Where ChatGPT Gets Doctor and Clinic Recommendations: Web, Maps and Reviews

This article is for clinic owners, practice managers and the marketers who support them. A patient asks ChatGPT for a dermatologist, a paediatrician or a dental clinic near them, and a few names come back. You want to know where those names came from and what, if anything, you can change.

It complements our guide to ranking in ChatGPT search, which covers the general approach. Here the focus is narrower: the layers of data that seem to feed a local answer, the evidence for each layer, and a diagnostic checklist for a clinic whose name does not appear. Where something is documented by OpenAI, we link it. Where it is inference from observation, we say so.

Why local answers matter

The BrightLocal Local Consumer Review Survey 2026 reports that 45% of the US consumers it surveyed had used generative AI tools for local business recommendations, up from 6% the previous year. It is a US sample of around a thousand adults, so it does not describe Indian patients. It does indicate that local discovery through AI assistants has moved from curiosity to routine behaviour in at least one large market, and Indian clinics should expect to follow.

The same survey is a reminder that reviews remain central to local decisions. If a clinic's review profile is thin or stale, an AI summary built on it will be thin too.

What OpenAI documents

OpenAI's help article on ChatGPT search describes four things worth knowing. Answers carry links to web pages. The system may rewrite a question into more targeted searches, and sometimes works with partner search providers. Sites that want to appear need to allow OAI-SearchBot and permit traffic from OpenAI's published IP ranges. Device location is optional and off by default; when a user turns it on, it can be used for local recommendations.

OpenAI's page on its crawlers is more specific. OAI-SearchBot is the crawler for search. Sites that opt out of it in robots.txt "will not be shown in ChatGPT search answers, though can still appear as navigational links." GPTBot is separate and relates to model training. ChatGPT-User handles actions that a user initiates, and robots.txt rules may not apply to it. OpenAI also notes it can take about 24 hours for a robots.txt change to take effect.

That is the documented core. It tells you one thing with confidence: a clinic that blocks the search crawler removes itself from the web layer. It does not tell you how local entities are chosen or ranked.

The layers that appear to be involved

Treat the following as a working model for diagnosis, not as a description of OpenAI's system.

Layer 1: web pages retrieved by search. The system issues one or more searches and reads pages. For a local query, this tends to surface clinic websites, directory profiles, and third-party list articles. Standard SEO fundamentals apply: crawlable, indexable, clear titles, answer-first content.

Layer 2: structured local data. Names, addresses, hours, categories and ratings often reach an assistant through map or listing data. Which provider supplies it, and how often, is not something we can verify, so do not assume Google Business Profile is the only input. Check Apple Business Connect, Bing Places and the major Indian directories as well as Google.

Layer 3: review text and ratings. Review platforms contribute star averages, counts and snippets. An assistant that summarises "patients mention short waiting times" is drawing on review text from somewhere.

Layer 4: reputation and editorial mentions. News articles, society pages, hospital pages and list posts that mention the clinic by name.

Layer 5: the model's prior knowledge. For well-known hospitals, the model may recall facts without retrieval. A small clinic should assume it has little prior presence and depends on layers 1 to 4.

Diagnose before you optimise

Run these checks in order. Each is quick, and each removes a common cause.

1. Crawler access

Open your robots.txt and look for rules that block OAI-SearchBot, or a blanket disallow aimed at AI crawlers. Many sites added broad blocks in 2024 and 2025 after reading advice about training data. A block on GPTBot is a training decision. A block on OAI-SearchBot removes you from ChatGPT search results. Decide on each separately, and record who made the decision and why.

Also check your CDN or firewall. Bot-protection rules sometimes block unfamiliar user agents even when robots.txt permits them. Your server logs will show whether OAI-SearchBot requests arrive and what status codes they receive.

2. Indexability and page quality

Confirm that service and doctor pages return 200, are not marked noindex, render their main text in HTML and have a canonical URL that points to themselves. A page that depends on client-side rendering for its main content can be partially read or missed by some crawlers.

3. Entity clarity

Does the website state the clinic's legal name, address, phone number, hours, specialties and the names and registration details of its doctors on pages that are easy to find? Does the name match the Business Profile and directory listings? Mismatches make it harder for any system to treat your listings as one entity.

4. Listing completeness across providers

Audit these at minimum: Google Business Profile, Bing Places, Apple Business Connect and two or three medical directories. Compare name, address, phone, categories and hours. Our local SEO guide for doctors and the post on ranking on Google Maps cover the Google side in detail, so we do not repeat it here.

5. Review profile

Count recent reviews across platforms, not only Google. Check that owners respond, that responses are specific rather than templated, and that no category of complaint is left unanswered. The post on Google reviews for doctors covers acquisition and handling. Do not offer incentives for reviews, and follow NMC advertising rules on how reviews and testimonials appear on your own site.

6. Third-party mentions

Search for the clinic's name and its doctors' names, in quotes. Count independent pages that describe you. A clinic with only its own website and listings provides little corroboration.

An observation protocol

Once the basics are fixed, observe. Choose 20 queries a patient would use in your catchment, such as "paediatrician near [area]" or "dental implants in [city]". Run each in ChatGPT with search enabled, once with location sharing off and once with it on, on two different days. Log:

• which clinics are named, in order
• the links shown and their domains
• whether the answer cites ratings or review counts
• whether your clinic appears

You will quickly see which domains recur. Use them as your priority list for listings and mentions. Do not report changes in visibility from fewer than several repeat runs, because answers differ between sessions.

What a clinic can influence

Rank the levers by how directly you control them.

The last row is the honest one. Nobody outside OpenAI can promise placement.

Two failure modes that look like AI problems

Many "ChatGPT got us wrong" complaints turn out to be data problems upstream.

Stale or conflicting facts. If the website says the clinic closes at 6 pm, the Business Profile says 8 pm and a directory says Sunday closed, an assistant has to pick one or hedge. Patients then arrive at a closed door and the review text says so. Fix the facts at source, and put a single owner and a quarterly check against the hours, phone numbers and doctor list.

Shared addresses and shared names. Multi-doctor clinics, polyclinics inside hospitals and doctors who practise at two sites create ambiguity. A doctor who consults at two hospitals needs a profile page for each location with distinct timings, and the clinic needs one canonical page for each doctor that links to both. Without that, an assistant may attach the wrong location to a name, which sends patients to the wrong building.

In both cases the cure is boring: consistent facts, published in more than one place, owned by a named person. That is also the cure that helps a patient using ordinary search.

Location on, location off

OpenAI documents that device location is optional and off by default. This matters for your test design. With location off, a query such as "paediatrician near me" has less to work with, and the assistant may ask for an area or use a general answer. With it on, the answer can be local. Patients vary in what they enable, so run both modes and note the differences. If your clinic appears only when the query names your locality, your problem is probably area relevance on your pages and listings, not a general lack of visibility.

Compliance points specific to doctors in India

An assistant's answer is not an advertisement by the doctor, but the pages it reads are. Anything you publish to be found must meet NMC rules on advertising: no superlatives, no comparisons, no solicitation, and clinically reviewed statements. Our summary of the 2026 NMC advertising changes is a good starting point. Do not coach patients to prompt an assistant with particular wording, and do not publish pages that exist only to be picked up by AI systems.

Where to start

• Check robots.txt, firewall rules and server logs for OAI-SearchBot this week.
• Build a one-page listing audit across Google, Bing, Apple and two directories, and fix mismatches.
• Count recent reviews on each platform and set a response standard.
• Run the 20-query observation on two days and record the recurring domains.
• Repeat in six weeks and compare against your log.

For tracking citations and publication workflows that support visibility across AI search, AiLLMz is one option. The build work itself sits under our ChatGPT ranking service.