Ask AI for a cosmetic surgeon in India: it names doctors, and rarely says how to check them

01 / 07Summary

People in Indian cities now ask AI engines who should operate on their nose, their eyelids or their scalp. We asked four engines 60 such questions, about eight cosmetic procedures in six cities, twice each, and read all 480 answers on 5 October 2026.

Asked for a surgeon, the engines usually named one, often four or five. Asked for a surgeon, they rarely told people how to check one. The safety advice was there, but mostly in answers to a different question.

02 / 07Why this matters

Cosmetic surgery is elective and usually paid for privately. In India, anyone can confirm that a doctor is registered by searching the Indian Medical Register that the National Medical Commission publishes, by name or registration number.¹ A patient who asks an AI engine for a surgeon is asking for a shortlist. What the answer says around that shortlist decides whether the patient also learns how to check it.

We wanted to measure that plainly: how often the answers name people, how often they tell the reader what to verify, whether they mention risk and cost, and whose pages they rely on.

The engines knew what to check. They said so when asked about safety, and rarely when asked for a name.

Study 005, Findings 2 and 3

03 / 07Method

We wrote 48 recommendation questions, one for each of eight procedures in each of six cities, in the words people type: “best surgeon for rhinoplasty in Delhi”, “who should I go to for liposuction in Delhi?”.² We added 12 safety questions such as “how do I choose a safe liposuction surgeon in Hyderabad”. Each question went to four engines from India, twice.³ We piloted 12 answers first; the prompts did not change.

Answers

48060 questions × 4 engines × 2 runs; no failures

Citations

5,5973,039 unique URLs on 540 domains

Questions

6048 recommendation, 12 safety

Engines

ChatGPT (search on), Perplexity, Gemini, Google AI Mode

Procedures and cities

8 procedures; Delhi, Mumbai, Bengaluru, Hyderabad, Chennai, Kolkata. Market: India

Fieldwork

5 October 2026

What we counted

A citation is a URL the engine presented as a source. We counted each unique URL once per answer. Links inside ChatGPT’s local business cards, map links and images were not counted.⁴ A doctor is named when a person appears in the answer text with the title “Dr”; we counted distinct people by surname within each answer, so counts are a lower bound.⁵

Each answer was coded against a written rubric: does it name a doctor or a clinic; does it advise checking registration (NMC or a state medical council), a plastic-surgery degree (MCh or DNB), society membership (such as IAAPS or ISAPS), before-and-after photos, reviews, or the hospital set-up; does it mention risks; does it mention cost or give a price. Advice counts only when the answer tells the reader to do it, not when it describes one doctor.

The coding is done by rules written in code, so anyone can re-run it. A second coder re-coded a random 72 answers (15%) without seeing the first codes. Before we revised the rules, agreement was 96% for registration advice and 90% across all measures. We then read a fresh random 40 answers in full: the final rules matched our reading in 39 of 40 for naming a doctor, 38 of 40 for registration advice and 39 of 40 for risks.⁶ Sources were classed by domain into types; every clinic and hospital domain is anonymised in the published data.⁷

For a random sample of 40 doctors named by full name, we searched for a public professional profile (the doctor’s own site, a hospital page or a directory listing) and recorded the qualification it states. We report counts only.⁸

Disclosure: some brands that appear in this study are clients of Citation Research. Their results were produced by the same questions, engines and rules as every other brand’s.

04 / 07Findings

1. Nine in ten answers named a surgeon

Of 384 answers to recommendation questions, 347 (90%) named at least one doctor. An answer that named doctors named four at the median. Across all 480 answers, at least 388 different doctors were named by full name. Only 18 of the 384 answers both named a doctor and told the reader to check registration.

The engines differed. Gemini named a doctor in 95 of 96 recommendation answers and ChatGPT in 94. Perplexity did so in 68 of 96. In 35 of its 96 recommendation answers Perplexity said it could not reliably pick or verify one surgeon; 23 of those still went on to name doctors.

2. Registration was the check least often mentioned

When the question asked for a surgeon, 54% of the 384 answers advised checking for a plastic-surgery degree and 52% advised asking for before-and-after photos. Registration came last: 19 answers, or 5%. Society membership came up in 31 answers (8%).

3. Ask about safety and the advice appears

The 96 answers to safety questions read very differently. Every one gave some verification advice. 55% told the reader to check registration, against 5% of recommendation answers, and 86% mentioned risks or complications, against 24%. Only 24% of safety answers named a doctor.

Risk was the other gap in recommendation answers. ChatGPT (search on) mentioned risks in 38 of 96 and Perplexity in 37, but Gemini did in 7 and Google AI Mode in 12. Among the 370 answers that named a doctor, 94 (25%) mentioned any risk or complication.

Nine in ten answers named a surgeon. One in twenty said how to check one was registered.

Study 005, Findings 1 and 2

4. Directories and Google profiles supplied most sources

The 480 answers cited 5,597 sources. Directories and booking platforms took 34% and Google business profiles 25%: 58% together. Clinic and surgeon sites took 15% and hospital sites 13%. Regulators hardly appeared: 11 answers (2.3% of 480) cited the National Medical Commission, a state medical council or another public body, and the NMC itself was cited in 4.

Each engine leaned on a different kind of page. 73% of Google AI Mode’s 1,885 citations were Google’s own business profiles, the listing cards with ratings and addresses. 76% of Perplexity’s 1,205 were directories. ChatGPT spread its 2,347 more widely, with 24% to hospital sites and 13% to reference sources. Gemini cited little, 160 sources across 120 answers, and 58% of those were clinic and surgeon sites. The most cited named platforms were Justdial (507 citations), Practo (437) and RealSelf (245).

5. Advice was thinnest for hair transplants

Hair transplants drew the most names and the least advice. 98% of the 48 recommendation answers named a doctor, 23% advised checking a qualification and 10% mentioned risks. For rhinoplasty, 2 answers of 48 mentioned risks. Tummy tuck answers carried the most verification advice, at 81%.

6. What the named doctors’ profiles say

We drew 40 doctors at random from those named by full name and looked for a public professional profile. For 24, the profile stated a plastic-surgery qualification. Eight were dermatologists or ophthalmologists, specialists who commonly perform hair transplants and eyelid surgery; five of the eight sampled for hair transplants were dermatologists. Six listed training in other fields: general surgery (2), oral and maxillofacial surgery (2), urology (1) and anaesthesia (1). For two we found no profile.

05 / 07What it means

For patients, for clinics and for the platforms the engines cite, the pattern points the same way.

1. Treat a named doctor as a lead, not a referral.The answers rarely said so. Check the name on the Indian Medical Register and ask the doctor which degree they hold in the specialty.
2. Ask the safety question as well.The same engines gave registration and risk advice in most answers when asked how to choose safely. One extra question changed what people were told.
3. Directory profiles carry weight.Most sources were directory listings and Google business profiles. For clinics and platforms, the qualification and registration details on those listings sit on the pages the engines cite most.
4. Official pages are cited when they answer the question.Surgical societies’ sites were cited in 77 answers; the regulator’s site in 4. Plain, linkable guidance from public bodies has room to be cited more.

06 / 07Limitations

• One day, one market. All 480 answers were collected on 5 October 2026 from India. Engines change without notice, and answers vary between runs.
• Our questions, not real logs. We wrote 60 questions in natural phrasing. People ask in many other ways, in other languages, and follow up.
• Rules can miss. The coding is by rule, checked by a second coder and a hand reading. Advice phrased in unusual ways may be missed, and doctors named without “Dr” are not counted.
• Profiles are not credentials. The qualification check used public profiles, which can be incomplete or out of date. We did not verify degrees or registration with the issuing bodies, and 40 doctors is a small sample.
• Specialty overlaps. Dermatologists, ophthalmologists and ENT surgeons legitimately perform some of these procedures. We report the field a profile states; we make no judgement about any individual.
• We observe, not explain. We see what was cited and said, not why an engine chose it.

07 / 07Data and reproducibility

The question list, the coded answers and the coded citations are published under CC BY 4.0. Doctors, clinics and hospitals are not named in any file: clinic and hospital domains carry anonymised ids, and the qualification sample uses ids D01 to D40. Version 1.0 corresponds to this publication.

• questions.csv · The prompts, procedure, city and type60 rowsCSV
• answers-coded.csv · Every answer’s codes, doctor count and citation count480 rowsCSV
• citations-coded.csv · Every counted citation with its source and type5,597 rowsCSV
• domain-types.csv · The source-type mapping540 rowsCSV
• doctor-qualification-check.csv · Anonymised qualification check40 rowsCSV

For questions about the data, or a reanalysis you would like us to check, write to research@citationresearch.com.

@techreport{citationresearch2026study005,
  author      = {{Citation Research}},
  title       = {Ask AI for a cosmetic surgeon in India: it names doctors, and rarely says how to check them},
  institution = {Citation Research},
  number      = {Study 005},
  year        = {2026},
  url         = {https://citationresearch.com/research/ai-cosmetic-surgeon-recommendations-india}
}