When a patient asks ChatGPT "Who is the best bariatric surgeon in Tijuana?", the model doesn't flip through a filing cabinet of doctor profiles and pick the most qualified one. When web search is triggered, it appears to pull from recent web results, evaluate available sources, and assemble an answer from the information it finds most structured and credible.
That process takes about two seconds. And it's happening at a scale most clinic owners haven't wrapped their heads around yet.
OpenAI reported in January 2026 that over 40 million people ask ChatGPT healthcare questions every single day. Globally, more than 230 million people submit health and wellness queries to the platform.
Those numbers explain why some of the most experienced surgeons in Tijuana don't show up at all, while clinics with half their credentials appear in every response.
We've been testing this for months. Hundreds of prompts across ChatGPT, Gemini, Claude, and Perplexity. Different specialties, different cities, different phrasing. What we found wasn't random. There are patterns. And once you see them, you can't unsee them.
What actually happens when you ask AI for a doctor
Let me walk you through the mechanics, because most people have the wrong mental model for this.
When ChatGPT Search is active (which it is by default for most users now), the model doesn't just rely on what it learned during training. It can supplement its training data with live web results, and OpenAI's documentation indicates those searches may be powered by Bing. Not Google. That's the first thing most people don't know.
Industry testing has suggested that ChatGPT often draws from pages surfaced in top Bing results, though OpenAI hasn't publicly documented a fixed retrieval range. What practitioners have observed is that it selects content based on its own criteria, not necessarily Bing's ranking order.
And here's something that surprised us: a February 2026 observational study by Sagapixel tracked 54 real users across 177 ChatGPT sessions as they searched for local service providers like dentists, chiropractors, and plastic surgeons. They found that 75% of users still typed keyword-style queries, things like "dentist near Boston" or "best affordable dentist Manhattan." The average session lasted just 2.1 prompts, and 45% of users left after a single query.
In Sagapixel's study, many users behaved more like traditional searchers than conversational researchers: short sessions, keyword-style prompts, and fast decisions. That pattern suggests the window to be the answer is even narrower than you'd think.
In an early benchmark study from late 2024, BrightLocal conducted 800 manual local searches in ChatGPT and recorded every source the model cited.
That breakdown is worth sitting with. More than half of what ChatGPT cites comes directly from business websites. Not from Google reviews. Not from social media. Not from paid ads. From the clinic's own website.
And here's the part that caught my attention: Yelp, Facebook, and Google Maps did not appear as directory sources in BrightLocal's study. At all. In that study, some of the platforms clinics spend the most time on were notably absent from ChatGPT's visible sources.
The signals that seem to matter most
Nobody has published a definitive ranking of what makes ChatGPT recommend one business over another. But research from Onely, BrightLocal, and our own testing points to a fairly consistent set of signals.
Onely analyzed ChatGPT's citation patterns and found that authoritative list mentions (industry rankings, "best of" compilations, expert roundups) accounted for roughly 41% of the influence on commercial recommendations. Awards and accreditations came in at 18%. Online reviews at 16%.
Their research also found something that should concern every clinic that's been focused exclusively on Google: 28% of the pages most frequently cited by ChatGPT have zero organic visibility on Google Search. Different platforms, different winners.
From our own prompt testing across medical tourism queries, here's what we've consistently observed:
- Your website appears to be one of the most important source types. BrightLocal's data backs this up: 58% of sources are business websites. But there's a catch. The website needs to contain structured, extractable information. A page that says "Our team is dedicated to excellence in patient care" gives ChatGPT nothing to work with. A page that says "Dr. Rodriguez is board-certified in bariatric surgery (FACS), completed her fellowship at [hospital], and has performed 7,800+ gastric sleeve procedures since 2012" gives it everything.
- Bing visibility matters more than you think. ChatGPT's search runs through Bing, not Google. Most clinics have never thought about Bing. They don't have a Bing Places profile. Their SEO has been entirely Google-focused. That means they're invisible in the exact index ChatGPT uses to find them.
- Freshness appears to matter. Onely found that 71% of ChatGPT's citations come from content published between 2023 and 2025. If your website hasn't been updated in two years, the model may skip it entirely in favor of a competitor who published last month.
- Third-party mentions amplify everything. When ChatGPT finds your clinic mentioned in an editorial article, a medical tourism guide, or a "best clinics in Tijuana" list from a credible publisher, that mention carries weight. Not because it's a backlink (ChatGPT doesn't care about backlinks the way Google does), but because it represents independent verification that your clinic exists and is worth mentioning.
- Schema markup makes content machine-readable. This is the technical layer most clinics are missing entirely. Physician schema, MedicalProcedure, MedicalOrganization, FAQPage. Without it, your content is just text on a page. With it, every fact is labeled and organized for extraction. We explain this in depth in our guide to GEO.
What we found testing prompts for Tijuana clinics
Theory is useful. But what actually happens when you test this with real queries in the medical tourism space?
Over the past several months, we've run hundreds of prompts across four major AI platforms. Queries like "best bariatric surgeon in Tijuana," "top facelift surgeon Tijuana Mexico," "is it safe to get dental work in Tijuana," and dozens of variations.
Here's what we observed.
ChatGPT and Gemini gave different answers to the same question. This isn't surprising when you understand the mechanics. ChatGPT searches Bing. Gemini searches Google. They're pulling from different indexes. A doctor who ranks well on Google might be invisible to ChatGPT, and vice versa. This is why we test both.
The doctors who appeared consistently had three things in common. They had personal websites with structured content (not just a page on the clinic's site). They had reviews that mentioned them by name and procedure. And they had third-party mentions on editorial sites, medical directories, or published articles.
The doctors who were invisible despite strong credentials were missing the same things. No personal website. No schema markup. Generic reviews that didn't mention their name or specialty. Academic credentials buried in PDF CVs that no AI can read. Published research that wasn't linked to their current practice. We documented one of these cases in detail.
Perplexity was the most aggressive at citing sources. Every answer came with numbered references. If your clinic's website was well-structured, Perplexity would cite it directly. If not, it would cite a directory or editorial that mentioned you secondhand, and you'd lose control of how you were described.
Google AI Overview pulled from Google's existing index. No surprise there. But the AI Overview often featured different clinics than the organic results below it. Being #1 on Google didn't guarantee being in the AI Overview. In our testing, the AI Overview seemed to favor pages with FAQ schema, clear procedural descriptions, and specific numerical data.
The uncomfortable pattern
After running enough prompts, a pattern became hard to ignore.
In practice, AI often surfaces the doctor whose credentials are most legible online, not necessarily the one with the strongest offline reputation.
A surgeon with 20 years of experience, 5,000 procedures, and training under a legendary mentor can be completely invisible to AI if that information isn't structured online. Meanwhile, a younger surgeon with a well-built personal website, consistent directory listings, active review management, and recent blog posts can appear in every response.
And here's what makes this urgent: according to GatherUp, 67% of consumers don't fact-check AI sources before choosing a local business. They read the recommendation, they trust it, and they book. No second opinion. No comparison shopping. For many users, the AI's answer heavily shapes the decision before they ever compare alternatives.
That's not a flaw in the technology. It's a flaw in how most clinics present their doctors. AI can only work with what it can find. If the best information about your doctor lives in a PDF on a hospital intranet or in the memories of grateful patients who never wrote a review, the AI will never know about it.
The gap isn't in medical talent. It's in digital infrastructure. And from what we've seen, it's fixable.
What this means for your clinic
If you're reading this as a clinic owner or a marketing director, here's what I'd take away from all of this.
First, check Bing. Not just Google. Go to bing.com and search your doctor's name, your clinic name, and your top procedures. If you're not showing up well on Bing, you're invisible to ChatGPT regardless of your Google rankings.
Second, look at your website through the lens of extractability. Can AI pull a complete, accurate profile of each doctor from your site? Name, credentials, board certifications, specialties, procedure counts, training, hospital affiliations? Or is your site full of "world-class care" language with no specific facts?
Third, check your review content. Not just the rating. Read through your reviews and ask: does this review mention the doctor by name? The procedure? The outcome? The patient's city of origin? If most of your reviews are "Great experience, 5 stars," they're doing almost nothing for your AI visibility.
Fourth, test the AI yourself. Run 10 prompts across ChatGPT, Gemini, and Perplexity for your specialty and city. Document who shows up, what information is cited, and what sources the AI pulls from. Then compare that to your own digital presence and identify the gaps.
This isn't a one-time fix. AI models update their behavior constantly. The sources they cite shift. New platforms emerge. But the underlying principle holds: the most structured, consistent, and well-documented digital presence tends to win.
We test 20+ prompts per doctor every month across all four platforms. It's the only way to know where you actually stand. And it's the starting point for everything else.