGPT, Gemini or Perplexity — how much is Botox, is masseter Botox worth it — and read the answer. It is granular about the drug: the molecule, the units, the manufacturer, the rival product. It is markedly vaguer about the person who would hold the needle.

That asymmetry has been measured, by an interested party. The 5W Med Spa and Aesthetic Medicine AI Visibility Index (June 18 2026) ran 65 patient-intent prompts across five AI engines against 92 candidate brands and reported AbbVie/Allergan at 47% of drug and device AI citation share and Galderma at 33% — 80% between them (5W). 5W is a PR and GEO agency selling AI-visibility work to brands in this exact category, so a finding that two manufacturers own the answer also sells manufacturer retainers. Directional, not peer-reviewed.

The same index put the top 25 brands at 95% of citation share, leaving roughly 11,500 facilities to split the other 5%. What has not been measured matters more: no published study covers how often med spa patients use an assistant to choose a clinic. None. The closest figure is healthcare-wide — rater8's June 2026 survey of about 1,000 US adults put the share of patients using AI to find a new provider at 47%, up from 31% in late 2025 (rater8), from a vendor selling review-collection software to medical practices. Any med-spa-specific percentage quoted at you is that number in disguise, or invented.

Can an assistant check who is legally allowed to inject in your clinic? No — and two state regulators spell out what it is missing

"Safe Botox near me with licensed injectors?" is a verification question, and the assistant answers it with an inference. It cannot open a license register, a medical director agreement or your ownership documents. It has your page text and your listings, and from those it writes a sentence that reads like verification.

The regulators are blunt about the terrain it is guessing across. The Illinois IDFPR/IDPH med spa memo, updated October 30 2025, states that med spa procedures include Botox, dermal fillers and lasers; that physicians may operate, supervise and delegate; that estheticians and cosmetologists are prohibited from techniques intended to affect the living layers of the skin; and that med spas must be physician- or APRN-owned within the structures it lists (Illinois IDFPR/IDPH). The Medical Board of California states that injecting Botox or filler and performing laser treatments is the practice of medicine, that no unlicensed staff may use prescriptive devices or drugs, and that "rent-a-license" medical director arrangements are not legal (Medical Board of California).

Those are two states. Scope of practice, supervision and the corporate practice of medicine vary state by state, and neither document is guidance about your clinic — confirm your structure with a healthcare attorney licensed where you practice, not with an article.

The visibility consequence is safe to state: every fact in those documents — who injects, what license they hold, who supervises, who owns the entity — is what the patient asks the assistant to confirm, and what most med spa sites never put into readable text. The engine fills the gap by guessing, in your name.

The assistant does not return "no data" on your Botox price. It returns a number.

A clinic that retired a laser two years ago, or never published a per-unit price, does not get a blank. Blanks are not what these systems produce. Asked "how much is Botox per unit in [city], and who is reputable?", the model returns a figure. Where that figure comes from, it does not say — and no published study has taken apart how assistants construct med spa prices, so nobody can tell you with authority. It may be inferred from a stale service page, an old directory entry, a review or a regional pricing article; run the prompt against your own clinic and you can at least see what it produced. Asked what you offer, it lists services, sometimes ones you dropped.

Accuracy has been measured. SOCi's 2026 Local Visibility Index, built on roughly 350,000 locations, put business-data accuracy at about 68% for ChatGPT and Perplexity, against 100% for Gemini, which is grounded in Google Maps (SOCi). SOCi sells local-marketing software, so a finding that local business data is broken sells local-marketing software. Discount it: roughly a third of what two of the three major assistants say about a local business is still wrong.

Patients check, but not reliably enough to spare you the error: BrightLocal's 2026 survey of 1,002 US consumers (BrightLocal sells local SEO and reputation tools) found 63% trust AI recommendations, 88% still fact-check and 42% always verify on a review site (BrightLocal). A patient arriving with a per-unit price you have never charged is a patient the assistant priced on your behalf.

What a med spa can put in readable page text so the engine has something to verify

No schema hack outruns this. The engine guesses because the verifiable facts are written nowhere it can read them — not in a booking widget, not baked into an image, not in a PDF price list. Aggarwal et al., "GEO: Generative Engine Optimization" (ACM SIGKDD 2024) benchmarked nine content methods and found adding statistics, quotations and cited sources were the top performers, the best lifting visibility by up to 41% on the Position-Adjusted Word Count metric, while keyword stuffing produced nothing (arXiv). Checkable writing is what the engines reward — and the only register a state medical board is unlikely to object to.

• Who injects, by name, credential and license type — physician, APRN, PA, RN — not the word "expert."
• Who provides medical oversight and what that person does on site. Both regulators above turn on this; how you may describe your own arrangement is a question for counsel.
• The treatments you offer today, plus an explicit line retiring what you no longer offer — the old service page is what the engine is still reading.
• What it costs: per unit, per area, a stated range, or a consult-only policy with a reason. The assistant states a price whether or not you do.
• The device and product names you use — Morpheus8, a named laser platform, a named filler line — not a category noun. Product names are the vocabulary these engines are fluent in.
• What happens if there is a complication: who the patient reaches, how fast, under whose clinical responsibility. Patients ask about safety explicitly; almost no clinic writes it down.

Ranked by what each source lets an engine verify about you — credentials, supervision, and what you actually still offer

No independent study measures which sources AI cites for med spas: the 5W index ranks brands, not sources, and publishes no citation-share table by directory. The one primary measurement is ours. In our July 2026 Dallas run — one market, one day, three engines — the three assistants cited 1,199 sources across med spa answers, and 948 of them (79%) were clinics' own websites; the largest platforms were RealSelf at 62, Yelp at 49 and Instagram at 46, while directories and professional sites contributed 47 between them (the Dallas med spa benchmark). Every raw answer behind those counts is stored and the method is published, so the run is checkable and repeatable — which is more than can be said for any other number in this vertical. Dallas, July 2026: one city, one day, so read it as the shape of the answer rather than a national average. The rest of the ranking rests on general and healthcare-wide citation measurements, so those lines remain untested hypotheses for this vertical.

The order is not by citation volume alone. It is by verifiability: the engine cannot open a license register or a medical director agreement, so it reads text, and treats a fact repeated identically across independent sources as confirmed. That puts your own site first — it is the only place the credential, the oversight arrangement, the price and the live service list can be asserted at all, and it is where 79% of the citations went — and the listings second, because a profile can corroborate a fact or, when it goes stale, contradict it, but it cannot state one you never published.

• 1. Your own website — the only source in this list where a credential, an oversight arrangement, a price or a live service list can be stated in the first place, and the source 79% of our Dallas citations came from (benchmark). The nearest third-party figure points the same way: Gemini takes 52.15% of its citations from brands' own sites, while ChatGPT is nearly the mirror image at 48.73% from third-party listings (Yext, October 2025: 6.8 million citations across 1.6 million AI answers; Yext sells listings management, so a finding that listings decide everything sells listings management). Do this: name the injector, the license type and who provides medical oversight, in text — the facts the regulators above turn on and the engine cannot look up.
• 2. Google Business Profile — the listing an engine most readily treats as current, and therefore the fastest corroboration of what your own pages say. Strongly preferred by Google's own models in BrightLocal's July 2025 study across four engines (BrightLocal; BrightLocal sells local SEO and reputation tools), consistent with Gemini's 100% data accuracy in SOCi's index (SOCi; SOCi sells local-marketing software). Do this: delete every discontinued service from the service list. A retired treatment left standing here is the single easiest hallucination to stop, because you can stop it today — and because your own page has already stopped selling it, the two records stop contradicting each other.
• 3. General healthcare directories (WebMD, Vitals, Healthgrades) — 52.6% of healthcare citations in that same Yext analysis (Yext; listings-management vendor). The largest measured share on this list and, in our own Dallas citations, among the smallest: directories of every kind drew 14 out of 1,199 (benchmark) — one city, one day, so read that as a caution against over-investing here, not as permission to skip it. Third because this is where a credential gets corroborated by a second independent source rather than merely asserted by you. Do this: make the injector's name, license type and treatment list on every profile match your own site word for word. A fact two sources state identically is the closest thing to verification an engine has.
• 4. RealSelf: the largest single platform in our own Dallas citations at 62, ahead of Yelp and Instagram, and one more place the credential can appear in writing. 5W scored it 86 and calls it "the gatekeeper," but publishes no citation-share table to support that (5W; a PR and GEO agency selling into this category) — and 62 citations against 948 to clinics' own sites is a platform, not a gate (benchmark). Do this: complete the profile so the injector credential is stated there too, and log the line as a hypothesis, not a channel.
• 5. Yelp — the largest body of text in which a claim of yours can be contradicted, and 49 citations in our Dallas run (benchmark). Foundation Marketing and AirOps analysed 28 million small-business queries across four engines in Q4 2025: Yelp 512,680 citations against BBB 149,700, Angi 145,600, Thumbtack 56,000, HomeAdvisor 33,600, Nextdoor 10,300 — 72.5% of directory citations in Google AI Mode, 62.1% in Perplexity (Foundation Marketing + AirOps; AirOps sells AI-search software). Do this: read your own reviews for services you no longer offer. That review text is what keeps a retired treatment alive in the answer after you have deleted it everywhere you control.
• 6. Reddit — last, because it is the one source here you can neither write nor correct. ChatGPT cites 15.4 sources per answer and leans on Reddit and Wikipedia, against 3.3 for Gemini (Semrush AI Visibility Index: 126 million prompts, 1,200+ brands, 22 verticals, January–April 2026; Semrush via PPC Land; Semrush sells AI-visibility tracking). Do this: read your city's threads for claims about your clinic's staffing or safety you would have to answer for. Read them; you cannot edit them.

Seven questions a Botox patient types, and the fifteen minutes it takes to run them yourself

Open ChatGPT, Gemini and Perplexity, start a fresh chat for each, paste the questions below with your city in the brackets, and write down every clinic named and in what order. Then the harder half: mark every claim the answers make about your clinic that you cannot substantiate — a service, a price, a credential, an oversight arrangement. Reading public answers touches no chart and no consent form. One pass is still one moment: the engines answer differently across runs, cities and phrasings, and a model update rewrites the picture without notice.

• "Safe Botox near me with licensed injectors?"
• "How much is Botox per unit in [city], and who is reputable?"
• "Best med spa for masseter Botox in [city]?"
• "Can I get filler while breastfeeding?"
• "Morpheus8 vs RF microneedling — which clinic in [city] has good reviews?"
• "What treatments does [your clinic name] offer, and what do they charge?" — this one shows the hallucination directly.
• "Who does the injections at [your clinic name], and is there a medical director?" — this one shows what the engines believe about your credentials.

What med spa owners ask before they change anything

Five objections come up in nearly every conversation with an owner. Each deserves a number or an honest "unmeasured," not a pitch.

• "If two drug companies own the answer, why bother?" The 80% is drug and device citation share — the part of the answer discussing products. Somebody still has to inject the patient, and that sentence is what you compete for. The bleaker figure is the top 25 brands at 95% (5W), from an agency that sells the remedy. Measure your own name before accepting either as fate.
• "Can I just tell the AI we stopped offering that treatment?" No. There is no console and no correction form. Engines rebuild answers from public sources, so the removal has to happen in the sources — service pages, Google Business Profile services, directory listings, old posts — then a lag, then re-measure. A one-click fix for a hallucinated service list does not exist.
• "Is publishing prices actually a good idea?" The commercial question is yours, and genuinely contested. The visibility question has one answer: a clinic that publishes nothing still gets a number, from somewhere else. You are choosing between your figure and a stranger's estimate of it.
• "Can I say 'licensed injectors' and 'medical director' on the site?" State what is true and checkable. What you may claim about supervision, delegation and ownership is governed by your state — Illinois and California, cited above, are strict and not identical. California goes further on the marketing copy itself: Business and Professions Code section 651 requires superiority claims to be substantiated with objective scientific evidence and prohibits false or misleading claims in the advertising of professional services (California Legislative Information). That is California; rules vary by state, and nothing here is advice about your clinic — have counsel read the page before it goes live.
• "Will anyone guarantee ChatGPT names my clinic?" No, and a vendor who does is writing exactly the unsubstantiated claim a medical board would strike from your own website. There is no ad slot inside an AI recommendation, and answers move with phrasing, city and the next model update. You can move the probability, not the certainty.

Run the prompts on your own clinic name before you spend a dollar on any of this

An owner sees every review the clinic receives and none of the answers an assistant gives about it. RecoSignal's free AI Visibility Snapshot runs patient-style questions across ChatGPT, Gemini and Perplexity and reports how often your clinic is named, which clinics are named instead, and which sources those answers lean on. A measurement, repeated — not a ranking, not a citation, not a promise. For the ordered playbook once you know your gaps, see how to improve your AI visibility. Suitability matching — skin type, downtime, budget, "natural, not frozen" — belongs to AI visibility for aesthetic clinics; the confusion between a board-certified surgeon, a cosmetic surgeon and a med spa injector belongs to AI visibility for plastic surgeons.