Daybreak - AI gets its health advice from hospitals. Hospitals get it from AI. Neither is right

Episode Date: July 19, 2026

If you search "diabetes blood sugar level" on Google, the AI overview gives you a number. The number is wrong, the source is a hospital website, and the article on that website was written by AI.This is the loop: AI writes flawed medical content for hospital websites, AI search tools treat it as authoritative, and the final result is patients make health decisions based on it. An audit of nearly 500 articles across Apollo, Max, Fortis, Medanta, and Artemis found incorrect drug recommendations, outdated protocols, and Western benchmarks applied to Indian patients.Nobody is legally responsible. Nobody is fixing it.Tune in.Daybreak is produced from the newsroom of The Ken, India’s first subscriber-only business news platform. Subscribe for more exclusive, deeply-reported, and analytical business stories.

On Google, if you search diabetes blood sugar level India, then Google's AI overview gives you three ranges. Normal, pre-diabetic and diabetic. At first glance, the numbers mostly look right. But if you look closer for a moment, then you'll notice something that's a little bit wrong. The AI overview lists the post-meal recommended sugar threshold as under 180 milligrams per decimeter.

The ICMR or the Indian Council of Medical Research on the other hand says that the threshold should be under 200. That is a 20 point difference. Now, most people searching for medical information online won't catch the error. And that is the danger. Maen Zawati, a medicine professor at McGill University, said in an interview with a Canadian broadcaster that the danger is subtle but it's significant. It's not just that the AI can be wrong. It's the fact that it can be wrong in a way that feels right. And what makes it riskier is this,

that the content usually comes from big name hospitals. Here's how. Corporate hospitals fill their websites with medical articles optimized to rank high on Google. And because they carry a trusted hospital's name, AI search tools treat them as credible to base their own output on. The deeper problem is this, that this content isn't just being read by AI. It's increasingly being written by AI as well.

In June, a healthcare content consultant audited nearly 500 articles across the websites of five top hospital chains, Apollo, Max, Fortis, Medanta and Artemis. And the findings which were shared with the Ken were quite concerning. The consultant found that these articles cited wrong wait time. for procedures, recommended discontinued drugs, listed outdated emergency protocols and used western clinical benchmarks on Indian patients. Many of them even had the telltale signs of AI writing. Menstrual discharge, for example, was mistranslated as sewage water. The month of May was changed into the verb can and the mineral iron was confused for the household

appliance. And there were also many AI-coded phrases like, like, let's break this down, that showed up repeatedly. So, basically, this is what was happening. AI writes flawed content, AI search tools read it, and then rank the flawed content quite prominently. This was creating a vicious, dangerous loop. My colleague, the Ken reporter Sideshna Ray, spoke to Sujit Katiar from KGS Consulting,

which advises healthcare firms on regulation and AI compliance. He told her that hospitals are very well. vicariously liable here. But unfortunately, that liability does not extend to legality. And that's just one reason why there is no clear answer to how this loop can be broken. But the fact that it exists at all, though, actually reveals a lot about why no one seems motivated enough to end it. Welcome to Daybreak, a business podcast from the Ken.

I'm your host, Rachel Weigies, and every day of the week, my co-host, Nika, and I will bring you one news story that is worth understanding and worth your time. Today is Monday, the 20th of July. If you scroll through the sources cited for that sugar-level AI overview, you will see that you can't find ICMR or the Research Society for the Study of Diabetes in India. The two bodies that actually set clinical standards here, which both have guidelines that are public and freely available online. Instead, the links that are cited go to hospital websites like,

Metropolis, Artemis, Apollo, a WHO page, an Asian heart journal paper and a block post from Roche's Diabetes Care brand called AccuC. If you scroll further, pass the overview, then the top search results still show blog posts from Max Healthcare, Medanta and smaller hospitals at the very top. Tabres Maner, a Mumbai-based healthcare investor and strategist, told Sudezschean that the entire system is already laid out. Hospitals try to rank their websites on top by having marketing agents generate entire blogs, even though they don't have any domain expertise. A copywriting consultant who works with multiple hospitals and clinics told us that corporate chains routinely outsource medical blogs to content agencies. But the final writer, he said, could be anyone whose research best friend used to be Google and now is Claude or Gemini. Naturally, the outcome of this is a flared of AI written content pushed in the name of the reputable healthcare brands and treated as authoritative by AI search tools even when it is unverified.

A further side effect of this is actually that smaller clinics and hospitals copy this content onto their own sides and end up reinforcing its credibility and extending the loop further. And now this incorrect information is everywhere. In fact, when the auditor shared these findings with the hospitals, most did not even respond. Some, like Fortis, quietly deleted tens of articles. Apollo was the only one that acknowledged the errors and actually removed some of the worst offenders from its website. Many flawed articles are still live, and that's simply because of how the model works. Maner explained to Sudezsna that the hospitals just go ahead and publish those blogs. Typically, there's no check for authenticity or any kind of review that happens.

But the consequences of this go further than just the theoretical. Let's take a look at the diabetes query again. Google's AI overview shows two different HBA1C thresholds. Now, HBA1C is the standard measurement for long-term blood sugar levels, and these thresholds are shown from two different sources. One says that the normal is below 6.5%. Another says 7%. A tech professional who is building AI tools for clinicians

explained why that is a problem. Now, imagine there's a patient whose report shows 6.7% and then they Google it and see the 7% figure. They could end up thinking that they don't have diabetes, delay diagnosis and let their condition get worse. Sujid Katiar from KGAS Consulting, who I mentioned earlier, told Soudeshna that everything is complaint-based. If you find something wrong, you file a complaint and only then can something actually happen. Which means the burden of correction falls on the patient who is being misled.

And that raises an important question. Why are hospitals putting out so much AI slop to begin with? More on this in the next segment. Hi there. This is Vitharri, the producer of Zero Shot, the Ken's podcast that covers everything about artificial intelligence. I'm here to tell you about our latest episode. It's an interesting one that covers something that you may have unread about. We wanted to find out just how Chinese labs have managed to thrive by building models that come close to performing like those from the US.

And when you think about how tightly regulated China's AI industry is, you've got to wonder whether these labs thrive because of that regulation or despite it. We invited Kenra Schaefer, the head of tech policy research, a trivial. China to tell us about that. She covered a lot, like the cost wars that regulators can't stop, how the government is beginning to crack down just as AI is getting really good at finding cybersecurity holes and how it might even go further to restrict Chinese frontier models from going overseas. That and more is in this episode, which is free to listen to until we put it behind a paywall.

The link is in the show notes. Actually, it all comes down to market incentive. Katya told Sadehna that doctors, and by extension hospitals cannot advertise. They can't say things like, I'm treating so-and-so disease, so you should come to me. But what they can do is educate patients,

which is what all these blogs are supposed to do. Except, as Katya said, they are also essentially calling patients in just in a different way. And AI just makes that whole process a lot easier for them. Hospitals use it to write content and also to optimize it for AI search tools so that their pages are kept in front of patients. Once again, let us go back to the diabetes query. Just below the flawed AI overview, search results point users to where to get tested in Javlpour

and name clinics like Metropolis, Dr. Lal Path Labs and Apollo. Now, Google itself shakes off any liability from this by adding a disclaimer in tiny font at the bottom of its AI over. that says this is for informational purposes only. For medical advice or diagnosis, consult a professional. AI responses may include mistakes. Now, even this disclaimer is a sign that the company has learned to cover its basis the hard way. What happened was this? In January, the Guardian reported that Google's AI overview pulled liver function test ranges from Max Healthcare's website and presented them as medical fact.

without any consideration for factors like age, sex or ethnicity. The story sparked backlash and Google pulled the AI overview for that specific query. The Max Healthcare article though stayed up. Now, it's not like people have not tried to fix this. Regulators have tried and failed. In 2023, the National Medical Council introduced rules requiring hospital published medical content to be factual and non-misleading. But doctors and institutions pushed back and the council eventually suspended the rules within 20 days. It went back to its 2002 Code of Conduct, which was written before smartphones, mobile internet or AI existed.

That means today, there is no legal framework that is holding anyone accountable for spreading misleading medical information. So how should consumer-facing health AI be regulated at all? Stay tuned. Ashish Rajura, the founder of Diagnostic AI company Scanbo, told Sudeshna that right now, AI interpretation should be clinician-facing and not consumer-facing. He believes that the tech and the people who use it need to walk before they run, because the stakes are simply too high to be moving faster. He argued that AI tools can misdiagnose symptoms or return false positives and negatives.

Right now, the technology simply is not mature enough to hand out medical advice, and neither is the data that it runs on. Even Manir, the healthcare strategist I mentioned earlier, agreed that a verification layer for AI-generated health content is not even on the horizon yet. He told Sudeshna that it's much too early for any startup or organization to even claim that they can bring any sort of authenticity to this. He also pointed to another deep set problem that Indian healthcare AI is just built on the wrong data. The thing is, Indian and Asian patients have different genomic sequences and thresholds, but most AI models are trained on Western data that don't know any of that. He explained that the health data AI models get trained on only represent 4% of Asia-Pacific populations right now.

That means much of the AI-generated medical advice simply is not built for Indian patients. Now, there are startups that are trying to close this gap, though Manir said that they tend to overestimate what they can deliver, partly because India does not have robust health data for training AI. Even AI Koch, the government's own health database is not clean enough at the moment. He even told us that he has seen many deals not to be able to. fly through after a pilot because they simply cannot take the risk in health care. Plus, investors get cold feet fast because they start questioning feasibility and scale.

Manir argued that that means that regulators need to move with the same caution. The accountability needs to be shared between hospitals, AI developers and regulators. And as long as accountability stays undefined, hospitals, AI companies and search plans, will keep pointing fingers at each other, all while the vicious loop keeps running. Daybreak is produced from the newsroom of the Ken India's first subscriber-focused business news platform. What you're listening to is just a small sample of our subscriber-only offerings. A full subscription offers daily long-form feature stories, newsletters and a whole bunch of premium podcasts. To subscribe, head to the ken.com and click on the red subscribe button on the

top of the Ken website. episode was hosted and produced by my colleague Rachel Vargis and edited by Rajiv Sien.