What AI assistants recommend when Indians ask for the best hospital

Data collected on 26 October 2026 by Gaurav Phogat. Method, coding rules and dataset are published below.

When I asked ChatGPT, Gemini and Perplexity for the “best hospital” for six specialties across nine Indian cities, every one of 91 answers gave a shortlist of three to eight hospitals, never a single winner. Private chains took 60% of the hospital mentions I could count, government hospitals 11%. The assistants disagreed on public hospitals: for the same Delhi and Mumbai questions, ChatGPT named one in 75% of answers and Gemini in 33%. And 86% of the sources the assistants linked were hospitals’ own websites.

New to the topic? Start here: AI in healthcare operations sets out the reading order.

Last updated: 26 October 2026.

Patients in India now type “best hospital for knee replacement in Pune” into a chat box as easily as into Google. The obvious question for hospital marketers: when that happens, who gets named, and why? I ran the questions myself, on one day, with one fixed prompt format, and logged every answer. Hospitals are anonymised by type. The aim is to show how these answers are built, not to rank anyone.

Disclosure. I work in hospital digital marketing. Under my standing conflict rule, two hospital groups are excluded from all my studies; their mentions were removed before counting, and the dataset shows how many were removed from each answer.

Download: the dataset (CSV, 97 rows, 23 columns). Free to reuse with credit under CC BY 4.0.

Key findings

1. Shortlists, not verdicts. The 91 answers contained 476 hospital mentions, an average of 5.2 per answer (range 3 to 8). None named a single “best” hospital without alternatives.
2. Private chains dominate. Of 401 counted mentions, national chains took 42% and regional chains 18%. Standalone private hospitals took 26%, government hospitals 11% and private teaching hospitals 2%.
3. Public hospitals depend on the assistant. For the same 12 Delhi and Mumbai prompts, ChatGPT named a government hospital in 9 answers, Perplexity in 5 of 11 and Gemini in 4.
4. Hospital websites are the evidence. 82 of 86 checkable answers showed sources, and 86% of 376 linked sources were hospital-owned websites. For ChatGPT it was 187 of 189.
5. Gemini leans on third parties. 36% of Gemini’s source labels were insurers, crowdfunding platforms, medical-tourism marketplaces or blogs. A health insurer was its most frequent source, appearing in 7 of 26 answers.
6. Read is not recommended. 42 of the 88 hospital-owned pages Perplexity linked belonged to hospitals that did not make its shortlist.
7. Specialty changes the mix. Cancer answers named a government institution 11 times out of 16. Maternity answers named one government hospital in 61 mentions.

How I did this

Prompt. One fixed wording: “best hospital for [specialty] in [city]”. Six specialties: cardiology, cancer treatment, orthopaedics, neurology, kidney disease and maternity. No follow-up questions, no location sharing, no extra context.

Assistants and settings. ChatGPT in a temporary, unpersonalised chat on its default model. Gemini in a temporary chat with the model the picker showed by default. Perplexity on a signed-in free plan in default search mode. All were run through a desktop Chrome browser on 26 October 2026, India time. Nobody logged in for the study; I used the sessions already open in the browser.

Sample. The core is a matched set: five metros (Delhi, Mumbai, Bengaluru, Chennai, Hyderabad) by six specialties, asked of ChatGPT and Gemini, giving 60 answers. ChatGPT was also asked for Kolkata, Pune and Ahmedabad (18 answers) and two Jaipur prompts before it stopped returning answers. Perplexity returned 11 answers (all Delhi and Mumbai prompts except Mumbai maternity) before it stopped responding. That makes 91 first-run answers. I also re-asked six prompts on Gemini to test consistency, for 97 rows in total.

Coding. From each answer I took the hospitals in its main shortlist: the numbered list, table or bolded list, not passing mentions. I classified each one by type using public information: national chain (private group operating in three or more states), regional chain (private group with several hospitals, mostly in one state or region), standalone private (single hospital, including charitable trusts), government (central, state or municipal, including government medical colleges and autonomous institutes) and private teaching (private medical college or university hospital).

Sources. For ChatGPT I counted unique linked domains inside the answer. For Perplexity, unique linked domains on the answer page. For Gemini, the source labels attached to the answer. Each source was coded as hospital-owned or third party (insurer, aggregator, crowdfunding, medical tourism, news, encyclopaedia, government directory).

Limits. One day, one prompt format, and answers change over time. The Perplexity sample is small and covers only two cities. Source counts are not strictly comparable because each interface shows sources differently, and four Gemini answers and one ChatGPT answer were recorded before I captured their sources. Classification is mine. Removing two excluded groups (75 mentions, 16% of the total) means chain shares here are lower than in the raw answers. This is not a quality ranking and says nothing about clinical outcomes.

This connects directly to GA4 setup for hospital websites, What AI assistants say about hospital marketing in India: Q3 2026 and What AI assistants say about your hospital, and how to check.

Every answer is a shortlist, and private chains fill most of it

The first surprise was how alike the answers looked. Each assistant opened with a line about the city being a strong healthcare destination, listed four to seven hospitals with a sentence on each, and closed by telling the reader to choose based on condition, location and doctor. ChatGPT averaged 4.8 names per answer, Gemini and Perplexity 5.7.

National chains took the largest share at 42% of counted mentions. Add regional chains, and private groups account for 60%. Standalone hospitals held their own at 26%, mostly long-established single-site names in Mumbai, Chennai, Pune and Kolkata, and specialty hospitals in orthopaedics, where standalone hospitals took 40% of mentions. Private teaching hospitals barely registered at 2%.

The assistants disagree about public hospitals

Government hospitals made up 11% of mentions overall, and 35 of 91 answers named at least one. But the assistant mattered more than the question. On the matched metro set, ChatGPT named a government hospital in 16 of 30 answers; Gemini in 7 of 30. Gemini also gave national chains a bigger share: 53% of its mentions against 33% for ChatGPT on the same prompts.

ChatGPT often framed the list by budget, with a line such as “government option” or “when treatment affordability matters”, and put a government institution first in 10 of its 50 answers. Gemini more often described its picks as top-tier corporate hospitals, sometimes adding one public institute at the end.

Hospital websites are the evidence AI reads

Almost every answer showed where it came from: 82 of the 86 answers I could check. The four without sources were three ChatGPT maternity answers and one Gemini cardiology answer.

ChatGPT’s links were nearly all official hospital pages, usually one per hospital it named. Perplexity linked more widely, a median of 9 sources per answer, and many were service pages titled “best [specialty] hospital in [city]” published by hospitals themselves. Of the 88 hospital-owned pages it linked, 42 belonged to hospitals that did not make its shortlist. Being read is not the same as being recommended.

Gemini was the outlier. 29 of its 80 source labels were third parties: health insurers, crowdfunding sites, medical-tourism marketplaces and parenting blogs. In a few answers the source label beside a hospital belonged to a different hospital. If you want to understand why official pages now matter so much, I wrote about Google pointing AI answers to official pages and how hospitals get cited by AI search.

Specialty and city change the answer

Cancer was where public institutions showed up most: 11 of 16 cancer answers named at least one government institution, and they took 18% of cancer mentions. Maternity was the opposite. One government hospital appeared in 61 maternity mentions, while national chains, including dedicated mother-and-child brands, took 57%.

City mattered as much as specialty. Delhi answers named a government hospital in 13 of 18 answers. Hyderabad (1 of 12) and Chennai (2 of 12) answers were almost entirely private, with regional chains taking 40% of Hyderabad mentions and standalone hospitals 40% of Chennai mentions. Mumbai split evenly between regional chains, standalone hospitals and national chains at about 30% each.

Ask twice, get a slightly different list

I re-asked six prompts on Gemini later the same day. On average 72% of hospitals overlapped between the two answers (shared names divided by all names across both). Four of six repeats kept the same hospital in first place, and one was identical. So one screenshot of an AI answer tells you little. Visibility here is a probability, and you need repeated checks to measure it.

What this means for hospital marketers

• Treat each specialty and city page as evidence. Assistants quote what your service pages say about services, technology, accreditation and teams. Keep those claims factual and verifiable, show doctors’ registration details, and avoid superlatives and outcome promises that NMC and ASCI guidance on healthcare advertising discourages. The Indian hospital website benchmark covers the basics most sites still miss.
• Make branch pages distinct. Answers named hospitals by locality, such as “Saket” or “Banjara Hills”. A clear page per facility, with address, specialties and timings, makes it easier for an assistant to place you.
• Fix your entity, not just your keywords. Consistent names, addresses and official profiles help assistants match a hospital to its pages. My guide to entity SEO for doctors and hospitals walks through it.
• Check the third-party pages assistants read. Insurer network lists, government directories and medical-tourism listings showed up as sources. Make sure your facts there are current. Do not create fake listings or pay for placements dressed up as reviews.
• Measure with a fixed prompt set, monthly. Pick 20 to 30 specialty and city prompts that matter to you, run each on two or three assistants, repeat each at least twice, and track your share of mentions and sources over time. A single run is noise.
• Specialists have room. Dedicated bone and joint, cancer and mother-and-child hospitals showed up next to large chains. A clearly stated specialty focus seems to travel well into AI answers.

Download the data

The full dataset (CSV) has one row per answer: assistant, settings, city, specialty, exact prompt, run, hospitals named, type counts, whether a government hospital was named or named first, and source counts by type. It contains no hospital or person names. Collection date is in every row. It is free to reuse under CC BY 4.0 with credit to Gaurav Phogat and a link to this page.

How to cite this study

Phogat, G. (2026). What AI assistants recommend when Indians ask for the best hospital: a study of 91 answers. gauravphogat.com. Data collected 26 October 2026. https://gauravphogat.com/ai-hospital-recommendations-study-india/

Questions people ask