Search for “diabetes blood sugar level India” and Google’s AI overview gives you the normal, pre-diabetic, and diabetic ranges.

At first glance, the numbers seem broadly correct. Linger a moment and you notice the post-meal recommended sugar threshold is mentioned as under 180 mg/dl. The Indian Council of Medical Research, in line with global guidelines, puts the figure at less than 200. A 20-point difference. 

To most people going online in search of medical information, this isn’t “obviously” wrong. And that is precisely the danger. “The danger is subtle, but it’s significant,” Ma’n Zawati, an associate professor of medicine at McGill University, said in an interviewCTV NewsWhy is there so much misleading health information online? with the Canadian broadcaster CTV News. “It’s not just that AI can be wrong, it can be wrong in a way that feels right.”

It only elevates the risk that such content often comes wrapped in a credible package in India: it’s published by big-name hospitals.

Corporate hospitals populate their online platforms with reams of medical content and optimise it for search engines like Google to display most prominently. Bearing the imprimatur of reputable brands, it is deemed credible enough by AI search tools to base their output on.

The problem isn’t just that the information pulled by AI can be wrong, it’s that it is increasingly also written by AI. 

In June, a healthcare content consultant audited nearly 500 articles that five top corporate hospital chains—Apollo, Max, Fortis, Medanta, and Artemis—carried on their websites. The findings, shared with The Ken, were startling. 

The articles mentioned incorrect wait times for procedures, recommended discontinued drugs and outdated emergency protocols, and used Western clinical benchmarks for Indian patients. 

Many of the articles covered in the audit also had “tells” of AI writing. “Menstrual discharge”, for example, was rendered by automated translation as “sewage water”, the month “May” was changed into the verb “can”, and the dietary mineral “iron” was mistaken for the household appliance. Typical AI phrases such as “let’s break this down” were legion.

In effect, AI produced erroneous content, read through it when it was searched for, and gave it prominence in the results.